--- name: gait-capture-runpack description: Capture and verify deterministic Gait runpacks from normalized run input. Use when asked to record a run, produce run_id or runpack artifacts, generate ticket-ready proof, or validate artifact integrity before handoff. disable-model-invocation: true --- # Capture Runpack Execute this workflow to record an artifact safely and deterministically. ## 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 captured run artifacts for reproduction - CI gate failures need verifiable artifact identity and integrity - receipt/evidence generation depends on runpack metadata and digests Do not use this skill when: - Gait CLI is unavailable in the environment - no Gait run input or artifact source is available ## Workflow 1. Validate required input path: ``. 2. Run record with JSON output (required): - `gait run record --input --json` 3. Parse output fields: - `ok`, `run_id`, `bundle`, `manifest_digest`, `ticket_footer` 4. Verify artifact integrity (required): - `gait verify --json` 5. Return a concise handoff block that includes: - `run_id` - `bundle` - `manifest_digest` - `ticket_footer` - verify status ## Safety Rules - Keep default capture mode as `reference`. - Do not switch to raw capture unless explicitly requested. - For replay workflows, prefer `gait run replay` (stub mode default); require explicit unsafe flags for real tool replay. - Do not invent `run_id`, digests, or verify results. - Treat non-zero exit from `gait run record` or `gait verify` as blocking errors. - Keep this skill wrapper-only: do not parse policy YAML or implement custom evaluator logic. ## Determinism Rules - Always use `--json` and parse structured fields. - Do not rely on text-only output for workflow decisions. - Keep output grounded in recorded artifact values only.