From c25b8c2f22faa69ecc583cf04f402486ee0d035c Mon Sep 17 00:00:00 2001 From: "fatih.bulut" Date: Fri, 26 Jun 2026 00:05:37 +0300 Subject: [PATCH] feat(clone-app): wire Phase 7 trigger + Phase 8 fidelity pass into SKILL Co-Authored-By: Claude Opus 4.8 --- plugins/clone-app/skills/clone-app/SKILL.md | 66 ++++++++++++++++---- plugins/clone-app/tests/test-skill-phases.sh | 7 +++ 2 files changed, 62 insertions(+), 11 deletions(-) diff --git a/plugins/clone-app/skills/clone-app/SKILL.md b/plugins/clone-app/skills/clone-app/SKILL.md index 817ceae..9f2e331 100644 --- a/plugins/clone-app/skills/clone-app/SKILL.md +++ b/plugins/clone-app/skills/clone-app/SKILL.md @@ -216,27 +216,68 @@ $WORK/clone-report-.md ## Phase 7: Decision Gate -Ask: "Report saved to `$WORK/clone-report-.md`. Proceed to build the -implementation plan?" -- **Yes** → proceed to Phase 8 to assemble the clone build spec, then hand off - to `superpowers:writing-plans`. -- **No** → stop; the report stands on its own. +Ask: "Feasibility report saved to `$WORK/clone-report-.md`. Proceed to +build the implementation plan? (This runs the deep **fidelity pass** — full +API payloads, in-app logic, navigation graph, and an inferred backend design — +and produces a second report.)" +- **Yes** → run Phase 8: the fidelity pass, then assemble the build spec and + hand off to `superpowers:writing-plans`. +- **No** → stop; the feasibility report stands on its own. The fidelity pass + (and its token cost) is never incurred. -## Phase 8: Assemble the Clone Build Spec +## Phase 8: Fidelity Pass + Build Spec + +Read `${CLAUDE_PLUGIN_ROOT}/skills/clone-app/references/fidelity-pass-guide.md`. + +### Phase 8a — Fidelity subagent (deep extraction) + +Dispatch one subagent (Agent tool, `general-purpose`). It reuses what Phase 2 +already decompiled to `$WORK/output` — **no re-download, no re-decompile**. Pass +it `$PKG`, `$WORK`, the clone-app scripts dir `$CA` +(`${CLAUDE_PLUGIN_ROOT}/skills/clone-app/scripts/`), and the paths to +`fidelity-pass-guide.md`, `logic-capture-guide.md`, `backend-recon-guide.md`. +Its instructions: + +1. **Full Tier-2 payloads.** Extend `$WORK/payloads.json` so every first-party + endpoint carries request/response/headers (third-party stays Tier-1). +2. **In-app logic.** Run + `python3 "$CA/extract-logic.py" "$WORK/output" --out "$WORK/logic-signals.json"`, + then write `$WORK/logic-digest.md` per `logic-capture-guide.md`. +3. **Navigation graph.** Run + `python3 "$CA/extract-nav-graph.py" "$WORK/output" --out "$WORK/nav-graph.json"`. +4. **Backend recon.** Write `$WORK/backend-recon.md` per `backend-recon-guide.md`. +5. **Unity (if RE Method indicated Unity).** Deepen `$WORK/unity-digest.md` with + game mechanics / formulas per `unity-re-guide.md`. +6. **Return** a short fidelity summary + the artifact paths — never raw sources. + +If the subagent fails, retry once; if it still fails, continue with whatever +artifacts exist and note the gap in the fidelity report. + +### Phase 8b — Fidelity report + +Write `$WORK/fidelity-report-.md` (actual run date): summarize the +logic digest, navigation graph, full API surface, and backend recon, each with +its confidence. This is a standalone report alongside the feasibility one. + +### Phase 8c — Build spec Read `${CLAUDE_PLUGIN_ROOT}/skills/clone-app/references/clone-build-spec-template.md`. -Assemble `$WORK/clone-build-spec.md` filling every section from the artifacts: +Assemble `$WORK/clone-build-spec.md`, filling every section from the artifacts: - §2 from `$WORK/design-tokens.json` (+ `design-digest.md`), -- §3 one entry per screen, each paired with `$WORK/screenshots/NN.png`, -- §5 from `$WORK/payloads.json`, §6 data model from the RE digest, +- §3 one entry per screen, each paired with `$WORK/screenshots/NN.png`, plus its + logic from `$WORK/logic-digest.md`, +- §3b user-flow diagrams from `$WORK/logic-digest.md`, +- §4 from `$WORK/nav-graph.json`, +- §5 from `$WORK/payloads.json` (full Tier-2), §5b + §6 from `$WORK/backend-recon.md`, - §7 asset inventory from `$WORK/output` (or `$WORK/game-assets/` for Unity), - §8 acceptance criteria per screen + flow, - §10 absolute paths to every `$WORK/` artifact. Use the **Game variant** sections when RE Method indicated Unity. Then invoke `superpowers:writing-plans`, passing `$WORK/clone-build-spec.md` as -the spec (NOT the feasibility report). The build spec is the single standalone -input — a fresh session with it + `$WORK/` can build the clone. +the spec and citing BOTH `$WORK/clone-report-.md` and +`$WORK/fidelity-report-.md` as reference. The build spec + `$WORK/` is the +standalone input — a fresh session with it can build an exact / near-exact clone. ## Error Handling Summary | Scenario | Action | @@ -254,3 +295,6 @@ input — a fresh session with it + `$WORK/` can build the clone. | Unity build detected | run IL2CPP/Mono branch + AssetRipper | | Unity tool missing | continue, partial digest, RE Method `limited: unity-no-tools` | | No screenshots on Play | note it, rely on design-tokens + web image search | +| Phase 7 = No | stop after feasibility report; skip the fidelity pass | +| Fidelity subagent fails | retry once, then continue with partial artifacts and note the gap | +| extract-logic/nav-graph finds nothing (Flutter/RN) | note low confidence, lean on screenshots + API contract | diff --git a/plugins/clone-app/tests/test-skill-phases.sh b/plugins/clone-app/tests/test-skill-phases.sh index c419392..844ca04 100644 --- a/plugins/clone-app/tests/test-skill-phases.sh +++ b/plugins/clone-app/tests/test-skill-phases.sh @@ -19,5 +19,12 @@ has "clone-build-spec.md" has "clone-build-spec-template.md" has "unity-re-guide.md" has "design-capture-guide.md" +has "extract-logic.py" +has "extract-nav-graph.py" +has "logic-digest.md" +has "nav-graph.json" +has "backend-recon.md" +has "fidelity-report-" +has "fidelity-pass-guide.md" exit $fail