SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsbugfix-quick

bugfix-quick

Coding & development

View repository

Updated 10-16 01:37

About this skill

AI-consumed reference. Optimized for Claude to read during execution. Human-readable explanation: see docs/architecture/HIERARCHICAL_PLANNING.md or docs/getting-started/ depending on topic.

Quick Bugfix

For bugs only. Features/refactors → run-orchestrator.

Core Principle

NO FIXES WITHOUT ROOT CAUSE. Fix at SOURCE, not symptom.

Process (4 Steps)

Step Name Agent Statusline
S1 Investigate lead (inline) — uses Read/Grep/Bash to reproduce bugfix S1
S2 Test RED tester — writes failing test bugfix S2
S3 Fix GREEN architect / frontend / mobile (per artifact, via agent-detector) bugfix S3
S4 Verify tester — runs full suite bugfix S4

Announce per step. When transitioning, surface to the user: ─── Step S{N} · {Name} ─── Dispatching {agent}…. Also update run-state.json#current_step (one of: investigate, test-red, fix-green, verify) and #active_agent so the statusline reflects reality.

1. Investigate

  • Read error + stack trace
  • Reproduce the bug
  • git log / git diff for recent changes
  • Trace backward: where did bad data originate?
  • Context economy (rules/core/context-economy.md) — Grep for the failing symbol first, then Read with offset+limit on the matched lines only. Don't read the full file unless the bug spans it.

2. Write Failing Test (RED)

  • Test that reproduces the bug exactly
  • Pick the right layer (skills/test-writer/SKILL.md#test-type-selection). If the bug is a UI / user-flow / auth / payment regression → e2e spec via Playwright MCP, not a unit test that won't actually reproduce it. If the bug is in pure logic → unit. The test must fail for the right reason.
  • User confirms test FAILS

3. Fix (GREEN)

  • Minimal, focused change at root cause
  • User confirms test PASSES

4. Verify

  • Run full test suite — and the e2e suite if S2 wrote an e2e spec (npx playwright test / npx cypress run). No regressions in either.
  • Read output, THEN claim result. "Tests passed" with no pass count = 0 tests collected = bug.
  • Wrong: "Should work now" / Right: "Tests pass: [output]"

Red Flags (Return to Step 1)

"Quick fix for now" / "Just try changing X" / "It's probably X" / "Seems fixed"

Output

## Bug Fix: [Description]
**Root Cause:** [Why — traced to source]
**Test Added:** [file + test name]
**Fix Applied:** [file + change summary]
**Verification:** Tests pass, no regressions

If complex → switch to run-orchestrator.

Escalation: If the bug resists quick fix (intermittent, race condition, "works on my machine", multiple plausible causes), escalate to skills/deep-debugging/SKILL.md for scientific-method root-cause analysis.

Related Rules

  • rules/core/tdd-workflow.md — Failing test before fix
  • rules/core/verification.md — Run tests, read output, then claim
  • rules/core/execution-rules.md — Root-cause discipline
  • rules/core/simplicity-over-complexity.md — Minimal fix at root cause, no speculative refactoring during bug fix