gait-incident-to-regression

name	gait-incident-to-regression
description	Convert a Gait run artifact into a deterministic regression workflow. Use when asked to initialize fixtures from run_id or runpack path, run graders, produce CI-friendly outputs, or summarize drift and failures.
disable-model-invocation	true

Incident To Regression

Execute this workflow to transform an observed run into repeatable CI checks.

Gait Context

Gait is the offline-first policy-as-code runtime for AI agent tool calls. It enforces tool-boundary policy, emits signed and verifiable evidence artifacts, and supports deterministic regressions.

Use this skill when:

• incident triage needs repeatable fixture creation
• CI gate failures require deterministic grader reruns
• receipt/evidence generation depends on regression outputs

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

Workflow

1. Resolve source run artifact:
   • use <run_id> or <runpack_path>
2. Initialize fixture deterministically (required):
   • explicit path: gait capture --from <run_id_or_path> --json
   • then gait regress add --from ./gait-out/capture.json --json
   • legacy fallback: gait regress init --from <run_id_or_path> --json
3. Parse and report:
   • ok, run_id, fixture_name, fixture_dir, config_path, next_commands
4. Run regression suite (required):
   • gait regress run --json
5. If CI output is requested, add JUnit:
   • gait regress run --json --junit junit.xml
6. Return concise summary:
   • source run
   • fixture path
   • pass/fail status
   • failed graders count
   • output paths

Safety Rules

• Keep replay deterministic defaults.
• For replay workflows, prefer gait run replay (stub mode default); require explicit unsafe flags for real tool replay.
• Do not pass --allow-nondeterministic unless explicitly requested.
• Treat non-zero regress run exits as regressions, not soft warnings.
• Keep this skill wrapper-only: no inline grading logic and no policy-evaluator behavior outside CLI calls.

Determinism Rules

• Always create a deterministic fixture before regress run for new incidents.
• Always consume --json output fields for decisions.
• Keep fixture names stable and explicit when user provides naming constraints.