Agent Skills: Quick Skill

Zero-ceremony atomic task execution. Use for tiny fixes or enhancements (minutes, not hours) that don't warrant SDLC ceremony. Flags: --discuss (disambiguate first), --full (delegate to /pre-commit after execution), --issue N (link commit to GitHub issue).

UncategorizedID: matteocervelli/llms/quick

Install this agent skill to your local

pnpm dlx add-skill https://github.com/matteocervelli/llms/tree/HEAD/.codex/skills/quick

Skill Files

Browse the full folder contents for quick.

Download Skill

Loading file tree…

.codex/skills/quick/SKILL.md

Skill Metadata

Name
quick
Description
"Zero-ceremony atomic task execution. Use for tiny fixes or enhancements (minutes, not hours) that don't warrant SDLC ceremony. Flags: --discuss (disambiguate first), --full (delegate to /pre-commit after execution), --issue N (link commit to GitHub issue)."

Quick Skill

Purpose

Atomic task execution without SDLC ceremony. Fills the gap between "raw Claude" and /fix//implementation:

  • No opusplan model switch
  • No task list creation
  • No TDD by default
  • No PR creation
  • Issue GitHub optional

Use when a change takes minutes, not hours.

When to Use vs Other Skills

| | /quick | /fix | /implementation | | -------------- | -------------- | ----------------- | -------------------- | | Ceremony | None | Minimal | Full (TDD, opusplan) | | Size | Tiny (minutes) | Small-medium | Medium-large | | Type | Anything | Bug / tech debt | Feature / refactor | | Issue required | No | No | No | | PR created | No | Optional (--pr) | Yes | | Tests required | No | Regression only | Full TDD |

Usage

/quick fix the failing auth test
/quick --discuss add rate limiting to the API   # disambiguate before executing
/quick --full refactor the user service         # run /pre-commit after execution
/quick --issue 42 update error message          # link to GitHub issue in commit

Workflow

Bare mode (default)

  1. Snapshot — Run git status and git branch --show-current to establish context
  2. Execute — Implement the minimal change
    • Trivial edit (1-2 files): work on current branch
    • Slightly larger: git checkout -b <type>/brief-description
  3. Smoke tests — After executing, always output a short checklist (3–5 items) the user can run manually before committing. Infer from modified files:
    • REST endpoint → curl commands (happy path + one error case)
    • Frontend → numbered browser steps
    • CLI → shell commands
    • Library/module → python -c "..." one-liner
  4. Gate — Run /security-verify scan before committing (non-negotiable per rules/security-gate.md)
  5. Commit — Atomic commit with conventional-commits message:
    • With issue: fix: <description> + Fixes #42 in commit body
    • Without issue: fix: <description> or feat: <description>
  6. Review suggestion — After committing, always suggest:

    "Run /review changes to get a companion second opinion before moving on."

  7. Track — Append entry to .claude/memory/local/quick-tasks.md

--discuss mode

Before executing:

  1. Ask 2-3 targeted disambiguation questions (scope, approach, edge cases)
  2. Save decisions to .claude/memory/local/CONTEXT.md
  3. Proceed with bare mode

--full mode

After executing:

  1. Delegate to /pre-commit (quality + tests + security + changelog)
  2. Use this when the change touches logic where regression is a real risk

Constraints

  • Stay on current model — no /model opus suggestion
  • No TaskCreate calls
  • No TDD unless --full flag
  • No gh pr create — use /fix or /implementation if a PR is needed
  • Security scan always mandatory before commit

Tracking: quick-tasks.md

After each execution, append to .claude/memory/local/quick-tasks.md:

### YYYY-MM-DD HH:MM — <task description>

- Branch: <branch-name> or "current branch (no new branch)"
- Commit: <sha> "<message>"
- Mode: bare | discuss | full
- Issue: #N or —

File is session-scoped and gitignored (memory/local/ is never committed).

Tips

  • If the change grows beyond ~30 minutes of work, switch to /fix or /implementation
  • --discuss is especially useful when the task is ambiguous or touches multiple areas
  • --full is for logic changes (not just config/text) where test coverage matters