Refine Gait skill context and workflow input handling

This commit is contained in:
David Ahmann 2026-02-15 16:36:15 -05:00
parent 50ddfd57aa
commit 0744f22f2a
3 changed files with 64 additions and 14 deletions

View File

@ -8,6 +8,19 @@ license: Apache-2.0
Use this skill to isolate failing CI causes with deterministic artifact checks.
## Gait Context
Gait is an offline-first runtime for AI agents that enforces tool-boundary policy, emits signed and verifiable evidence artifacts, and supports deterministic regressions.
Use this skill when:
- incident triage requires artifact-first root-cause isolation
- CI gate failures need deterministic reruns tied to a failing artifact
- evidence outputs must be generated from Gait artifacts
Do not use this skill when:
- Gait CLI is unavailable in the environment
- no Gait run/pack artifact or run identifier is available as input
## Required Inputs
- `failure_target`: failing run id, runpack path, or pack path.
@ -18,13 +31,16 @@ Use this skill to isolate failing CI causes with deterministic artifact checks.
1. Verify the failing artifact first:
- `gait verify <failure_target> --json`
2. If failure came from regress grading, rerun deterministically:
2. Enter an isolated triage workspace:
- `mkdir -p <workdir> && cd <workdir>`
3. If failure came from regress grading, bind reruns to the requested target:
- `gait regress init --from <failure_target> --json`
- `gait regress run --json`
3. If baseline evidence exists, compute deterministic diff:
4. If baseline evidence exists, compute deterministic diff:
- `gait pack diff <baseline_target> <failure_target> --json`
4. If environment health is uncertain, run diagnostics:
5. If environment health is uncertain, run diagnostics:
- `gait doctor --json`
5. Return triage summary:
6. Return triage summary:
- integrity status
- failing stage and reason codes
- diff highlights (if provided)
@ -41,12 +57,15 @@ Use this skill to isolate failing CI causes with deterministic artifact checks.
```bash
gait verify ./artifacts/runpack_failed.zip --json
gait doctor --json
mkdir -p ./triage && cd ./triage
gait regress init --from ../artifacts/runpack_failed.zip --json
gait regress run --json
gait doctor --json
```
Expected result:
- verify output reports integrity status for the target artifact
- regress init output references the requested `failure_target`
- doctor output reports actionable diagnostics
- regress output reports stable pass/fail status and failures

View File

@ -8,6 +8,19 @@ license: Apache-2.0
Use this skill to generate integrity-backed receipts that are portable across CI, tickets, and audit handoff.
## Gait Context
Gait is an offline-first runtime for AI agents that enforces tool-boundary policy, emits signed and verifiable evidence artifacts, and supports deterministic regressions.
Use this skill when:
- incident triage needs portable receipt handoff proof
- CI gate failures require receipt/evidence attachment
- evidence outputs must be generated from Gait artifacts
Do not use this skill when:
- Gait CLI is unavailable in the environment
- no Gait run/pack artifact or run identifier is available as input
## Required Inputs
- `source`: run id, runpack path, or pack path.
@ -18,8 +31,9 @@ Use this skill to generate integrity-backed receipts that are portable across CI
1. Verify integrity before receipt generation:
- `gait verify <source> --json`
2. Generate receipt from source artifact:
- `gait run receipt --from <source> --json`
3. Parse receipt fields for handoff:
- when `out_path` is provided: `gait run receipt --from <source> --json > <out_path>`
- when `out_path` is omitted: `gait run receipt --from <source> --json`
3. Parse receipt fields for handoff from `out_path` (if set) or stdout:
- `ok`, `run_id`, `manifest_digest`, `ticket_footer`, `bundle`
4. Return concise evidence block containing:
- source identifier
@ -38,12 +52,12 @@ Use this skill to generate integrity-backed receipts that are portable across CI
```bash
gait verify ./artifacts/runpack.zip --json
gait run receipt --from ./artifacts/runpack.zip --json
gait run receipt --from ./artifacts/runpack.zip --json > ./artifacts/receipt.json
```
Expected result:
- verify returns `ok=true` for intact artifacts
- receipt returns a non-empty `ticket_footer` suitable for issue or PR evidence
- receipt file contains a non-empty `ticket_footer` suitable for issue or PR evidence
## Validation Example

View File

@ -8,6 +8,19 @@ license: Apache-2.0
Use this skill to transform an observed incident into deterministic regression checks.
## Gait Context
Gait is an offline-first runtime for AI agents that enforces tool-boundary policy, emits signed and verifiable evidence artifacts, and supports deterministic regressions.
Use this skill when:
- incident triage needs repeatable regression fixtures
- CI gate failures need deterministic reproduction
- evidence outputs must be generated from Gait artifacts
Do not use this skill when:
- Gait CLI is unavailable in the environment
- no Gait run/pack artifact or run identifier is available as input
## Required Inputs
- `run_source`: run id, runpack path, or equivalent source accepted by `gait regress init`.
@ -15,15 +28,17 @@ Use this skill to transform an observed incident into deterministic regression c
## Workflow
1. Initialize deterministic fixture from the incident source:
1. Enter the declared working directory before generating artifacts:
- `mkdir -p <workdir> && cd <workdir>`
2. Initialize deterministic fixture from the incident source:
- `gait regress init --from <run_source> --json`
2. Parse fields from init output and record them:
3. Parse fields from init output and record them:
- `ok`, `run_id`, `fixture_name`, `fixture_dir`, `config_path`, `next_commands`
3. Execute regression graders:
4. Execute regression graders:
- `gait regress run --json`
4. If CI evidence is needed, rerun with JUnit output:
5. If CI evidence is needed, rerun with JUnit output:
- `gait regress run --json --junit <junit_path>`
5. Return a concise summary with:
6. Return a concise summary with:
- source identifier
- fixture directory and config path
- status and failed grader count
@ -39,6 +54,7 @@ Use this skill to transform an observed incident into deterministic regression c
## Usage Example
```bash
mkdir -p ./regress-workdir && cd ./regress-workdir
gait regress init --from run_demo --json
gait regress run --json --junit ./artifacts/junit.xml
```
@ -51,6 +67,7 @@ Expected result:
## Validation Example
```bash
mkdir -p ./regress-workdir && cd ./regress-workdir
gait regress run --json > ./artifacts/regress_result.json
python3 - <<'PY'
import json