fallow

name	fallow
description	Use after implementation or before review prep when changed code needs a Fallow code-quality self-review.
role	quality-gate

Fallow

Use after implementation, cleanup, or review prep when JS/TS code needs Fallow analyzer evidence.

Skill Route Index

• Implemented work / PR prep: start with changed-code audit --plain when target fit and scope are plausible.
• Current uncommitted task slice on a dirty branch: prefer scoped --root plus --base-ref HEAD; use default base for whole-branch PR review.
• Blocked PR evidence: run doctor, follow the first safe repair hint, then retry the same evidence command.
• Current-task reporting: list current-task findings first; keep pre-existing findings as count or status context.
• Changed-code review: use audit; escalate to JSON only for issue references, repair planning, structured evidence, or before/after comparison.
• Audit triage: read the attribution split first; act on introduced findings, treat inherited as base context. Zero introduced means the changeset added nothing; stop without per-finding triage.
• Introduced remove-export finding: follow the advertised Finding resolver action (the why target) for reachability evidence before treating removal as a candidate.
• Resolver evidence: act on the evidence grade, not absence of references as proof.
• Noisy non-audit scan on a skill/CLI folder: dead-code, health, and dupes carry no attribution; add-tests and remove-export run mostly false-positive against contract exports and integration-tested code; intersect with coverage per references/workflows.md before treating findings as real.
• Cleanup / refactor scan: use dead-code, dupes, or health from the request shape; use health first for bare cleanup asks.
• Readiness check: use doctor when setup, JS/TS target fit, git readiness, JSON capability, or config scope is unknown.
• Fix request: run fix-preview before source mutation.
• Apply request: stop unless current-task source-mutation authorization exists; read references/safety.md; use runner help for the apply marker.
• Suspect target: challenge the premise or retarget before treating readiness or evidence as useful.

Owner

• Repo-local front door: skills/fallow/package.json#scripts (fallow-runner).
• Runner implementation: skills/fallow/src/fallow-runner.ts.
• Public command contract, result vocab, and repair action ids: skills/fallow/src/command-contract.ts.
• Runner tests: skills/fallow/src/fallow-runner.test.ts.
• Live compatibility smoke: skills/fallow/src/fallow-runner.live.test.ts.
• Command recipes: skills/fallow/references/commands.md.
• Workflow recipes: skills/fallow/references/workflows.md.
• Safety policy: skills/fallow/references/safety.md.
• CI adoption notes: skills/fallow/references/ci.md.
• Source lineage: skills/fallow/PROVENANCE.md.

Workflow

• Challenge suspect targets before readiness checks.
• Run doctor when Fallow availability, repo shape, git readiness, or config scope is unknown on a plausible JS/TS target.
• Choose one evidence command for the current question.
• Use commands.md for mode selection and --help for exact syntax.
• Read plain summary output first for routine judgment.
• Parse runner JSON when issue references or structured evidence are needed.
• Follow runner repair hints before retrying blocked runs.
• Rerun the same evidence command after code changes.
• Report before/after summary when a rerun exists.

Verification

• Run bun --filter fallow-scripts test after runner, command-contract, or workflow behavior changes.
• Run bun --filter fallow-scripts typecheck after TypeScript edits.
• Run live smoke only when checking installed Fallow compatibility.

Safety

• Read references/safety.md before mutation.
• Treat references/safety.md as owner for apply policy, excluded behavior, and config trust.

References

• Read references/commands.md for mode selection and help pointers.
• Read references/workflows.md for self-review, audit attribution, cleanup, coverage-intersect, preview, apply, and rerun loops.
• Read references/safety.md for the mutation boundary and trust rules.
• Read references/ci.md only for adoption guidance; CI setup is reference-only in v1.