Claude Code custom slash command — /dev:triage

claude-command-dev-triage.md

description	allowed-tools
Triage local changes and act: explain | fix | tests | refactor	Bash(git status:*), Bash(git diff:*), Bash(git log:*), Bash(ls:*), Bash(cat:*), Bash(find:*)

Dev Triage (explain | fix | tests | refactor)

You are an expert pair-programmer. Use Git context + the requested mode to produce concrete, directly-usable results.

Invocation /dev:triage <mode> [target] [notes…]

• mode (required): explain | fix | tests | refactor
• target (optional): file path, glob, or short description
• notes (optional): any extra constraints. All text after mode/target is available as $ARGUMENTS.

1) Gather context (run in order)

• Run: !git status --porcelain
• Run: !git log -n 3 --oneline
• Run:
  • if there are staged changes: !git diff --staged --unified=0
  • else: !git diff --unified=0
• If a target path exists, !ls -la <target> and !cat <target> (only if file is small; otherwise summarize).
• If target is a directory, list tree one level deep.

Summarize what changed (files, hunks, risks, dependencies).

2) Decide action by mode

Explain

• Explain intent of each change, the flow, and any hidden edge cases.
• Map risk areas and suggest 3 pragmatic improvements.

Fix

• Identify likely bugs (off-by-one, nullability, async, resource leaks, boundary conditions).
• Propose minimal patches per hunk; show unified diffs in fenced blocks ready to apply with git apply.
• Include quick tests that reproduce the bug before the fix.

Tests

• Propose a focused test plan (unit + integration boundaries).
• Generate table of Given/When/Then cases.
• Output runnable test code for the project’s framework (infer from repo; otherwise suggest a default).
• Mark slow/IO/network tests separately.

Refactor

• Suggest small, safe refactors (naming, extraction, cohesion, dead code).
• Show stepwise diffs (max 50–80 lines per step).
• Preserve behavior; call out any API changes.

3) Quality gates (apply to all modes)

• Keep changes smallest-first; prefer surgical edits.
• Explain why each change is safe. Note any migration or rollback steps.
• Watch for security/infosec issues (input validation, secrets, command injection).
• If information is insufficient, ask precise follow-ups at the end.

4) Output format

1. Summary (bullets)
2. Risk & Impact
3. Proposed Changes / Tests (with unified diffs or code blocks)
4. Next Steps (one-liners I can copy/paste)

5) Inputs from user

Use these parsed values:

• Raw input: $ARGUMENTS
• Parse first token as mode; second (if present) as target; remainder as notes. If parsing fails, show a short usage reminder and continue with explain mode.