#!/usr/bin/env python3 """ Agentic OS — FastAPI Backend Multi-agent orchestration server for opencode, Hermes, Gemini CLI """ import argparse import asyncio import json import os import re import shlex import shutil import signal import subprocess import sys import tarfile import threading import time import uuid from datetime import datetime, timezone from pathlib import Path from typing import Optional from concurrent.futures import ThreadPoolExecutor from fastapi import FastAPI, HTTPException, Query, WebSocket, WebSocketDisconnect from fastapi.middleware.cors import CORSMiddleware from fastapi.responses import HTMLResponse, JSONResponse, Response from fastapi.staticfiles import StaticFiles from pydantic import BaseModel BASE_DIR = Path(__file__).parent.resolve() # Agents supported across chat, routing, health, and kanban dispatch. AGENTS = ["opencode", "hermes", "gemini"] app = FastAPI(title="Agentic OS", version="1.1.0") # Load OpenRouter API key from Hermes .env HERMES_ENV = Path.home() / ".hermes" / ".env" if HERMES_ENV.exists(): for line in HERMES_ENV.read_text().splitlines(): line = line.strip() if line and not line.startswith("#") and "=" in line: k, v = line.split("=", 1) if k == "OPENROUTER_API_KEY": os.environ[k] = v # last value wins (matches shell sourcing) def get_cors_origins() -> list[str]: """Return local dashboard origins allowed to call the API.""" port = 8080 settings_file = BASE_DIR / "data" / "settings.json" if settings_file.exists(): try: settings = json.loads(settings_file.read_text(encoding="utf-8")) port = int(settings.get("dashboard", {}).get("port", port)) except (json.JSONDecodeError, OSError, TypeError, ValueError): port = 8080 origins = { "http://127.0.0.1:8080", "http://localhost:8080", f"http://127.0.0.1:{port}", f"http://localhost:{port}", } extra_origins = os.environ.get("AGENTIC_OS_CORS_ORIGINS", "") origins.update( origin.strip() for origin in extra_origins.split(",") if origin.strip() ) return sorted(origins) # CORS for local dev app.add_middleware( CORSMiddleware, allow_origins=get_cors_origins(), allow_credentials=True, allow_methods=["*"], allow_headers=["*"], ) # ─── Models ─────────────────────────────────────────────────────── class BrainUpdate(BaseModel): content: str class SkillRunRequest(BaseModel): input: Optional[str] = "" agent: Optional[str] = "auto" class OrchestrateRequest(BaseModel): goal: str angles: Optional[dict] = None # {agent: custom_angle_prompt} agents: Optional[list] = None # override default [opencode, gemini, hermes] converge_agent: Optional[str] = "hermes" converge_prompt: Optional[str] = None # custom synthesis instruction class SkillCreate(BaseModel): name: str skill_md: str = "" class SkillUpdate(BaseModel): skill_md: str class SkillContextFileWrite(BaseModel): content: str = "" class ScheduleJobRequest(BaseModel): name: str skill: str cron: str enabled: bool = True class SettingsUpdate(BaseModel): settings: dict class BackupRestoreRequest(BaseModel): file: str class ChatRequest(BaseModel): agent: str message: str # ─── Helper Functions ───────────────────────────────────────────── def read_file(path: Path): if not path.exists(): return "" return path.read_text(encoding="utf-8") def write_file(path: Path, content: str): path.write_text(content, encoding="utf-8") return True def list_dir(path: Path): if not path.exists(): return [] return sorted([p.name for p in path.iterdir() if not p.name.startswith(".")]) def read_json(path: Path, default=None, best_effort=False): """Load JSON from path, returning ``default`` when the file is missing. On corrupt or unreadable content a descriptive ``HTTPException(500)`` is raised so the error is propagated to the client instead of surfacing as an opaque 500. Aggregate/listing callers can set ``best_effort=True`` to tolerate one bad file: the corruption is logged and ``default`` is returned instead of aborting the whole view. """ if not path.exists(): return default try: return json.loads(path.read_text(encoding="utf-8")) except (json.JSONDecodeError, OSError, UnicodeDecodeError) as e: if best_effort: print(f"[load] skipping corrupt {path.name}: {e}") return default raise HTTPException(500, f"Failed to read {path.name}: {e}") def write_json(path: Path, data, indent: int = 2): """Serialize data as pretty JSON to path.""" path.write_text(json.dumps(data, indent=indent), encoding="utf-8") def iter_skill_dirs(): """Yield skill directories, skipping hidden and underscore-prefixed ones.""" skills_dir = BASE_DIR / "skills" if not skills_dir.exists(): return for d in sorted(skills_dir.iterdir()): if d.is_dir() and not d.name.startswith("_"): yield d def new_id(): return str(uuid.uuid4())[:8] def get_timestamp(): return datetime.now(timezone.utc).isoformat() def append_audit(entry: dict): audit_dir = BASE_DIR / "audit" audit_dir.mkdir(parents=True, exist_ok=True) audit_file = audit_dir / "audit.log" entry["timestamp"] = get_timestamp() entry["id"] = new_id() try: audit_file.parent.mkdir(parents=True, exist_ok=True) with open(audit_file, "a", encoding="utf-8") as f: f.write(json.dumps(entry) + "\n") except OSError as e: # Auditing is best-effort: never let a logging failure abort the # underlying operation, but surface it on the server console. print(f"[audit] failed to write entry {entry.get('action')!r}: {e}") def record_brain_learning(source: str, source_path: str, title: str, content: str, agent: str | None = None, updated_at: str | None = None) -> bool: """Push a single document into the centralized brain index. Best-effort: a brain-index failure must never break the calling operation (e.g. a skill run). Calls the same engine the CLI/HTTP endpoints use, synchronously — no HTTP round-trip. """ try: mod = _brain_index_module() conn = mod.get_conn() try: mod.upsert_doc( conn, source=source, source_path=source_path, title=title, content=content, agent=agent, updated_at=updated_at, ) conn.commit() return True finally: conn.close() except Exception as e: print(f"[brain-index] learning upsert failed ({source}/{source_path}): {e}") return False # ─── Agent Stats (real, persisted) ────────────────────────────── AGENT_STATS_FILE = BASE_DIR / "data" / "agent-stats.json" def load_agent_stats() -> dict: if AGENT_STATS_FILE.exists(): try: return json.loads(AGENT_STATS_FILE.read_text()) except (json.JSONDecodeError, KeyError): return {} return {} def save_agent_stats(stats: dict): AGENT_STATS_FILE.parent.mkdir(parents=True, exist_ok=True) AGENT_STATS_FILE.write_text(json.dumps(stats, indent=2)) def record_agent_run(agent: str, success: bool, duration: float): """Persist real invocation stats for an agent (consumed by /api/agents/health).""" stats = load_agent_stats() now = get_timestamp() s = stats.get(agent, { "total_runs": 0, "successful_runs": 0, "failed_runs": 0, "first_seen": now, "last_seen": now, "total_time": 0.0, }) s["total_runs"] += 1 if success: s["successful_runs"] += 1 else: s["failed_runs"] += 1 s["last_seen"] = now s["total_time"] = s.get("total_time", 0.0) + duration s["avg_response_time"] = round(s["total_time"] / s["total_runs"], 3) stats[agent] = s save_agent_stats(stats) def is_error_response(text: str) -> bool: """Detect the error/refusal/busy sentinels this server emits on agent failures. Must match COMPLETE sentinel phrases, never bare glyphs — the warning sign (⚠) and clock (⏱) can appear inside legitimate agent prose, so matching them alone would false-positive real successes. """ t = (text or "").strip() if not t: return True # Full-phrase sentinels only (start-anchored where appropriate). hard = ( "timed out", # '...timed out.' / 'timed out after' "CLI not installed", # '⚠ Agent X CLI not installed' "Error communicating", # '⚠ Error communicating with X' "did not return", # 'Gemini CLI did not return a response' "needs setup", # '**Hermes needs setup**' "needs re-auth", # '**Gemini needs re-auth**' "Error executing skill", # '⚠ Error executing skill' "not configured", # hermes setup markers "no api key", "api_key not", "config file not found", "command not found", ) if any(p in t for p in hard): return True # Whole-line / start-of-response sentinels. if t.startswith("Unknown agent"): return True if t.startswith("**Hermes") and "error" in t.lower(): return True if t.startswith("**Gemini needs"): return True return False # ─── Skill Eval Scoring (heuristic self-improvement metric) ─────── def compute_skill_score(name: str, response_text: str) -> int: """Weighted 0-100 quality score from eval.json criteria + response features.""" skill_dir = BASE_DIR / "skills" / name criteria = [] eval_path = skill_dir / "eval.json" if eval_path.exists(): try: criteria = json.loads(eval_path.read_text()).get("criteria", []) except Exception: criteria = [] if not criteria: criteria = [{"name": "completeness", "weight": 0.4}, {"name": "accuracy", "weight": 0.3}, {"name": "clarity", "weight": 0.3}] total_w = sum(c.get("weight", 0) for c in criteria) or 1.0 text = response_text or "" low = text.lower() is_err = is_error_response(text) completeness = 0 if is_err else min(100, int(len(text) / 8)) accuracy = 0 if is_err else 60 if not is_err: if "```" in text: accuracy += 15 if any(m in low for m in ["step", "1.", "first", "example", "because"]): accuracy += 15 accuracy = min(100, accuracy) clarity = 10 if is_err else 30 if not is_err: if "\n#" in text or "##" in text: clarity += 25 if "- " in text or "* " in text: clarity += 20 if "```" in text: clarity += 15 clarity = min(100, clarity) score_map = {"completeness": completeness, "accuracy": accuracy, "clarity": clarity} overall = sum(score_map.get(c["name"], 50) * c.get("weight", 0) for c in criteria) / total_w return max(0, min(100, int(round(overall)))) def record_skill_score(name: str, score: int, agent: str): """Append a run's score to skills//score-history.json.""" skill_dir = BASE_DIR / "skills" / name if not skill_dir.exists(): return hist_path = skill_dir / "score-history.json" hist = json.loads(hist_path.read_text()) if hist_path.exists() else [] hist.append({ "date": get_timestamp()[:10], "timestamp": get_timestamp(), "score": score, "agent": agent, }) hist_path.write_text(json.dumps(hist, indent=2)) # ─── Agent Registry (dynamic, user-extensible) ───────────────────── AGENT_REGISTRY_FILE = BASE_DIR / "data" / "agent-registry.json" # Built-in agents (always present) BUILTIN_AGENTS = { "opencode": { "name": "opencode", "display_name": "OpenCode", "description": "Code generation, DevOps, file operations", "binary": "opencode", "type": "cli", "run_args": ["opencode", "run", "--format", "json", "{message}"], "check_type": "binary", "builtin": True, }, "hermes": { "name": "hermes", "display_name": "Hermes Agent", "description": "Memory, scheduling, multi-agent coordination", "binary": "hermes", "type": "cli", "run_args": ["hermes", "chat", "-q", "{message}"], "check_type": "binary", "builtin": True, }, "gemini": { "name": "gemini", "display_name": "Gemini CLI", "description": "Research, analysis, multi-modal understanding", "binary": "gemini", "type": "cli", "run_args": ["gemini", "-y", "-m", "gemini-2.5-flash", "{message}"], "check_type": "oauth_file", "oauth_file": ".gemini/gemini-credentials.json", "builtin": True, }, } def load_agent_registry() -> dict: """Load agent registry from disk, merging with builtins.""" registry = dict(BUILTIN_AGENTS) if AGENT_REGISTRY_FILE.exists(): try: custom = json.loads(AGENT_REGISTRY_FILE.read_text()) for name, agent in custom.items(): if name not in BUILTIN_AGENTS: agent["builtin"] = False registry[name] = agent except (json.JSONDecodeError, KeyError): pass return registry def save_agent_registry(agents: dict): """Save custom agents to disk (excluding builtins).""" AGENT_REGISTRY_FILE.parent.mkdir(parents=True, exist_ok=True) custom = {k: v for k, v in agents.items() if not v.get("builtin", False)} AGENT_REGISTRY_FILE.write_text(json.dumps(custom, indent=2)) def _cli_has_subcommand(base_args: list, subcommand: str) -> bool: try: r = subprocess.run([*base_args, "--help"], capture_output=True, text=True, timeout=10) return subcommand in ((r.stdout or "") + (r.stderr or "")) except Exception: return False def hermes_cli_args(*args: str) -> list: """Build the command to invoke Hermes, bridging through WSL if the real agent only lives there. The dashboard commonly runs as a native Windows process while Hermes (whose official installer is Bash-only) lives inside WSL - a plain PATH lookup on Windows will never find it there. Windows machines can also have an unrelated tool also named 'hermes' on PATH (e.g. the academic softwarepub/HERMES metadata-publishing project, which coincidentally shares the name), so don't just trust that a native 'hermes' is the right one - confirm it exposes the NousResearch agent's `chat` subcommand before using it, falling back to WSL otherwise. """ if shutil.which("hermes") is not None and _cli_has_subcommand(["hermes"], "chat"): return ["hermes", *args] if shutil.which("wsl") is not None: quoted = " ".join(shlex.quote(a) for a in args) return ["wsl", "-e", "bash", "-lc", f"hermes {quoted}"] return ["hermes", *args] _hermes_available_cache = {"checked_at": 0.0, "result": False} HERMES_AVAILABLE_CACHE_TTL = 60 def hermes_available() -> bool: """Cached: this spawns a subprocess (possibly via WSL), and /api/status is polled every 15s.""" now = time.time() if now - _hermes_available_cache["checked_at"] < HERMES_AVAILABLE_CACHE_TTL: return _hermes_available_cache["result"] try: r = subprocess.run(hermes_cli_args("--version"), capture_output=True, text=True, timeout=10) result = r.returncode == 0 except Exception: result = False _hermes_available_cache["checked_at"] = now _hermes_available_cache["result"] = result return result def check_agent(name: str) -> dict: """Dynamic agent check based on registry configuration.""" registry = load_agent_registry() agent = registry.get(name) if not agent: return {"name": name, "status": "unknown", "display_name": name} try: check_type = agent.get("check_type", "binary") binary = agent.get("binary", name) exists = shutil.which(binary) is not None if check_type == "binary": status = "online" if exists else "offline" elif check_type == "oauth_file": oauth_path = Path.home() / agent.get("oauth_file", "") logged_in = oauth_path.exists() status = "online" if exists and logged_in else "offline" if not exists else "warning" elif check_type == "http": # HTTP health check endpoint import urllib.request try: url = agent.get("health_url", "") req = urllib.request.Request(url, method="GET") with urllib.request.urlopen(req, timeout=5) as resp: status = "online" if resp.status == 200 else "warning" except Exception: status = "offline" if not exists else "warning" elif check_type == "custom": # Run a custom check command. IMPORTANT: never use shell=True — # check_command comes from agent-registry config and a malicious # value ("; rm -rf ~") would otherwise execute arbitrary shell. check_cmd = agent.get("check_command", "") if check_cmd: import shlex try: cmd = shlex.split(check_cmd) result = subprocess.run( cmd, shell=False, capture_output=True, timeout=10 ) status = "online" if result.returncode == 0 else "offline" except (ValueError, OSError): status = "offline" else: status = "online" if exists else "offline" else: status = "online" if exists else "offline" except Exception: status = "offline" return { "name": name, "display_name": agent.get("display_name", name), "status": status, "description": agent.get("description", ""), "type": agent.get("type", "cli"), "builtin": agent.get("builtin", False), } def execute_agent_dynamic(agent_name: str, message: str) -> str: """Execute a message on any registered agent.""" registry = load_agent_registry() agent = registry.get(agent_name) if not agent: return f"Unknown agent: '{agent_name}'" agent_type = agent.get("type", "cli") if agent_type == "cli": return execute_agent_cli(agent, message) elif agent_type == "http": return execute_agent_http(agent, message) elif agent_type == "mcp": return execute_agent_mcp(agent, message) else: return f"Unsupported agent type: '{agent_type}'" def execute_agent_cli(agent: dict, message: str) -> str: """Execute via CLI binary.""" binary = agent.get("binary", agent["name"]) run_args_template = agent.get("run_args", [binary, "{message}"]) # Build command with message substituted cmd = [] for arg in run_args_template: if "{message}" in arg: cmd.append(arg.replace("{message}", message)) else: cmd.append(arg) timeout = agent.get("timeout", 60) try: result = subprocess.run(cmd, capture_output=True, text=True, timeout=timeout) if result.returncode == 0: output = (result.stdout or "").strip() if output: return output return f"**{agent.get('display_name', binary)}**\n\nProcessed your message.\n\n**Message:** {message[:100]}" err = (result.stderr or "").strip() return err or f"{binary} returned exit code {result.returncode}" except subprocess.TimeoutExpired: return f"⏱ Agent '{agent.get('display_name', binary)}' timed out after {timeout}s.\n\n**Message:** {message[:100]}" except FileNotFoundError: return f"⚠ Agent '{agent.get('display_name', binary)}' CLI not installed. Install it and try again." except Exception as e: return f"⚠ Error communicating with {binary}: {str(e)}" def execute_agent_http(agent: dict, message: str) -> str: """Execute via HTTP API endpoint.""" import urllib.request url = agent.get("api_url", "") method = agent.get("api_method", "POST") headers = agent.get("api_headers", {"Content-Type": "application/json"}) timeout = agent.get("timeout", 60) body = json.dumps({"message": message}).encode() try: req = urllib.request.Request(url, data=body, headers=headers, method=method) with urllib.request.urlopen(req, timeout=timeout) as resp: result = json.loads(resp.read()) return result.get("response", result.get("output", str(result))) except Exception as e: return f"⚠ HTTP agent error: {str(e)}" def execute_agent_mcp(agent: dict, message: str) -> str: """Execute via MCP server.""" return f"⚠ MCP agent type not yet implemented for '{agent.get('display_name', agent['name'])}'" # ─── Routes: Status ─────────────────────────────────────────────── @app.get("/api/status") def get_status(): registry = load_agent_registry() agents = [check_agent(name) for name in registry] skills = list_dir(BASE_DIR / "skills") return { "status": "healthy", "agents": agents, "skills_count": len(skills), "uptime": time.time(), } # ─── Routes: Brain ──────────────────────────────────────────────── @app.get("/api/brain") def list_brain(): files = list_dir(BASE_DIR / "brain") brain_data = {} for f in files: path = BASE_DIR / "brain" / f if path.is_file(): brain_data[f] = read_file(path) return brain_data @app.get("/api/brain/{file_name}") def get_brain_file(file_name: str): path = BASE_DIR / "brain" / file_name if not path.exists() or path.is_dir(): raise HTTPException(404, "File not found") return {"name": file_name, "content": read_file(path)} @app.put("/api/brain/{file_name}") def update_brain_file(file_name: str, data: BrainUpdate): path = BASE_DIR / "brain" / file_name write_file(path, data.content) append_audit({"action": "brain_update", "file": file_name}) return {"status": "ok", "file": file_name} # ─── Routes: Centralized Brain Index (unified information pipeline) ─ # Aggregates brain/** notes, skills/*/learnings.md, and chat history into # one SQLite+FTS5 index. Agents can write to it (ingest/upsert); you can # query it (search/stats). def _brain_index_module(): """Lazy import of brain-core so a missing dir never breaks server boot.""" import importlib.util spec = importlib.util.spec_from_file_location( "brain_index", BASE_DIR / "brain-core" / "brain_index.py" ) mod = importlib.util.module_from_spec(spec) spec.loader.exec_module(mod) return mod @app.post("/api/brain-index/ingest") def ingest_brain_index(): mod = _brain_index_module() conn = mod.get_conn() try: summary = mod.ingest_all(conn) conn.commit() return {"status": "ok", "ingested": summary, "stats": mod.stats(conn)} finally: conn.close() @app.post("/api/brain-index/upsert") def upsert_brain_doc(data: dict): """Agents push a single document into the brain. Fields: source, source_path, title, content, agent (optional).""" for field in ("source", "source_path", "title", "content"): if not data.get(field): raise HTTPException(400, f"missing required field: {field}") mod = _brain_index_module() conn = mod.get_conn() try: doc_id = mod.upsert_doc( conn, source=data["source"], source_path=data["source_path"], title=data["title"], content=data["content"], agent=data.get("agent"), updated_at=data.get("updated_at"), ) conn.commit() return {"status": "ok", "doc_id": doc_id} finally: conn.close() @app.get("/api/brain-index/search") def search_brain_index(q: str = "", limit: int = 20, source: str = None): if not q or not q.strip(): raise HTTPException(400, "query parameter 'q' is required") mod = _brain_index_module() conn = mod.get_conn() try: hits = mod.search(conn, q, limit=limit, source=source) return {"query": q, "count": len(hits), "results": hits} finally: conn.close() @app.get("/api/brain-index/stats") def stats_brain_index(): mod = _brain_index_module() conn = mod.get_conn() try: return mod.stats(conn) finally: conn.close() # ─── Routes: Skills ─────────────────────────────────────────────── SKILL_NAME_RE = re.compile(r"^[a-zA-Z0-9_-]{1,64}$") SKILL_CONTEXT_FILENAME_RE = re.compile(r"^[a-zA-Z0-9_][a-zA-Z0-9_.-]{0,127}$") def skill_dir_path(name: str) -> Path: if not SKILL_NAME_RE.fullmatch(name or ""): raise HTTPException(400, "Invalid skill name") base = (BASE_DIR / "skills").resolve() candidate = (base / name).resolve() if candidate.parent != base: raise HTTPException(400, "Invalid skill name") return candidate def resolve_skill_dir(name: str) -> Path: """Return the directory of an existing skill by matching ``name`` against the actual directory entries. Using the entry from ``iterdir()`` (rather than a path built from ``name``) means traversal input can never escape the skills directory. Raises 404 if no skill matches. """ base = BASE_DIR / "skills" if base.exists(): for entry in base.iterdir(): if entry.is_dir() and entry.name == name: return entry raise HTTPException(404, "Skill not found") def skill_context_file_path(name: str, filename: str) -> Path: if not SKILL_CONTEXT_FILENAME_RE.fullmatch(filename or ""): raise HTTPException(400, "Invalid file name") context_dir = (skill_dir_path(name) / "context").resolve() candidate = (context_dir / filename).resolve() if candidate.parent != context_dir: raise HTTPException(400, "Invalid file name") return candidate @app.get("/api/skills") def list_skills(): skills = [] for d in iter_skill_dirs(): skill_md = read_file(d / "SKILL.md") learnings = read_file(d / "learnings.md") eval_data = read_json(d / "eval.json", {}, best_effort=True) score_history = read_json(d / "score-history.json", [], best_effort=True) skills.append({ "name": d.name, "description": skill_md[:200] if skill_md else "", "has_learnings": bool(learnings), "eval_criteria": eval_data.get("criteria", []), "scores": score_history, }) return skills @app.get("/api/skills/{name}") def get_skill(name: str): path = resolve_skill_dir(name) return { "name": name, "skill": read_file(path / "SKILL.md"), "learnings": read_file(path / "learnings.md"), "eval": read_json(path / "eval.json", default={}), "score_history": read_json(path / "score-history.json", default=[]), "context": [f.name for f in (path / "context").iterdir()] if (path / "context").exists() else [], } @app.post("/api/skills") def create_skill(data: SkillCreate): path = skill_dir_path(data.name) if path.exists(): raise HTTPException(409, "Skill already exists") path.mkdir(parents=True) (path / "context").mkdir() (path / "SKILL.md").write_text(data.skill_md, encoding="utf-8") (path / "learnings.md").write_text("", encoding="utf-8") append_audit({"action": "skill_created", "skill": data.name}) return {"name": data.name} @app.put("/api/skills/{name}") def update_skill(name: str, data: SkillUpdate): path = skill_dir_path(name) if not path.exists(): raise HTTPException(404, "Skill not found") (path / "SKILL.md").write_text(data.skill_md, encoding="utf-8") append_audit({"action": "skill_updated", "skill": name}) return {"status": "ok"} @app.get("/api/skills/{name}/context/{filename}") def get_skill_context_file(name: str, filename: str): path = skill_context_file_path(name, filename) if not path.exists(): raise HTTPException(404, "File not found") return {"filename": filename, "content": read_file(path)} @app.put("/api/skills/{name}/context/{filename}") def put_skill_context_file(name: str, filename: str, data: SkillContextFileWrite): skill_path = skill_dir_path(name) if not skill_path.exists(): raise HTTPException(404, "Skill not found") path = skill_context_file_path(name, filename) path.parent.mkdir(parents=True, exist_ok=True) path.write_text(data.content, encoding="utf-8") append_audit({"action": "skill_context_updated", "skill": name, "file": filename}) return {"status": "ok"} @app.delete("/api/skills/{name}/context/{filename}") def delete_skill_context_file(name: str, filename: str): path = skill_context_file_path(name, filename) if not path.exists(): raise HTTPException(404, "File not found") path.unlink() append_audit({"action": "skill_context_deleted", "skill": name, "file": filename}) return {"status": "deleted"} @app.post("/api/skills/{name}/run") def run_skill(name: str, req: Optional[SkillRunRequest] = None): path = resolve_skill_dir(name) agent_choice = req.agent if req else "auto" skill_input = req.input if req else "" # Read skill files skill_md = read_file(path / "SKILL.md") learnings = read_file(path / "learnings.md") # Determine which agent based on skill type if agent_choice == "auto": devops_keywords = ["devops", "audit", "deploy", "k8s", "gcp", "infra", "terraform"] research_keywords = ["research", "synthesis", "analyze", "search", "compare"] if any(k in name for k in devops_keywords): agent_choice = "opencode" elif any(k in name for k in research_keywords): agent_choice = "gemini" else: # Check SKILL.md for explicit agent assignment for line in skill_md.split('\n'): line = line.strip() if "Primary:" in line: candidate = line.split(":")[-1].strip().lower() if candidate in load_agent_registry(): agent_choice = candidate break if agent_choice == "auto": agent_choice = "opencode" # Build prompt from skill instructions + learnings + user input prompt = f"Execute the '{name}' skill.\n\n" if skill_md: prompt += f"## Skill Instructions\n{skill_md}\n\n" if learnings and learnings.strip(): prompt += f"## Past Learnings\n{learnings}\n\n" if skill_input: prompt += f"## User Input\n{skill_input}" run_id = new_id() # Execute via agent (track real duration for agent-stats) import time as _t _t0 = _t.time() try: response_text = execute_agent(agent_choice, prompt) except Exception as e: # last-resort guard (execute_agent usually returns an error string) response_text = f"⚠ Error executing skill: {str(e)}" _elapsed = _t.time() - _t0 _agent_used = agent_choice or "auto" # Record REAL agent stats (consumed by /api/agents/health) success = not is_error_response(response_text) record_agent_run(_agent_used, success, _elapsed) # Record REAL skill eval score (consumed by /api/analytics/skills) score = compute_skill_score(name, response_text) record_skill_score(name, score, _agent_used) append_audit({ "action": "skill_scored", "skill": name, "agent": agent_choice, "score": score, "duration_s": round(_elapsed, 2), "success": success, }) # Save output to learnings.md timestamp = get_timestamp()[:10] existing = read_file(path / "learnings.md") new_entry = ( f"\n## {timestamp} (Run {run_id})\n" f"- Agent: {agent_choice}\n" f"- Input: {skill_input or '(none)'}\n" f"- Output: {response_text[:500]}\n" ) write_file(path / "learnings.md", existing + new_entry) # Also push this learning into the centralized brain index so it is # immediately searchable alongside brain/ notes and chat history. record_brain_learning( source="skill-learning", source_path=str(path.relative_to(BASE_DIR) / "learnings.md") + f"#{run_id}", title=f"{name} — learning {run_id}", content=f"## {timestamp} (Run {run_id})\n- Agent: {agent_choice}\n- Input: {skill_input or '(none)'}\n- Output: {response_text[:500]}", agent=agent_choice, updated_at=get_timestamp(), ) # Log execution append_audit({ "action": "skill_run", "skill": name, "agent": agent_choice, "run_id": run_id, "output_preview": response_text[:100], }) return { "status": "completed", "run_id": run_id, "skill": name, "agent": agent_choice, "output": response_text, "message": f"Skill '{name}' completed via {agent_choice}", } # ─── Routes: Multi-Agent Orchestration (fan-out + converge) ───────── ORCHESTRATION_DIR = BASE_DIR / "data" / "orchestration-runs" def _run_angle(agent: str, angle_prompt: str): """Run a single agent on its angle. Returns (agent, output, error).""" try: out = execute_agent(agent, angle_prompt) return agent, out, None except Exception as e: return agent, f"⚠ Error from {agent}: {e}", str(e) @app.post("/api/orchestrate") def orchestrate_multi_agent(req: OrchestrateRequest): """Spin up multiple agents on the same project goal from different angles, then converge their outputs into a single synthesis. Default angles (one per built-in agent): - opencode -> implementation plan / code for the goal - gemini -> research, analysis, risk/feasibility - hermes -> coordination, memory, integration plan A convergence step feeds all angle outputs to `converge_agent` (default hermes) to produce the unified result. """ if not req.goal or not req.goal.strip(): raise HTTPException(422, "goal is required") agents = req.agents or ["opencode", "gemini", "hermes"] # Validate agents exist in the registry registry = load_agent_registry() unknown = [a for a in agents if a not in registry] if unknown: raise HTTPException(400, f"unknown agent(s): {unknown}") # Default angle prompts per role default_angles = { "opencode": ( f"PROJECT GOAL: {req.goal}\n\n" "You are the IMPLEMENTATION agent. Produce a concrete implementation " "approach: architecture, files to create/modify, key functions, and a " "step-by-step build plan. Be specific and actionable. Do not do the " "other agents' jobs — focus on engineering." ), "gemini": ( f"PROJECT GOAL: {req.goal}\n\n" "You are the RESEARCH / ANALYSIS agent. Investigate the problem space: " "relevant approaches, trade-offs, risks, feasibility, and any external " "facts or references. Do not write implementation code — focus on " "analysis and evidence." ), "hermes": ( f"PROJECT GOAL: {req.goal}\n\n" "You are the COORDINATION / MEMORY agent. Define how the work should be " "sequenced, what shared state/memory is needed, how the implementation " "and research results integrate, and any scheduling/coordination plan. " "Focus on orchestration, not on coding or research yourself." ), } angle_prompts = {} for a in agents: if req.angles and a in req.angles and req.angles[a]: angle_prompts[a] = req.angles[a] else: angle_prompts[a] = default_angles.get(a, f"PROJECT GOAL: {req.goal}\n\nAddress this goal from your perspective as {a}.") # Fan-out: run all angles in parallel outputs = {} with ThreadPoolExecutor(max_workers=len(agents)) as ex: futures = {ex.submit(_run_angle, a, angle_prompts[a]): a for a in agents} for fut in futures: a = futures[fut] _ag, out, _err = fut.result() outputs[a] = out # Converge converge = req.converge_agent or "hermes" if converge not in registry: converge = agents[0] angle_block = "\n\n".join( f"### {a.upper()} ANGLE OUTPUT:\n{outputs[a]}" for a in agents ) synthesis_prompt = req.converge_prompt or ( f"PROJECT GOAL: {req.goal}\n\n" "You are the CONVERGENCE agent. Below are the outputs of multiple agents " "who each tackled the same project goal from a different angle. Synthesize " "them into ONE coherent plan/result:\n" "- Resolve contradictions between angles.\n" "- Produce a unified, prioritized action plan (or final answer).\n" "- Call out open risks and the single next step.\n\n" f"{angle_block}" ) synthesis, conv_err = _run_angle(converge, synthesis_prompt) # Persist the run run_id = uuid.uuid4().hex[:12] timestamp = get_timestamp() run_record = { "run_id": run_id, "timestamp": timestamp, "goal": req.goal, "agents": agents, "angles": angle_prompts, "outputs": outputs, "converge_agent": converge, "synthesis": synthesis, "converge_error": conv_err, } ORCHESTRATION_DIR.mkdir(parents=True, exist_ok=True) run_file = ORCHESTRATION_DIR / f"{run_id}.json" write_json(run_file, run_record) # Make the run searchable in the brain try: bi = _brain_index_module() conn = bi.get_conn() bi.upsert_doc( conn, source="agent-note", source_path=f"data/orchestration-runs/{run_id}.json", title=f"Multi-agent run: {req.goal[:80]}", content=f"GOAL: {req.goal}\n\nSYNTHESIS:\n{synthesis[:2000]}", agent=f"orchestrate/{converge}", updated_at=timestamp, ) conn.commit(); conn.close() except Exception as e: print(f"[orchestrate] brain upsert failed: {e}") return { "status": "completed", "run_id": run_id, "goal": req.goal, "agents": agents, "outputs": outputs, "converge_agent": converge, "synthesis": synthesis, "converge_error": conv_err, "run_file": str(run_file.relative_to(BASE_DIR)), } @app.get("/api/orchestrate/runs") def list_orchestration_runs(limit: int = 50): ORCHESTRATION_DIR.mkdir(parents=True, exist_ok=True) files = sorted(ORCHESTRATION_DIR.glob("*.json"), reverse=True)[:limit] runs = [] for f in files: try: d = read_json(f, default={}) runs.append({ "run_id": d.get("run_id"), "timestamp": d.get("timestamp"), "goal": d.get("goal"), "agents": d.get("agents"), "converge_agent": d.get("converge_agent"), }) except Exception: continue return {"runs": runs} @app.get("/api/skills/{name}/eval") def get_skill_eval(name: str): path = resolve_skill_dir(name) / "score-history.json" return {"scores": read_json(path, default=[])} # ─── Routes: Scheduler ──────────────────────────────────────────── @app.get("/api/scheduler/jobs") def list_jobs(): jobs_dir = BASE_DIR / "scheduler" / "jobs" jobs = [] for f in sorted(jobs_dir.glob("*.json")): job = read_json(f, default=None, best_effort=True) if job is not None: jobs.append(job) return jobs @app.post("/api/scheduler/jobs") def create_job(job: ScheduleJobRequest): jobs_dir = BASE_DIR / "scheduler" / "jobs" jobs_dir.mkdir(parents=True, exist_ok=True) job_data = { "id": new_id(), "name": job.name, "skill": job.skill, "cron": job.cron, "enabled": job.enabled, "created": get_timestamp(), "last_run": None, "next_run": None, } (jobs_dir / f"{job.name.replace(' ', '_')}.json").write_text( json.dumps(job_data, indent=2) ) append_audit({"action": "job_created", "job": job.name}) return job_data @app.delete("/api/scheduler/jobs/{job_id}") def delete_job(job_id: str): jobs_dir = BASE_DIR / "scheduler" / "jobs" for f in jobs_dir.glob("*.json"): data = read_json(f, default=None, best_effort=True) if data and data.get("id") == job_id: f.unlink() append_audit({"action": "job_deleted", "job_id": job_id}) return {"status": "deleted"} raise HTTPException(404, "Job not found") # ─── Routes: Audit ──────────────────────────────────────────────── @app.get("/api/audit") def get_audit(limit: int = Query(100, le=500)): audit_file = BASE_DIR / "audit" / "audit.log" if not audit_file.exists(): return {"entries": []} lines = audit_file.read_text().strip().split("\n") entries = [] for l in lines: if not l.strip(): continue try: entries.append(json.loads(l)) except (json.JSONDecodeError, ValueError): # Skip malformed lines in the audit log continue return {"entries": entries[-limit:]} # ─── Routes: Cost Analytics ─────────────────────────────────────── @app.get("/api/cost") def get_cost(): cost_file = BASE_DIR / "data" / "cost-history.json" return read_json(cost_file, {"entries": [], "daily_totals": {}, "monthly_projection": 0, "free_tier_alerts": []}) @app.post("/api/cost/record") def record_cost(data: dict): cost_file = BASE_DIR / "data" / "cost-history.json" cost_data = read_json(cost_file, {"entries": [], "daily_totals": {}, "monthly_projection": 0, "free_tier_alerts": []}) cost_data["entries"].append({ "timestamp": get_timestamp(), "agent": data.get("agent", "unknown"), "tokens": data.get("tokens", 0), "cost": data.get("cost", 0.0), "model": data.get("model", "unknown"), }) write_json(cost_file, cost_data) return {"status": "recorded"} # ─── Routes: Registry/Plugins ───────────────────────────────────── @app.get("/api/plugins") def list_plugins(): reg_file = BASE_DIR / "registry" / "plugins.json" return read_json(reg_file, {"plugins": []}) @app.post("/api/plugins/install") def install_plugin(data: dict): name = data.get("name", "").strip() repo_url = data.get("repo_url", "").strip() if not name and not repo_url: raise HTTPException(400, "Plugin name or repo_url required") # Derive name from repo_url if not given if not name and repo_url: name = repo_url.rstrip("/").split("/")[-1].replace(".git", "") reg_file = BASE_DIR / "registry" / "plugins.json" reg = read_json(reg_file, {"plugins": []}) if any(p["name"] == name for p in reg["plugins"]): return {"status": "already_installed", "plugin": name} plugin_dir = BASE_DIR / "skills" / name # If repo_url provided, clone it if repo_url: import subprocess if plugin_dir.exists(): # Pull latest if already cloned subprocess.run(["git", "pull"], cwd=str(plugin_dir), capture_output=True) else: subprocess.run(["git", "clone", "--depth=1", repo_url, str(plugin_dir)], capture_output=True) # If no repo_url, create from template elif not plugin_dir.exists(): import shutil template_dir = BASE_DIR / "skills" / "_template" shutil.copytree(template_dir, plugin_dir) # Detect metadata from SKILL.md frontmatter skill_md = plugin_dir / "SKILL.md" version = "1.0.0" description = "" plugin_type = "skill" if skill_md.exists(): content = skill_md.read_text() import re ver_match = re.search(r'^version:\s*(.+)$', content, re.MULTILINE) desc_match = re.search(r'^description:\s*(.+)$', content, re.MULTILINE) type_match = re.search(r'^type:\s*(.+)$', content, re.MULTILINE) if ver_match: version = ver_match.group(1).strip().strip('"').strip("'") if desc_match: description = desc_match.group(1).strip().strip('"').strip("'") if type_match: plugin_type = type_match.group(1).strip() reg["plugins"].append({ "name": name, "version": version, "description": description, "installed": get_timestamp(), "type": plugin_type, "source": repo_url or "local", }) write_json(reg_file, reg) append_audit({"action": "plugin_installed", "plugin": name}) return {"status": "installed", "plugin": name} @app.delete("/api/plugins/{plugin_name}") def uninstall_plugin(plugin_name: str): reg_file = BASE_DIR / "registry" / "plugins.json" reg = json.loads(reg_file.read_text()) if reg_file.exists() else {"plugins": []} reg["plugins"] = [p for p in reg["plugins"] if p["name"] != plugin_name] reg_file.write_text(json.dumps(reg, indent=2)) append_audit({"action": "plugin_uninstalled", "plugin": plugin_name}) return {"status": "uninstalled", "plugin": plugin_name} # ─── Routes: Connected Apps ──────────────────────────────────────── @app.get("/api/integrations") def list_integrations(): """List connected external apps/integrations.""" int_file = BASE_DIR / "data" / "integrations.json" if not int_file.exists(): return {"integrations": []} return json.loads(int_file.read_text()) @app.post("/api/integrations") def add_integration(data: dict): """Connect an external app via URL/webhook.""" name = data.get("name", "").strip() url = data.get("url", "").strip() int_type = data.get("type", "webhook") if not name or not url: raise HTTPException(400, "name and url required") int_file = BASE_DIR / "data" / "integrations.json" integrations = [] if int_file.exists(): integrations = json.loads(int_file.read_text()).get("integrations", []) if any(i["name"] == name for i in integrations): raise HTTPException(409, f"Integration '{name}' already exists") integrations.append({ "name": name, "url": url, "type": int_type, "added": get_timestamp(), "status": "active", }) int_file.write_text(json.dumps({"integrations": integrations}, indent=2)) append_audit({"action": "integration_added", "name": name, "url": url}) return {"status": "connected", "name": name} # ─── Routes: Backup ─────────────────────────────────────────────── @app.get("/api/backups") def list_backups(): backup_dir = BASE_DIR / "backups" backups = [] for f in sorted(backup_dir.glob("*.tar.gz"), reverse=True): backups.append({ "name": f.name, "size": f.stat().st_size, "created": datetime.fromtimestamp(f.stat().st_mtime).isoformat(), }) return backups @app.post("/api/backup") def create_backup(): backup_dir = BASE_DIR / "backups" backup_dir.mkdir(exist_ok=True) ts = datetime.now().strftime("%Y%m%d_%H%M%S") backup_file = backup_dir / f"agentic-os-{ts}.tar.gz" with tarfile.open(backup_file, "w:gz") as tar: for dir_name in ["brain", "skills", "agents", "registry", "standards", "prompts"]: d = BASE_DIR / dir_name if d.exists(): tar.add(d, arcname=dir_name) append_audit({"action": "backup_created", "file": backup_file.name}) return {"status": "ok", "file": backup_file.name, "size": backup_file.stat().st_size} def _resolve_backup_file(name: str) -> Path: """Resolve a restore request to a real .tar.gz inside backups/, rejecting traversal. The returned path is taken from the directory listing (never built from the raw request value), so a caller can only ever select an existing backup file. """ if not name or name != Path(name).name or not name.endswith(".tar.gz"): raise HTTPException(400, "Invalid backup file name") backup_dir = (BASE_DIR / "backups").resolve() for candidate in backup_dir.glob("*.tar.gz"): if candidate.name == name: return candidate raise HTTPException(404, "Backup file not found") def _safe_extractall(tar: tarfile.TarFile, dest: Path): """Extract a tar archive, refusing members that would escape dest (CVE-2007-4559).""" dest = dest.resolve() for member in tar.getmembers(): target = (dest / member.name).resolve() if target != dest and dest not in target.parents: raise HTTPException(400, f"Unsafe path in archive: {member.name}") if member.issym() or member.islnk(): link_target = (target.parent / member.linkname).resolve() if link_target != dest and dest not in link_target.parents: raise HTTPException(400, f"Unsafe link in archive: {member.name}") tar.extractall(path=dest, filter="data") @app.post("/api/backup/restore") def restore_backup(data: BackupRestoreRequest): backup_file = _resolve_backup_file(data.file) if not backup_file.exists(): raise HTTPException(404, "Backup file not found") with tarfile.open(backup_file, "r:gz") as tar: _safe_extractall(tar, BASE_DIR) append_audit({"action": "backup_restored", "file": data.file}) return {"status": "restored"} # ─── Routes: Prompts ────────────────────────────────────────────── @app.get("/api/prompts") def list_prompts(): prompts_dir = BASE_DIR / "prompts" prompts = {} for f in sorted(prompts_dir.glob("*.md")): prompts[f.stem] = read_file(f) return prompts # ─── Routes: Settings ───────────────────────────────────────────── @app.get("/api/settings") def get_settings(): sf = BASE_DIR / "data" / "settings.json" return read_json(sf, {}) @app.put("/api/settings") def update_settings(data: SettingsUpdate): sf = BASE_DIR / "data" / "settings.json" # Merge with existing existing = read_json(sf, {}) existing.update(data.settings) write_json(sf, existing) append_audit({"action": "settings_updated"}) return {"status": "ok"} # ─── Routes: Standards ──────────────────────────────────────────── @app.get("/api/standards") def list_standards(): std_dir = BASE_DIR / "standards" if not std_dir.exists(): return {"standards": []} standards = [] index_file = std_dir / "index.yml" index_content = read_file(index_file) for f in std_dir.glob("*.md"): standards.append({ "name": f.stem, "content": read_file(f), }) return {"standards": standards, "index": index_content} @app.post("/api/standards/discover") def discover_standards(): # Stub: scans codebase for patterns append_audit({"action": "standards_discovery_run"}) return {"status": "discovery_started", "message": "Scanning codebase for patterns..."} # ─── Routes: Chat ───────────────────────────────────────────────── CHAT_HISTORY_FILE = BASE_DIR / "data" / "chat-history.json" def load_chat_history(): return read_json(CHAT_HISTORY_FILE, {"messages": []}) def save_chat_message(msg: dict): history = load_chat_history() history["messages"].append(msg) if len(history["messages"]) > 200: history["messages"] = history["messages"][-200:] write_json(CHAT_HISTORY_FILE, history) def run_cli(args: list, timeout: int = 30) -> tuple: r = subprocess.run(args, capture_output=True, text=True, timeout=timeout) return r.returncode, r.stdout, r.stderr def clean_hermes_output(raw: str) -> str: """Strip CLI metadata from Hermes output, returning only the AI response.""" if not raw: return "" lines = raw.split('\n') in_box = False content_lines = [] for line in lines: if '╭─' in line: in_box = True continue if '╰─' in line: in_box = False continue if in_box: # Remove ANSI escape codes and leading whitespace cleaned = line.strip() if cleaned: content_lines.append(cleaned) if content_lines: return '\n'.join(content_lines) # Fallback: if no box found, return last non-metadata line non_meta = [l.strip() for l in lines if l.strip() and not l.startswith(('Query:', 'Initializing', '──', 'Resume', 'Session:', 'Duration:', 'Messages:'))] return '\n'.join(non_meta[-5:]) or raw def execute_agent(agent: str, message: str) -> str: import time as _t _t0 = _t.time() _ok = True try: if agent == "opencode": try: code, out, err = run_cli(["opencode", "run", "--format", "json", message], timeout=30) except subprocess.TimeoutExpired: return f"⏱ Agent 'opencode' timed out.\n\nOpenCode's model is taking too long. Try running `opencode run \"{message[:60]}\"` directly in your terminal.\n\n**Message:** {message[:100]}" if code == 0: response_text = "" for line in (out or "").split('\n'): line = line.strip() if not line: continue try: event = json.loads(line) if event.get("type") == "text": text = event.get("part", {}).get("text", "") if text: response_text += text + "\n" except (json.JSONDecodeError, KeyError): continue if response_text: return response_text.strip() return f"**opencode**\n\nProcessed your message.\n\n**Message:** {message[:100]}" err_msg = (err or "").strip() return err_msg or f"opencode returned exit code {code}" elif agent == "hermes": try: code, out, err = run_cli(hermes_cli_args("chat", "-q", message), timeout=180) except subprocess.TimeoutExpired: return f"⏱ Hermes timed out.\n\nThe model took too long to respond. Try a shorter query or check your OpenRouter rate limits.\n\n**Message:** {message[:100]}" if code == 0: cleaned = clean_hermes_output(out or "") if cleaned: return cleaned # Empty response from model - return useful fallback return f"**Hermes**\n\nReceived your message but the model returned an empty response. Try rephrasing your query.\n\n**Message:** {message}" err_msg = (err or "").strip() # Only flag as "needs setup" when the error clearly indicates Hermes # is genuinely unconfigured (missing key/config). Transient failures # (rate limits, timeouts, empty model output, API hiccups) must NOT # be mislabeled as a setup problem. setup_markers = ("not configured", "no api key", "api_key not", "missing config", "config file not found", "command not found") if code != 0 and any(m in err_msg.lower() for m in setup_markers): return f"**Hermes needs setup**\n\nRun `hermes setup` or check your config.\n\n**Details:** {err_msg[:200]}" # Otherwise surface the real error so failures are diagnosable. if err_msg: return f"**Hermes error (exit {code})**\n\n{err_msg[:400]}\n\nIf this looks like a config problem, run `hermes setup`." return f"hermes returned exit code {code}" elif agent == "gemini": for attempt, (args, to) in enumerate([ (["-y", "-m", "gemini-2.5-flash"], 60), (["-y"], 40), ]): try: code, out, err = run_cli(["gemini", *args, message], timeout=to) except subprocess.TimeoutExpired: if attempt == 0: continue return f"⏱ Gemini timed out.\n\nTry running `gemini \"{message[:60]}\"` directly.\n\n**Message:** {message[:100]}" if code == 0: return (out or "").strip() or f"**Gemini CLI**\n\nProcessed your query.\n\n**Message:** {message}" err_msg = (err or "").strip() if attempt == 0 and ("model" in err_msg.lower() or "not found" in err_msg.lower()): continue if "auth" in err_msg.lower() or "login" in err_msg.lower(): return f"**Gemini needs re-auth**\n\nRun `gemini auth login` to re-authenticate.\n\n**Details:** {err_msg[:200]}" return err_msg or f"gemini returned exit code {code}" return "Gemini CLI did not return a response." else: return f"Unknown agent: {agent}" except subprocess.TimeoutExpired: return f"⏱ Agent '{agent}' timed out.\n\nRun `{agent} --help` in your terminal for CLI usage.\n\n**Message:** {message[:100]}" except FileNotFoundError: return f"⚠ Agent '{agent}' CLI not installed. Install it and try again." except Exception as e: return f"⚠ Error communicating with {agent}: {str(e)}" @app.post("/api/chat") def chat(req: ChatRequest): agent = req.agent.lower().strip() registry = load_agent_registry() if agent not in registry: raise HTTPException(400, f"Agent must be one of: {', '.join(registry.keys())}") user_msg = { "id": new_id(), "role": "user", "agent": agent, "content": req.message, "timestamp": get_timestamp(), } save_chat_message(user_msg) response_text = execute_agent(agent, req.message) # Record REAL agent stats (consumed by /api/agents/health) record_agent_run(agent, not is_error_response(response_text), 0.0) agent_msg = { "id": new_id(), "role": "assistant", "agent": agent, "content": response_text, "timestamp": get_timestamp(), } save_chat_message(agent_msg) append_audit({"action": "chat_message", "agent": agent, "msg_preview": req.message[:50]}) return {"status": "ok", "response": agent_msg} @app.get("/api/chat/history") def get_chat_history(): return load_chat_history() # ─── Routes: Terminal (real interactive PTY over WebSocket) ────── class PtySession: """Wraps a real pseudo-terminal shell process, POSIX (stdlib pty) or Windows (pywinpty).""" def __init__(self): self.master_fd = None self.pid = None self.winpty_process = None def start(self, cwd: str): if os.name == "nt": import winpty shell = shutil.which("powershell.exe") or "powershell.exe" self.winpty_process = winpty.PtyProcess.spawn([shell, "-NoLogo"], cwd=cwd) else: import pty pid, fd = pty.fork() if pid == 0: os.chdir(cwd) shell = os.environ.get("SHELL", "/bin/bash") os.execvp(shell, [shell]) else: self.master_fd = fd self.pid = pid def read(self, size: int = 4096): if self.winpty_process is not None: return self.winpty_process.read(size) return os.read(self.master_fd, size) def write(self, data: str): if self.winpty_process is not None: self.winpty_process.write(data) else: os.write(self.master_fd, data.encode()) def resize(self, cols: int, rows: int): if self.winpty_process is not None: self.winpty_process.setwinsize(rows, cols) else: import fcntl import struct import termios fcntl.ioctl(self.master_fd, termios.TIOCSWINSZ, struct.pack("HHHH", rows, cols, 0, 0)) def close(self): if self.winpty_process is not None: try: self.winpty_process.close(force=True) except Exception: pass else: try: os.close(self.master_fd) except OSError: pass if self.pid: # Interactive shells commonly ignore SIGTERM; SIGKILL can't be blocked/ignored. try: os.kill(self.pid, signal.SIGKILL) except OSError: pass try: os.waitpid(self.pid, 0) # reap the killed child so it doesn't stay a zombie except ChildProcessError: pass @app.websocket("/ws/terminal") async def ws_terminal(websocket: WebSocket): # CORSMiddleware does not protect WebSocket handshakes, so this endpoint - which spawns a # full interactive shell - must check the Origin header itself, or any webpage could open # this socket and get command execution on the machine running the dashboard. origin = websocket.headers.get("origin") if origin not in get_cors_origins(): await websocket.close(code=1008) return await websocket.accept() session = PtySession() try: session.start(str(BASE_DIR)) except Exception as e: await websocket.send_json({"type": "output", "data": f"Failed to start terminal: {e}\r\n"}) await websocket.close() return append_audit({"action": "terminal_session_started"}) loop = asyncio.get_event_loop() stop = False def reader(): while not stop: try: data = session.read(4096) except (OSError, EOFError): break if not data: break text = data.decode(errors="replace") if isinstance(data, bytes) else data fut = asyncio.run_coroutine_threadsafe(websocket.send_json({"type": "output", "data": text}), loop) try: fut.result(timeout=5) except Exception: break asyncio.run_coroutine_threadsafe(websocket.close(), loop) threading.Thread(target=reader, daemon=True).start() try: while True: msg = await websocket.receive_json() if msg.get("type") == "input": session.write(msg.get("data", "")) elif msg.get("type") == "resize": session.resize(int(msg.get("cols", 80)), int(msg.get("rows", 24))) except WebSocketDisconnect: pass except Exception: pass finally: stop = True session.close() append_audit({"action": "terminal_session_ended"}) # ═══════════════════════════════════════════════════════════════════ # v0.2.0 — New Feature Endpoints # ═══════════════════════════════════════════════════════════════════ # ─── Models ───────────────────────────────────────────────────── class KanbanTaskCreate(BaseModel): title: str body: str = "" status: str = "triage" priority: str = "medium" assignee: str = "" class KanbanTaskUpdate(BaseModel): title: Optional[str] = None body: Optional[str] = None status: Optional[str] = None priority: Optional[str] = None assignee: Optional[str] = None class KanbanComplete(BaseModel): summary: str = "" class KanbanBlock(BaseModel): reason: str = "" class KanbanCommentCreate(BaseModel): message: str class KanbanLinkCreate(BaseModel): parent_id: str child_id: str class GoalCreate(BaseModel): title: str description: str = "" category: str = "general" target_date: str = "" class GoalUpdate(BaseModel): title: Optional[str] = None description: Optional[str] = None category: Optional[str] = None target_date: Optional[str] = None progress: Optional[int] = None status: Optional[str] = None class JournalSave(BaseModel): content: str class RouterSuggest(BaseModel): task: str class RouterRoute(BaseModel): task: str agent: str # ─── Data Helpers ─────────────────────────────────────────────── KANBAN_DIR = BASE_DIR / "data" / "kanban" GOALS_FILE = BASE_DIR / "data" / "goals.json" JOURNAL_DIR = BASE_DIR / "brain" / "journal" def ensure_dir(d: Path): d.mkdir(parents=True, exist_ok=True) def load_kanban_tasks(): ensure_dir(KANBAN_DIR) tasks = [] for f in sorted(KANBAN_DIR.glob("*.json")): task = read_json(f, default=None, best_effort=True) if task is not None: tasks.append(task) return tasks KANBAN_ID_RE = re.compile(r"^[0-9a-f]{6,16}$") def kanban_task_path(task_id: str) -> Path: """Resolve a task id to its file path, rejecting anything that isn't a plain generated id.""" if not KANBAN_ID_RE.fullmatch(task_id or ""): raise HTTPException(400, "Invalid task id") base = KANBAN_DIR.resolve() candidate = (base / f"{task_id}.json").resolve() if candidate.parent != base: raise HTTPException(400, "Invalid task id") return candidate def save_kanban_task(task: dict): ensure_dir(KANBAN_DIR) kanban_task_path(task["id"]).write_text(json.dumps(task, indent=2)) KANBAN_AGENTS = set(AGENTS) def dispatch_kanban_task(task_id: str): """Move a task to in_progress and hand it to its assignee agent in the background.""" path = kanban_task_path(task_id) if not path.exists(): return task = json.loads(path.read_text()) if task.get("status") in ("in_progress", "done"): return if task.get("assignee") not in KANBAN_AGENTS: return task["status"] = "in_progress" task["updated"] = get_timestamp() save_kanban_task(task) append_audit({"action": "kanban_task_dispatched", "task_id": task_id, "agent": task["assignee"]}) threading.Thread(target=_run_kanban_agent, args=(task_id,), daemon=True).start() def _run_kanban_agent(task_id: str): # Runs in a daemon thread: any unhandled exception would be lost and leave # the task stuck in "in_progress" forever, so catch failures and surface # them by marking the task blocked with the error. try: path = kanban_task_path(task_id) if not path.exists(): return task = json.loads(path.read_text()) agent = task.get("assignee") prompt = task["title"] if not task.get("body") else f"{task['title']}\n\n{task['body']}" response = execute_agent(agent, prompt) failed = response.startswith(("⏱", "⚠", "Unknown agent")) task = json.loads(path.read_text()) # reload in case it changed while the agent ran task.setdefault("comments", []).append({ "id": new_id(), "message": f"🤖 **{agent}**\n\n{response}", "timestamp": get_timestamp(), }) if failed: task["status"] = "blocked" task["block_reason"] = response[:300] append_audit({"action": "kanban_task_dispatch_failed", "task_id": task_id, "agent": agent}) else: task["status"] = "done" task["summary"] = response[:300] task["completed_at"] = get_timestamp() append_audit({"action": "kanban_task_dispatch_completed", "task_id": task_id, "agent": agent}) task["updated"] = get_timestamp() save_kanban_task(task) except Exception as e: print(f"[kanban] dispatch for task {task_id} crashed: {e}") try: path = kanban_task_path(task_id) task = json.loads(path.read_text()) task["status"] = "blocked" task["block_reason"] = f"Dispatch crashed: {e}"[:300] task["updated"] = get_timestamp() save_kanban_task(task) append_audit({"action": "kanban_task_dispatch_error", "task_id": task_id, "error": str(e)[:200]}) except Exception as inner: print(f"[kanban] could not mark task {task_id} as blocked: {inner}") def load_goals(): return read_json(GOALS_FILE, []) def save_goals(goals: list): write_json(GOALS_FILE, goals) # ─── Routes: Kanban Board (13 endpoints) ──────────────────────── @app.get("/api/kanban/board") def kanban_board(status: Optional[str] = None): try: tasks = load_kanban_tasks() if status: tasks = [t for t in tasks if t.get("status") == status] columns = {"triage": [], "todo": [], "ready": [], "in_progress": [], "blocked": [], "done": []} for t in tasks: s = t.get("status", "triage") if s in columns: columns[s].append(t) return {"columns": columns, "total": len(tasks)} except Exception as e: return {"error": str(e), "columns": {}, "total": 0} @app.get("/api/kanban/tasks/{task_id}") def kanban_get_task(task_id: str): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") return json.loads(path.read_text()) @app.delete("/api/kanban/tasks/{task_id}") def kanban_delete_task(task_id: str): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") path.unlink() append_audit({"action": "kanban_task_deleted", "task_id": task_id}) return {"status": "deleted", "task_id": task_id} @app.post("/api/kanban/tasks") def kanban_create_task(data: KanbanTaskCreate): try: task = { "id": new_id(), "title": data.title, "body": data.body, "status": data.status, "priority": data.priority, "assignee": data.assignee, "comments": [], "links": [], "created": get_timestamp(), "updated": get_timestamp(), } save_kanban_task(task) append_audit({"action": "kanban_task_created", "title": data.title}) if task["assignee"] in KANBAN_AGENTS: dispatch_kanban_task(task["id"]) task = json.loads(kanban_task_path(task["id"]).read_text()) return task except Exception as e: raise HTTPException(500, str(e)) @app.patch("/api/kanban/tasks/{task_id}") def kanban_update_task(task_id: str, data: KanbanTaskUpdate): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) assignee_changed = data.assignee is not None and data.assignee != task.get("assignee") for field in ["title", "body", "status", "priority", "assignee"]: val = getattr(data, field, None) if val is not None: task[field] = val task["updated"] = get_timestamp() save_kanban_task(task) append_audit({"action": "kanban_task_updated", "task_id": task_id}) if assignee_changed and task["assignee"] in KANBAN_AGENTS: dispatch_kanban_task(task_id) task = json.loads(path.read_text()) return task @app.post("/api/kanban/tasks/{task_id}/dispatch") def kanban_dispatch_task(task_id: str): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) if task.get("assignee") not in KANBAN_AGENTS: raise HTTPException(400, "Task must be assigned to opencode, hermes, or gemini to dispatch") dispatch_kanban_task(task_id) return {"status": "dispatched", "task_id": task_id} @app.post("/api/kanban/tasks/{task_id}/complete") def kanban_complete_task(task_id: str, data: KanbanComplete): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) task["status"] = "done" task["summary"] = data.summary task["completed_at"] = get_timestamp() task["updated"] = get_timestamp() save_kanban_task(task) append_audit({"action": "kanban_task_completed", "task_id": task_id}) return task @app.post("/api/kanban/tasks/{task_id}/block") def kanban_block_task(task_id: str, data: KanbanBlock): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) task["status"] = "blocked" task["block_reason"] = data.reason task["updated"] = get_timestamp() save_kanban_task(task) append_audit({"action": "kanban_task_blocked", "task_id": task_id}) return task @app.post("/api/kanban/tasks/{task_id}/unblock") def kanban_unblock_task(task_id: str): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) task["status"] = "ready" task["block_reason"] = "" task["updated"] = get_timestamp() save_kanban_task(task) append_audit({"action": "kanban_task_unblocked", "task_id": task_id}) return task @app.post("/api/kanban/tasks/{task_id}/comments") def kanban_add_comment(task_id: str, data: KanbanCommentCreate): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) comment = { "id": new_id(), "message": data.message, "timestamp": get_timestamp(), } task.setdefault("comments", []).append(comment) task["updated"] = get_timestamp() save_kanban_task(task) return task @app.post("/api/kanban/links") def kanban_add_link(data: KanbanLinkCreate): for tid in [data.parent_id, data.child_id]: path = kanban_task_path(tid) if not path.exists(): raise HTTPException(404, f"Task {tid} not found") t = json.loads(path.read_text()) t.setdefault("links", []) link = {"parent": data.parent_id, "child": data.child_id} if link not in t["links"]: t["links"].append(link) t["updated"] = get_timestamp() save_kanban_task(t) append_audit({"action": "kanban_link_added", "parent": data.parent_id, "child": data.child_id}) return {"status": "linked"} @app.delete("/api/kanban/links") def kanban_remove_link(parent_id: str = Query(...), child_id: str = Query(...)): for tid in [parent_id, child_id]: path = kanban_task_path(tid) if path.exists(): t = json.loads(path.read_text()) t.setdefault("links", []) t["links"] = [l for l in t["links"] if not (l.get("parent") == parent_id and l.get("child") == child_id)] t["updated"] = get_timestamp() save_kanban_task(t) return {"status": "unlinked"} @app.delete("/api/kanban/tasks/{task_id}") def kanban_delete_task(task_id: str): path = KANBAN_DIR / f"{task_id}.json" if not path.exists(): raise HTTPException(404, "Task not found") path.unlink() append_audit({"action": "kanban_task_deleted", "task_id": task_id}) return {"status": "deleted"} @app.post("/api/kanban/dispatch") def kanban_dispatch(): dispatched = [] for task in load_kanban_tasks(): if task.get("status") in ("todo", "ready") and task.get("assignee") in KANBAN_AGENTS: dispatch_kanban_task(task["id"]) dispatched.append(task["id"]) append_audit({"action": "kanban_dispatch_triggered", "task_ids": dispatched}) return {"status": "dispatch_triggered", "dispatched": dispatched, "message": f"Dispatched {len(dispatched)} task(s)"} @app.post("/api/kanban/tasks/{task_id}/specify") def kanban_specify_task(task_id: str): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) if task.get("status") == "triage": task["status"] = "todo" task["updated"] = get_timestamp() save_kanban_task(task) return task @app.post("/api/kanban/tasks/{task_id}/decompose") def kanban_decompose_task(task_id: str): path = kanban_task_path(task_id) if not path.exists(): raise HTTPException(404, "Task not found") task = json.loads(path.read_text()) children = [] for i, subtask in enumerate(task.get("body", "").split("\n")): subtask = subtask.strip().lstrip("-* ") if subtask: child = { "id": new_id(), "title": subtask[:80], "body": subtask, "status": "todo", "priority": task.get("priority", "medium"), "assignee": "", "comments": [], "links": [{"parent": task_id, "child": ""}], "created": get_timestamp(), "updated": get_timestamp(), } child["links"][0]["child"] = child["id"] save_kanban_task(child) children.append(child) return {"parent": task_id, "children": children} # ─── Routes: Goals (4 endpoints) ───────────────────────────────── @app.get("/api/goals") def list_goals(): try: return {"goals": load_goals()} except Exception as e: return {"goals": [], "error": str(e)} @app.post("/api/goals") def create_goal(data: GoalCreate): try: goals = load_goals() goal = { "id": new_id(), "title": data.title, "description": data.description, "category": data.category, "target_date": data.target_date, "status": "active", "progress": 0, "created": get_timestamp(), "updated": get_timestamp(), } goals.append(goal) save_goals(goals) # Auto-sync to brain/active-projects.md active_path = BASE_DIR / "brain" / "active-projects.md" if active_path.exists(): existing = active_path.read_text() existing += f"\n- [{goal['title']}](goal:{goal['id']}) — {goal['description'][:80]}\n" active_path.write_text(existing) append_audit({"action": "goal_created", "title": data.title}) return goal except Exception as e: raise HTTPException(500, str(e)) @app.put("/api/goals/{goal_id}") def update_goal(goal_id: str, data: GoalUpdate): try: goals = load_goals() for g in goals: if g["id"] == goal_id: for field in ["title", "description", "category", "target_date", "progress", "status"]: val = getattr(data, field, None) if val is not None: g[field] = val g["updated"] = get_timestamp() save_goals(goals) return g raise HTTPException(404, "Goal not found") except HTTPException: raise except Exception as e: raise HTTPException(500, str(e)) @app.delete("/api/goals/{goal_id}") def delete_goal(goal_id: str): try: goals = load_goals() goals = [g for g in goals if g["id"] != goal_id] save_goals(goals) append_audit({"action": "goal_deleted", "goal_id": goal_id}) return {"status": "deleted"} except Exception as e: raise HTTPException(500, str(e)) # ─── Routes: Journal (4 endpoints) ─────────────────────────────── @app.get("/api/journal/entries") def list_journal_entries(): try: ensure_dir(JOURNAL_DIR) entries = [] for f in sorted(JOURNAL_DIR.glob("*.md"), reverse=True): entries.append({ "date": f.stem, "preview": f.read_text()[:200], "modified": datetime.fromtimestamp(f.stat().st_mtime).isoformat(), }) return {"entries": entries} except Exception as e: return {"entries": [], "error": str(e)} @app.get("/api/journal/entries/{entry_date}") def get_journal_entry(entry_date: str): try: path = JOURNAL_DIR / f"{entry_date}.md" ensure_dir(JOURNAL_DIR) content = path.read_text() if path.exists() else "" return {"date": entry_date, "content": content} except Exception as e: return {"date": entry_date, "content": "", "error": str(e)} @app.put("/api/journal/entries/{entry_date}") def save_journal_entry(entry_date: str, data: JournalSave): try: ensure_dir(JOURNAL_DIR) path = JOURNAL_DIR / f"{entry_date}.md" path.write_text(data.content) append_audit({"action": "journal_saved", "date": entry_date}) return {"status": "saved", "date": entry_date} except Exception as e: raise HTTPException(500, str(e)) @app.get("/api/journal/search") def search_journal(q: str = Query("")): try: ensure_dir(JOURNAL_DIR) if not q: return {"results": []} results = [] for f in JOURNAL_DIR.glob("*.md"): content = f.read_text() if q.lower() in content.lower(): results.append({"date": f.stem, "preview": content[:200]}) return {"results": results, "query": q} except Exception as e: return {"results": [], "error": str(e)} # ─── Routes: Agent Health (3 endpoints) ────────────────────────── @app.get("/api/agents/health") def get_agent_health(): try: stats = load_agent_stats() agents = [] for name in load_agent_registry(): info = check_agent(name) s = stats.get(name, {}) total = s.get("total_runs", 0) info["total_runs"] = total info["successful_runs"] = s.get("successful_runs", 0) info["failed_runs"] = s.get("failed_runs", 0) info["success_rate"] = round(100.0 * s.get("successful_runs", 0) / total, 1) if total else 0.0 info["avg_response_time"] = s.get("avg_response_time", 0.0) info["uptime"] = 0 # process-level uptime not tracked per-agent info["last_seen"] = s.get("last_seen", "") agents.append(info) return {"agents": agents, "updated": get_timestamp()} except Exception as e: return {"agents": [], "error": str(e), "updated": get_timestamp()} @app.get("/api/agents/{name}/stats") def get_agent_stats(name: str): try: if name not in load_agent_registry(): raise HTTPException(400, "Invalid agent") info = check_agent(name) stats = load_agent_stats().get(name, {}) total = stats.get("total_runs", 0) return { "name": name, "status": info["status"], "total_runs": total, "successful_runs": stats.get("successful_runs", 0), "failed_runs": stats.get("failed_runs", 0), "success_rate": round(100.0 * stats.get("successful_runs", 0) / total, 1) if total else 0.0, "avg_response_time": stats.get("avg_response_time", 0.0), "first_seen": stats.get("first_seen", ""), "last_seen": stats.get("last_seen", ""), } except HTTPException: raise except Exception as e: raise HTTPException(500, str(e)) @app.post("/api/agents/health/refresh") def refresh_agent_health(): try: agents = [] for name in load_agent_registry(): info = check_agent(name) agents.append(info) append_audit({"action": "agent_health_refreshed"}) return {"agents": agents, "updated": get_timestamp()} except Exception as e: return {"agents": [], "error": str(e)} # ─── Routes: Agent Registry Management ─────────────────────────── class AgentRegisterRequest(BaseModel): name: str display_name: str = "" description: str = "" binary: str = "" type: str = "cli" # cli, http, mcp run_args: list = [] check_type: str = "binary" # binary, oauth_file, http, custom oauth_file: str = "" health_url: str = "" api_url: str = "" api_method: str = "POST" check_command: str = "" timeout: int = 60 router_keywords: list = [] @app.get("/api/agents") def list_agents(): """List all registered agents (built-in + custom).""" registry = load_agent_registry() agents = [] for name, agent in registry.items(): info = check_agent(name) agents.append(info) return {"agents": agents} @app.post("/api/agents/register") def register_agent(req: AgentRegisterRequest): """Register a new custom agent.""" name = req.name.lower().strip().replace(" ", "_").replace("-", "_") if not name: raise HTTPException(400, "Agent name required") if name in BUILTIN_AGENTS: raise HTTPException(409, f"'{name}' is a built-in agent") registry = load_agent_registry() if name in registry: raise HTTPException(409, f"Agent '{name}' already exists") if not req.display_name: req.display_name = req.name.replace("_", " ").replace("-", " ").title() if not req.binary: req.binary = req.name if not req.run_args: req.run_args = [req.binary, "{message}"] agent_config = { "name": name, "display_name": req.display_name, "description": req.description, "binary": req.binary, "type": req.type, "run_args": req.run_args, "check_type": req.check_type, "timeout": req.timeout, "builtin": False, } if req.oauth_file: agent_config["oauth_file"] = req.oauth_file if req.health_url: agent_config["health_url"] = req.health_url if req.api_url: agent_config["api_url"] = req.api_url if req.api_method: agent_config["api_method"] = req.api_method if req.check_command: agent_config["check_command"] = req.check_command registry[name] = agent_config save_agent_registry(registry) if req.router_keywords: rkf = BASE_DIR / "data" / "router-keywords.json" kw = json.loads(rkf.read_text()) if rkf.exists() else {} kw[name] = req.router_keywords rkf.write_text(json.dumps(kw, indent=2)) append_audit({"action": "agent_registered", "agent": name}) return {"status": "registered", "agent": check_agent(name)} @app.delete("/api/agents/{agent_name}") def unregister_agent(agent_name: str): """Remove a custom agent (built-ins cannot be removed).""" if agent_name in BUILTIN_AGENTS: raise HTTPException(403, "Cannot remove built-in agents") registry = load_agent_registry() if agent_name not in registry: raise HTTPException(404, "Agent not found") del registry[agent_name] save_agent_registry(registry) rkf = BASE_DIR / "data" / "router-keywords.json" if rkf.exists(): kw = json.loads(rkf.read_text()) kw.pop(agent_name, None) rkf.write_text(json.dumps(kw, indent=2)) append_audit({"action": "agent_unregistered", "agent": agent_name}) return {"status": "unregistered", "agent": agent_name} # ─── Routes: Smart Router (2 endpoints) ───────────────────────── ROUTER_RULES = { "opencode": ["code", "devops", "deploy", "git", "file", "terraform", "docker", "test", "build", "infra", "script"], "hermes": ["memory", "schedule", "channel", "skill", "cron", "reminder", "brain", "plugin", "backup"], "gemini": ["research", "analyze", "search", "compare", "explain", "study", "learn", "document", "report", "review"], } @app.post("/api/router/suggest") def router_suggest(data: RouterSuggest): try: task_lower = data.task.lower() # Include custom agent router keywords rkf = BASE_DIR / "data" / "router-keywords.json" if rkf.exists(): custom_kw = json.loads(rkf.read_text()) for agent, kw in custom_kw.items(): if agent not in ROUTER_RULES: ROUTER_RULES[agent] = kw scores = {} for agent, keywords in ROUTER_RULES.items(): scores[agent] = sum(1 for k in keywords if k in task_lower) best = max(scores, key=scores.get) confidence = "high" if scores[best] >= 2 else "medium" if scores[best] == 1 else "low" return { "suggested_agent": best, "confidence": confidence, "scores": scores, "task": data.task, } except Exception as e: return {"suggested_agent": "opencode", "confidence": "low", "error": str(e)} @app.post("/api/router/route") def router_route(data: RouterRoute): try: agent = data.agent.lower() if agent not in load_agent_registry(): return {"status": "error", "message": f"Invalid agent: {agent}"} append_audit({"action": "task_routed", "agent": agent, "task_preview": data.task[:50]}) return { "status": "routed", "agent": agent, "task": data.task, "message": f"Task routed to {agent}", } except Exception as e: return {"status": "error", "message": str(e)} # ─── Routes: Learning Analytics (2 endpoints) ─────────────────── @app.get("/api/analytics/skills") def get_skill_analytics(): try: analytics = [] for d in iter_skill_dirs(): scores = read_json(d / "score-history.json", []) avg_score = sum(s.get("score", 0) for s in scores) / len(scores) if scores else 0 analytics.append({ "name": d.name, "total_runs": len(scores), "avg_score": round(avg_score, 1), "last_score": scores[-1].get("score", 0) if scores else 0, "trend": "up" if len(scores) >= 2 and scores[-1].get("score", 0) > scores[-2].get("score", 0) else "down" if len(scores) >= 2 else "stable", }) return {"skills": sorted(analytics, key=lambda x: x["total_runs"], reverse=True)} except Exception as e: return {"skills": [], "error": str(e)} @app.get("/api/analytics/trends") def get_trend_analytics(): try: trends = [] for d in iter_skill_dirs(): scores = read_json(d / "score-history.json", []) if scores: trends.append({ "name": d.name, "scores": [s.get("score", 0) for s in scores[-10:]], "labels": [s.get("date", "") for s in scores[-10:]], }) return {"trends": trends} except Exception as e: return {"trends": [], "error": str(e)} # ─── Routes: Session Replay (2 endpoints) ─────────────────────── @app.get("/api/sessions/list") def list_sessions(): try: sessions = [] sessions_dir = Path.home() / ".local" / "share" / "opencode" log_dir = sessions_dir / "log" if log_dir.exists(): for f in sorted(log_dir.glob("*.log"), reverse=True)[:20]: sessions.append({ "id": f.stem, "name": f.stem, "size": f.stat().st_size, "modified": datetime.fromtimestamp(f.stat().st_mtime).isoformat(), "source": "opencode", }) hermes_sessions = Path.home() / ".hermes" / "sessions.json" if hermes_sessions.exists(): sessions.append({ "id": "hermes-sessions", "name": "Hermes Session Archive", "size": hermes_sessions.stat().st_size, "modified": datetime.fromtimestamp(hermes_sessions.stat().st_mtime).isoformat(), "source": "hermes", }) return {"sessions": sessions} except Exception as e: return {"sessions": [], "error": str(e)} @app.get("/api/sessions/{session_id}/replay") def get_session_replay(session_id: str): try: sessions_dir = Path.home() / ".local" / "share" / "opencode" log_file = sessions_dir / "log" / f"{session_id}.log" if log_file.exists(): content = log_file.read_text() lines = content.split("\n") messages = [] for line in lines: if "user:" in line.lower() or "assistant:" in line.lower(): messages.append(line) return { "session_id": session_id, "lines": len(lines), "messages": messages[:100], "content": content[:5000], } return {"session_id": session_id, "messages": [], "content": "Session log not found"} except Exception as e: return {"session_id": session_id, "messages": [], "error": str(e)} # ─── WebSocket Terminal ─────────────────────────────────────────── import asyncio import websockets import pty import os import select import struct import fcntl import signal # NOTE: The interactive terminal is provided by the in-app WebSocket # endpoint `/ws/terminal` (see PtySession above). The previous standalone # WebSocket terminal server on port 8082 was removed to avoid two # redundant terminal implementations. # ─── Routes: Dashboard Static Files ────────────────────────────── dashboard_dir = BASE_DIR / "dashboard" if dashboard_dir.exists(): app.mount("/dashboard", StaticFiles(directory=str(dashboard_dir)), name="dashboard") # graphify knowledge-graph output (built by `graphify . --code-only`). # Served so dashboard pages can link/open the interactive graph. graphify_dir = BASE_DIR / "graphify-out" if graphify_dir.exists(): app.mount("/graphify-out", StaticFiles(directory=str(graphify_dir)), name="graphify-out") @app.get("/", response_class=HTMLResponse) def index(): html_file = BASE_DIR / "dashboard" / "index.html" if html_file.exists(): content = html_file.read_text(encoding="utf-8") content = content.replace('href="styles.css"', 'href="/dashboard/styles.css"') content = content.replace('src="utils.js"', 'src="/dashboard/utils.js"') content = content.replace('src="api.js"', 'src="/dashboard/api.js"') content = content.replace('src="app.js"', 'src="/dashboard/app.js"') content = content.replace('pages/', '/dashboard/pages/') return HTMLResponse(content=content) return HTMLResponse("

Agentic OS

Dashboard not built yet. Run the installer for your platform first (./install.sh on Linux/macOS or .\\install.ps1 on Windows).

") @app.get("/test", response_class=HTMLResponse) def test_page(): html_file = BASE_DIR / "dashboard" / "test.html" if html_file.exists(): return HTMLResponse(html_file.read_text()) return HTMLResponse("

Test page not found

") # ─── Routes: Agent Time Monitor ─────────────────────────────── # Tracks total time AI agents spent on the project. Time is DERIVED from # event logs (audit.log, chat-history.json, cost-history.json) via a # session-gap estimator — see scripts/analyze_agent_time.py. # NOTE: no recompute happens at startup or import; the report is only # regenerated on explicit /api/agent-time/recompute (or if the cache # file is missing). This keeps app boot fast and side-effect free. AGENT_TIME_REPORT = BASE_DIR / "data" / "agent-time.json" AGENT_TIME_ANALYZER = BASE_DIR / "scripts" / "analyze_agent_time.py" def _recompute_agent_time() -> dict: """Run the analyzer to (re)generate data/agent-time.json. Returns report.""" import subprocess try: subprocess.run( [sys.executable, str(AGENT_TIME_ANALYZER)], cwd=str(BASE_DIR), capture_output=True, text=True, timeout=120, ) except Exception: pass # The analyzer always writes the file (even with 0 events), but guard # in case the script is missing. if AGENT_TIME_REPORT.exists(): return json.loads(AGENT_TIME_REPORT.read_text(encoding="utf-8")) return {"project": "agentic-os", "total_seconds": 0, "total_human": "0h 0m 0s", "agents": {}, "event_count": 0, "method": "unavailable"} @app.get("/api/agent-time") def get_agent_time(recompute: bool = False): if recompute or not AGENT_TIME_REPORT.exists(): return _recompute_agent_time() try: return json.loads(AGENT_TIME_REPORT.read_text(encoding="utf-8")) except Exception: return _recompute_agent_time() @app.post("/api/agent-time/recompute") def recompute_agent_time(): report = _recompute_agent_time() return {"ok": True, "total_human": report.get("total_human"), "agents": len(report.get("agents", {})), "event_count": report.get("event_count", 0)} # Skin-friendly view for Rainmeter (WebParser). Returns plain text: # Line 1: human total, e.g. "5h 34m 45s" # Line 2: total_seconds + fixed-order per-agent seconds (0 if absent) # system hermes opencode gemini codex jarvis kilocode test_claude test # This fixed-width layout lets a single RegExp grab every value reliably. _AGENT_ORDER = ["system", "hermes", "opencode", "gemini", "codex", "jarvis", "kilocode", "test_claude", "test"] @app.get("/api/agent-time/skin") def agent_time_skin(): if not AGENT_TIME_REPORT.exists(): report = _recompute_agent_time() else: try: report = json.loads(AGENT_TIME_REPORT.read_text(encoding="utf-8")) except Exception: report = _recompute_agent_time() agents = report.get("agents", {}) total = report.get("total_seconds", 0) per = " ".join(str(agents.get(a, {}).get("total_seconds", 0)) for a in _AGENT_ORDER) body = f"{report.get('total_human', '0h 0m 0s')}\n{total} {per}" return Response(content=body, media_type="text/plain; charset=utf-8") # ─── Routes: Agent Insights (AI usage analytics) ────────────────── # Aggregates real usage data across chat history, cost history, derived # agent-time, the agent registry, and router keywords to answer: # - which agent do I chat with most # - which agent spends the most time completing tasks # - which agent seems best suited for what # - which models I use most with each agent # All source data is read-only; no writes happen on this endpoint. ROUTER_KEYWORDS_FILE = BASE_DIR / "data" / "router-keywords.json" # Human-readable role hints keyed by agent name (falls back to registry # description / router keywords when unknown). AGENT_ROLE_HINTS = { "opencode": "Code generation, file ops, DevOps/infra, git, software engineering", "hermes": "Persistent memory, cron scheduling, messaging channels, skill hub, coordination", "gemini": "Web research, multi-modal/image/PDF analysis, reasoning, data analysis", "jarvis": "Local-first personal AI: deep research, knowledge, memory, general reasoning", "kilocode": "AI coding assistant: implement, refactor, programming tasks", "codex": "OpenAI Codex: coding, debugging, tests, builds", } @app.get("/api/agent-insights") def get_agent_insights(recompute: bool = False): # Pull the four required datasets from disk. chat = load_chat_history().get("messages", []) cost = read_json( BASE_DIR / "data" / "cost-history.json", {"entries": []}, ).get("entries", []) registry = load_agent_registry() router_kw = read_json(ROUTER_KEYWORDS_FILE, {}) if recompute or not AGENT_TIME_REPORT.exists(): time_report = _recompute_agent_time() else: try: time_report = json.loads(AGENT_TIME_REPORT.read_text(encoding="utf-8")) except Exception: time_report = _recompute_agent_time() time_agents = time_report.get("agents", {}) # 1) Chat volume per agent (count user+assistant messages; pair count # = number of user turns, which is the most meaningful "chat" metric). chat_counts = {} user_turns = {} first_chat = {} last_chat = {} for m in chat: a = m.get("agent") or "unknown" chat_counts[a] = chat_counts.get(a, 0) + 1 if m.get("role") == "user": user_turns[a] = user_turns.get(a, 0) + 1 ts = m.get("timestamp") if ts: if a not in first_chat or ts < first_chat[a]: first_chat[a] = ts if a not in last_chat or ts > last_chat[a]: last_chat[a] = ts # 2) Time per agent (from derived agent-time report). time_per_agent = { a: { "total_seconds": d.get("total_seconds", 0), "sessions": d.get("sessions", 0), "touches": d.get("touches", 0), } for a, d in time_agents.items() } # 3) Models used per agent (from cost history). models_per_agent = {} tokens_per_agent = {} cost_per_agent = {} for e in cost: a = e.get("agent") or "unknown" model = e.get("model") or "unknown" models_per_agent.setdefault(a, {}) models_per_agent[a][model] = models_per_agent[a].get(model, 0) + 1 tokens_per_agent[a] = tokens_per_agent.get(a, 0) + (e.get("tokens", 0) or 0) cost_per_agent[a] = cost_per_agent.get(a, 0) + (e.get("cost", 0) or 0) # 4) Best-suited-for: combine registry description, router keywords, # and AGENTS.md role hints into a per-agent "suite" summary. suite = {} for name in set(list(registry.keys()) + list(router_kw.keys()) + list(chat_counts.keys()) + list(time_per_agent.keys())): desc = (registry.get(name, {}) or {}).get("description", "") role = AGENT_ROLE_HINTS.get(name, "") kws = router_kw.get(name, []) suite[name] = { "description": desc or role, "role_hint": role, "keywords": kws, } # Build a per-agent consolidated view. all_agents = set( list(chat_counts.keys()) + list(time_per_agent.keys()) + list(models_per_agent.keys()) + list(suite.keys()) ) per_agent = {} for a in all_agents: per_agent[a] = { "agent": a, "display_name": (registry.get(a, {}) or {}).get("display_name", a), "chat_messages": chat_counts.get(a, 0), "user_turns": user_turns.get(a, 0), "time_seconds": time_per_agent.get(a, {}).get("total_seconds", 0), "sessions": time_per_agent.get(a, {}).get("sessions", 0), "touches": time_per_agent.get(a, {}).get("touches", 0), "tokens": tokens_per_agent.get(a, 0), "cost": round(cost_per_agent.get(a, 0), 6), "top_models": sorted( models_per_agent.get(a, {}).items(), key=lambda kv: kv[1], reverse=True, ), "suite": suite.get(a, {}), "first_seen": first_chat.get(a), "last_seen": last_chat.get(a), } # Rankings by_chats = sorted( per_agent.values(), key=lambda x: x["user_turns"], reverse=True ) by_time = sorted( per_agent.values(), key=lambda x: x["time_seconds"], reverse=True ) # Most-used model overall per agent already in top_models; also a global # model popularity map. global_models = {} for a, models in models_per_agent.items(): for m, c in models.items(): global_models[m] = global_models.get(m, 0) + c return { "generated_at": get_timestamp(), "totals": { "total_chat_messages": sum(chat_counts.values()), "total_user_turns": sum(user_turns.values()), "total_time_seconds": time_report.get("total_seconds", 0), "total_time_human": time_report.get("total_human", "0h 0m 0s"), "total_tokens": sum(tokens_per_agent.values()), "total_cost": round(sum(cost_per_agent.values()), 6), "agents_observed": len(per_agent), }, "most_chatted": [ {"agent": x["agent"], "display_name": x["display_name"], "user_turns": x["user_turns"], "chat_messages": x["chat_messages"]} for x in by_chats if x["user_turns"] > 0 ], "most_time": [ {"agent": x["agent"], "display_name": x["display_name"], "time_seconds": x["time_seconds"], "sessions": x["sessions"], "touches": x["touches"]} for x in by_time if x["time_seconds"] > 0 ], "global_models": sorted( global_models.items(), key=lambda kv: kv[1], reverse=True ), "per_agent": per_agent, } # ─── Favicon ────────────────────────────────────────────────────── FAVICON_SVG = '' @app.get("/favicon.ico") def favicon(): return Response(content=FAVICON_SVG, media_type="image/svg+xml") @app.get("/favicon.svg") def favicon_svg(): return Response(content=FAVICON_SVG, media_type="image/svg+xml") # ─── Main ───────────────────────────────────────────────────────── if __name__ == "__main__": import argparse import socket import uvicorn parser = argparse.ArgumentParser() parser.add_argument("--port", type=int, default=8081) parser.add_argument("--host", type=str, default="0.0.0.0") args = parser.parse_args() # ── Double-launch guard ─────────────────────────────────────────── # Refuse to start if the API port is already bound by another process. # Without this, two server.py instances collide on 8081 (and 8082), # producing phantom "register doesn't persist" bugs and bind crashes. probe = socket.socket(socket.AF_INET, socket.SOCK_STREAM) try: probe.bind((args.host, args.port)) probe.close() except OSError: print( f"ERROR: port {args.port} on {args.host} is already in use. " f"Agentic OS appears to be running already — stop it first " f"(./start.sh --stop) before launching another instance." ) raise SystemExit(1) uvicorn.run(app, host=args.host, port=args.port)