commit-trailers

Commit Trailers -- Structured Decision Context

Supplement conventional commits with structured trailers that preserve the reasoning behind a change. Future maintainers (and future agents) read commit history to understand not just what changed, but why a specific approach was taken.

Standard Trailers

Add these after the commit body, separated by a blank line, when they add value:

Constraint: <external limitation that shaped this approach>
Rejected: <alternative considered and the reason it was rejected>
Confidence: high | medium | low  <how certain we are this is correct>
Scope-risk: low | medium | high  <blast radius — how much can this break>
Not-tested: <what was not tested and why>

When to Use Each Trailer

| Trailer | When to Include | When to Skip | |---------|----------------|--------------| | Constraint | External limitation forced a non-obvious decision | No external constraints applied | | Rejected | A reasonable alternative was considered and discarded | Only one approach was viable | | Confidence | Correctness is uncertain or untested in key areas | Straightforward well-tested change | | Scope-risk | Change touches shared infrastructure, middleware, auth, or core data paths | Isolated, single-purpose change | | Not-tested | A gap in test coverage is intentional and known | All paths are tested |

Not every commit needs trailers. A fix: typo in error message needs none. A new auth middleware affecting every API call needs all five.

Trailer Reference

Constraint

An external limitation — browser API, hosting restriction, library limitation, business rule, or environment constraint — that ruled out the obvious solution.

Constraint: Safari 15 does not support the CSS :has() selector
Constraint: Redis unavailable in staging — in-memory fallback only
Constraint: Must stay on Node 18, no upgrade path in scope
Constraint: Third-party payment API does not support idempotency keys

Rejected

An alternative that was seriously considered. Helps future maintainers understand why the current approach exists instead of the simpler-looking one they are thinking of.

Rejected: Redux — too much boilerplate for this feature's scope
Rejected: JWT stored in localStorage — XSS exposure, using httpOnly cookie instead
Rejected: Direct DB query — bypasses cache layer, use repository instead
Rejected: Polling every 5s — too much server load, went with SSE instead

Confidence

Signal the quality of certainty around the correctness of the change. low should trigger a code review comment or discussion before merge.

Confidence: high
Confidence: medium — edge case with concurrent writes not covered by tests
Confidence: low — reversing this migration is complex, needs review before merge

Scope-risk

Signal how much of the system this change can affect if it is wrong. Helps reviewers and the team calibrate how carefully to review and test.

Scope-risk: low
Scope-risk: medium — touches payment processing, needs manual smoke test
Scope-risk: high — auth middleware change, every authenticated endpoint affected

Not-tested

Explicitly acknowledge a test gap. This is honest engineering, not an excuse. A named gap is better than an invisible one.

Not-tested: email delivery — no SMTP server in test environment
Not-tested: concurrent request behavior — requires load testing setup
Not-tested: Safari-specific rendering — no CI browser suite yet
Not-tested: DB rollback on payment failure — requires Stripe test mode setup

Full Example

feat: add rate limiting to API endpoints

Implement token bucket algorithm for /api/* routes.
Limits: 100 requests/min for free tier, 1000 for paid tier.
Exceeding the limit returns 429 with Retry-After header.

Constraint: Redis required for distributed rate limiting — in-memory does not work across multiple instances
Rejected: express-rate-limit — lacks per-user tier support without significant custom logic
Confidence: high
Scope-risk: medium — affects all /api/* routes, auth and webhook routes excluded
Not-tested: sustained 10K+ req/min behavior — requires dedicated load test environment

Format Rules

• Trailers go AFTER the commit body, separated by a blank line
• One trailer per line, no line wrapping
• Rejected may appear multiple times (one per rejected alternative)
• Only include trailers that add genuine value — avoid mechanical inclusion
• Confidence: low is a signal for the team, not a reason to skip the commit
• This format supplements, not replaces, the conventional commit type prefix

Integration with Other Rules

• safety-and-quality.md defines the base commit format (<type>: <description>)
• This rule adds optional trailers on top of that format
• full-autonomy.md requires commit approval — trailers give Batuhan the context to approve confidently
• Trailers are the commit-level equivalent of Priority Notes in the notepad system