Validate Fix

SkillDev tools

Makes your agent prove a bug fix actually works by rerunning failed tests, running the whole suite, typechecking, and linting.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the Validate Fix skill

About this skill

Prove a fix works before declaring done, re-run the failing test, run the full suite, typecheck, lint, and harden against recurrence. Use after implementing a bug fix, when user says "is this fixed?", or before closing an investigation.

What this skill tells your AI

The instructions your AI receives, as published by danielvm-git/bigpowers in skills/validate-fix/SKILL.md and read by ahel’s review.

HARD GATE — Fix must not regress. Run full test suite and manual UAT before declaring success.

Prove the fix works. "I think it works" is not evidence. Run the suite, show the output, then harden against recurrence.

Two-commit red/green policy (e45s08) — Bug fixes follow the same two-commit discipline as develop-tdd: first commit adds/adjusts the failing test (test(<scope>): …), second commit applies the fix (fix(<scope>): …). Do not squash RED and GREEN before review.

Checklist

1. Re-run the originally failing test

# Run the specific test that captured the bug
<test command for the failing test>
  • Previously failing test now passes

2. Run the full test suite

# Run all tests — no filtering
<full test command from CLAUDE.md>
  • All tests pass (zero regressions)

3. Type check

<typecheck command>
  • No type errors introduced

4. Lint

<lint command>
  • No lint violations introduced

5. Harden against recurrence

For every bug fixed, add at least one prevention layer:

MechanismWhen to use
Type guardInput could be the wrong shape
Schema validation (Zod, Pydantic, etc.)External data crossing a boundary
Invariant assertionInternal state that must always hold
Lint rulePattern that's easy to repeat by mistake
Environment check at startupMissing config causes silent failure
  • At least one hardening mechanism added
  • Hardening mechanism is tested

5b. Generalize-fix (HARD GATE — e80s04 / GH #98)

Sweep the defect class across the codebase after local hardening — see REFERENCE-generalize-fix.md.

  • Defect class documented (not just the one-line root cause)
  • Grep sweep run and match_count recorded
  • verify-generalize-sweep.sh passes on the artifact

6. Update the bug file and registry.yaml

Find the most recent specs/bugs/BUG-*.md file and append the resolution:

## Resolution

**Fixed:** [date]
**Root cause confirmed:** [one sentence]
**Fix applied:** [what was changed]
**Hardening added:** [type guard / schema / assertion / lint rule]
**Evidence:** all tests pass (`<verify command>`)
**Commit:** `fix(<scope>): <description>`

Also update the corresponding row in specs/bugs/registry.yaml: set status to fixed, fill in files_changed, approach, risk_level, commit_message, and any other resolution fields.

  • specs/bugs/BUG-*.md updated with resolution
  • specs/bugs/registry.yaml row updated with resolution fields

7. Behavioral Proof (HARD GATE)

Mechanical verification (tests passing) is only half the fix. You must prove behavioral correctness.

  • Manually demonstrate the fixed behavior (e.g., via run_shell_command or web_fetch)
  • Compare the output/state against the "Expected Behavior" in the bug file
  • Show the user evidence of the behavior, not just the test logs

Rules

  • Loop until behavioral correctness is verified: if any checklist item fails, or if the behavior is still incorrect despite passing tests, return to step 1 and run all checks again from the top — do not declare done until every item is green and the behavior is proven correct in a single run.
  • Never use @ts-ignore, as any, or // eslint-disable to "fix" a bug — these suppress the symptom without fixing the root cause
  • Never mark the task done if any test is still failing
  • The verify command from specs/bugs/BUG-*.md or the active epic task verify field must pass

Suggest next skill: audit-code → commit-message.

Verify

→ verify: grep -q 'generalize-fix' skills/validate-fix/SKILL.md && test -x scripts/verify-generalize-sweep.sh && bash scripts/verify-generalize-sweep.sh --self-test && echo OK

Signals

GitHub stars
248
Forks
19
Last commit
Sep 2026
Advanced
Item type
skill
Key
validate-fix
Source
github.com/danielvm-git/bigpowers