agit AGENTS.md

AGENTS.md

This repo uses agit. Use agit commit instead of git commit.

Setup

go install github.com/Madhurr/[email protected]
agit init

How to commit

agit commit \
  -m "feat: add rate limiting" \
  --intent "..." \
  --confidence 0.85 \
  --tried "approach: reason" \
  --risk "severity:area:description" \
  --unknowns "..." \
  --ripple "..."

What to write in each field

--intent — why you made this change, not what it does. The commit message says what. Intent says why.

• ❌ "implement rate limiting middleware"
• ✅ "protect the API from abuse without adding a Redis dependency"

--confidence — how sure you are the implementation is correct. Be honest.

• 0.95+ only if you've tested it and there are no known gaps
• 0.70–0.85 for most changes: logic is sound but edge cases exist
• < 0.70 if you're making assumptions or skipping something important
• Do not default to 0.9 every time. Calibrate it.

--confidence-rationale — one sentence on why you gave that score. What did you verify? What didn't you test?

• "core path tested, token refresh not covered"
• "logic is straightforward but not benchmarked under load"

--tried — other approaches you considered and rejected. Format: "approach: why rejected".

• "session cookies: needs shared store, won't work across pods"
• "Redis rate limiter: adds infra dependency we don't have"
• Only include things you actually considered. Don't fabricate alternatives.

--risk — what could go wrong. Format: "severity:area:description". Severity: low, medium, high.

• "high:data-loss:rollback deletes rows, no soft delete"
• "medium:performance:N+1 query on user lookup, not optimized yet"
• "low:ux:error message is generic, could be more specific"
• Flag real risks. Don't write "low:none:no risks".

--unknowns — things you weren't sure about and didn't resolve.

• "token revocation strategy not decided"
• "unclear if this needs to be idempotent"
• "haven't checked behavior when user has no permissions"
• This is not a weakness. It's useful information.