minhtran3124

feature-intake

The routing entry point for any change request. Classifies a prompt by input type, runs a 10-flag risk checklist + hard gates to assign a lane (tiny / normal / high-risk), scores confidence to decide whether a human is needed, writes the result to specs/<slug>/SUMMARY.md, and routes to the matching workflow path. Use FIRST, before xia2 / writing-plans / any edit, to decide how much ceremony and how much human oversight a task needs.

Code & Development Listed

review-diff

Use when you want to visualize what changed before review or PR — generates a Markdown review file with C4 architecture diagrams and a code walkthrough of the current git diff. Not a correctness review; for bug-hunting use /code-review.

visual-planner

Use after a PLAN.md is written, when you want to review it visually before execution (auto-invoked by /writing-plans; also standalone as /visual-planner <slug>). Renders specs/<slug>/PLAN.md to a self-contained HTML page; optional --review mode overlays graph-derived blast-radius and risk.

xia2

Portable research-first feature discovery — investigates what exists locally, upstream on GitHub, and in version-matched official docs before any implementation. Classifies risk from built-in common signals (no per-project config). Use before adding new features, capabilities, or integrations to answer what already exists and what is the lightest credible path forward.

compound

Knowledge compounding skill — transforms session learnings into persistent, discoverable documentation. Use after any session where a bug was solved, a non-obvious pattern was discovered, an architectural decision was made, or an approach was tried and abandoned (a failure worth not repeating). Trigger: /compound

Code & Development Listed

correctness-review

Run one adversarial correctness review over a diff — assumes ≥1 runtime bug exists and hunts for it (None/async/DB/auth/concurrency/contract breaks), independent of any plan or spec. Find→score→threshold(75)→classify→fix-loop. Invokable standalone on any diff, and called by subagent-driven-development as its final pre-ship gate. Not a style/cleanup pass; for that use /code-review.

create-pr

Generates a PR description template — title, summary, tasks, and notes (with an optional diagram for flow-shaped changes). Use when the user asks to write a PR description, create a PR template, or prepare a pull request write-up. Does NOT push code or create a real PR on GitHub.

executing-plans

Use when you have a written implementation plan to execute in a separate session with review checkpoints

finishing-a-development-branch

Use when implementation is complete and tests pass — runs the targeted test suite, then pushes the branch and opens a PR against the base branch. Creates the PR only; it never merges and never discards work.

subagent-driven-development

Use when executing implementation plans with independent tasks in the current session

using-git-worktrees

Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via the harness's native worktree tool or a git worktree fallback, with detection-first and safety verification

writing-plans

Use when you have a spec or requirements for a multi-step task, before touching code

brainstorming

Use when feature-intake routes here (high-risk lane, a real design fork, or ambiguous direction) - explores user intent, requirements and design before implementation. Lane routing decides WHETHER to brainstorm; this skill governs HOW.

Code & Development Listed

intent-review

Run one review that checks a finished diff against the user's ORIGINAL request verbatim — the third oracle. Deliberately blind to PLAN.md, it catches "passed the plan, passed the tests, but not what the user asked for." Findings classify as gap / excess / drift with explicit routing (fix-loop · escalate · report-only) and a residual gate. Invokable standalone on any diff that has an intent statement, and called by subagent-driven-development as its last pass before shipping. Not a bug hunt (use /correctness-review) and not a style pass (use /code-review).