Decide fallow gating: run fallow audit in CI / pre-push, and whether it replaces knip

[glary-bot]

Follow-up split out of #13284 (docs/adr/0011-adopt-fallow.md), requested in #13284 (comment).

Problem

After #13284, fallow is configured (.fallowrc.jsonc + .fallow-baselines/) and runnable (pnpm fallow, pnpm fallow:audit), but nothing enforces it. The gates still belong entirely to knip:

• CI: pnpm knip runs in .github/workflows/ci-lint-format.yaml via .github/actions/lint-format-verify — a required check.
• Local: .husky/pre-push runs pnpm knip, with an explicit CI bail-out because bot workflows populate ComfyUI/ and contaminate knip's project glob.

So we carry two overlapping tools, and the fallow-only coverage (duplication, complexity, diff-aware audit) can regress unnoticed. ADR 0011 records this as undecided:

For now Fallow runs through the pnpm scripts and sits alongside knip. We have not wired fallow audit into CI or the pre-push hook, and we have not retired knip.

Decisions needed

1. Where fallow audit runs. Own job vs. a step inside CI: Lint Format. Note ci-lint-format.yaml is mirrored by ci-lint-format-queue.yaml for merge_group, and the workflow/job names must match on both sides for branch protection to resolve one required check — a new job needs the same treatment.
2. Whether audit needs the base branch. fallow audit --gate new-only diffs against the base, so CI checkout depth / base-ref fetching has to be set up accordingly.
3. Pre-push or not. Does fallow audit join pnpm knip in .husky/pre-push, replace it, or stay opt-in? Runtime on a full local tree is the deciding factor; the same CI bail-out applies.
4. Knip's fate. Retire it once fallow covers the same findings (unused files/exports/deps), or keep both permanently and document which one owns what. If it is retired, knip.config.ts, the knip/knip:no-cache scripts, the dependency, the pre-push hook, and .github/actions/lint-format-verify all need updating, and the knip exception ledger must be fully represented in .fallowrc.jsonc.
5. Baseline hygiene. Who re-saves .fallow-baselines/* as code is cleaned up, and how a stale baseline is prevented from masking new instances of an already-baselined pattern in an already-baselined file (a known hole called out in the ADR).

Acceptance criteria

• A decision is recorded in docs/adr/0011-adopt-fallow.md (status moves off "Proposed", or a follow-up ADR supersedes it).
• Whatever is decided is implemented: the CI/pre-push wiring exists (with the merge_group mirror kept in sync), or the ADR states explicitly that fallow stays advisory and why.
• No duplicate enforcement is left undocumented: for each finding class (unused files, unused exports, unused deps, duplication, complexity), exactly one tool is named as the gate.

Context

• ADR: docs/adr/0011-adopt-fallow.md (added in chore: make repo fallow-compliant #13284), "Notes" section.
• Prior art on gradual rule enablement: [Tech Debt]: Gradually enable knip rules for better code health #4888 (knip), Prepush Knip Checks Unstaged #5443 (pre-push knip staging behaviour).
• Blocked on chore: make repo fallow-compliant #13284 landing.

[coderabbitai]

🔗 Related PRs

#13275 - chore: remove unused dependencies [open]
#14152 - fix(ci): stop backport partial failures from discarding clean backports [open]
#14429 - feat(cicd): fail when fixes sit unshipped on a consumed release line [open]

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!