跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
其他技能

ask-supermatt

Ask which skill or flow fits your situation. A router over the skills in this repo.

svyatov · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add svyatov/supermatt --skill ask-supermatt

技能说明


name: ask-supermatt description: Ask which skill or flow fits your situation. A router over the skills in this repo. disable-model-invocation: true argument-hint: "[your situation]" license: MIT

Ask SuperMatt

You don't remember every skill, so ask.

Skills are named here as /name, which is also the Cursor and Claude linked-skill form. From the Claude Code plugin, type /supermatt:name: a bare /code-review runs Claude Code's bundled review, not this one. In Codex, type $name; in OMP, type /skill:name. Preserve arguments in every form.

A flow is a path through the skills. Most paths run along one main flow, and four on-ramps merge onto it. Everything else is standalone, or a vocabulary layer that runs underneath.

The main flow: idea → ship

The route most work travels. You have an idea and want it built.

  1. /grill-with-docs sharpens the idea by interview. Start here whenever you are working in a working directory: it's stateful, retaining what it learns in GLOSSARY.md and ADRs. (No working directory? Use /grill-me instead, covered under Standalone. Both run the same /grilling primitive; grill-with-docs is the one that leaves a paper trail, which makes it the better of the two whenever a repo is there to leave it in.)

  2. Branch: can you settle every question in conversation? If a question needs a runnable answer (state, business logic, a UI you have to see), detour through a prototype, bridged by /handoff in both directions (the prototype is a side task forked mid-phase, one of the narrow cases /handoff is for; see Phase boundaries):

    • /handoff out, then open a fresh session against that file,
    • /prototype to answer the question with throwaway code,
    • /handoff back what you learned, and reference it from the original idea thread.
  3. Branch: is this a multi-session build?

    • Yes → /to-spec (turn the thread into a spec), then /to-tickets to split it into independently green tracer-bullet tickets, each declaring its blocking edges. For per-ticket work, start /implement on a ticket whose blockers are done, resetting context between tickets with the host's native control or a fresh session. On a local tracker the tickets are files under .scratch/<feature>/issues/; hosted trackers use native blocking links when available, otherwise the configured Blocked by: fallback. For a whole-spec build, use /implement-spec: native subagents work the ready frontier in separate worktrees and integrate into one branch for a single PR, without Herdr. It runs QA and review over the whole spec and leaves shipping to you. Keep tracker issues open until the default-branch merge; dependencies inside that run advance when their commits reach the integration branch.
    • No → /implement right here, in the same context window.

    Both implementation flows drive /tdd internally (focused checks during red-green cycles, then the full required gate for a completed behavior), commit, run /qa in a fresh subagent or session when available (otherwise in the current context with the independence gap reported), and close out with /code-review, a three-axis review (Standards, Spec, Adversarial) of the diff. QA drives the running program the way its user would, a terminal program in tmux or a web app in a browser, and checks every requested behavior. /code-review reads docs/agents/cross-model.json: host-only keeps every reader on the current host, auto uses only hardened Claude or Codex peers, and explicit modes can select Claude, Codex, Cursor Agent, or OMP under their disclosed limits. /implement verifies each ticket; /implement-spec verifies the combined integration branch against the whole spec. Follow-up QA and review carry the prior reports and check the fixes and affected behavior. When the work goes up as a pull request, /pr shapes the body: a summary diagram, before/after evidence, and the merge danger. Reach for /tdd on its own when you just want to build a concrete behaviour test-first without a full spec, /qa on its own whenever you want to see a branch actually work before anyone reads its code, and /code-review on its own whenever you want to review a branch or PR against a fixed point.

  4. Ship. /fix-findings applies the findings of the last review or check (a finding's text works too, from another session). /refactor the changes on this branch since origin/main cleans up what the branch added without changing behavior, and stops when the code is already clean. /ship-pr commits, pushes, opens the pull request, diagnoses and repairs CI within bounded attempts, squash merges only on green checks, verifies post-merge workflows, and closes the issues the branch finished, which unblocks the tickets waiting on them. /babysit-pr is the same road without the merge: it opens the PR when the branch has none, handles review feedback before CI, keeps the branch current, and stops at a ready PR for you to merge; run it again on the same PR to resume with the same budgets. Both recover CI through /recover-ci: two repair cycles and one transient rerun. /commit commits on the current branch when you are not ready to ship. Then /retro on the session, so the next one starts from what this one learned.

Both implementation flows consult codebase-design before changing data contracts, state transitions, shared mutable state, or retryable effects. Adversarial review checks duplicate execution and recovery after interruption against the intended contract.

Unattended: /orchestrate

When the issues are triaged and you want the queue worked without you, /orchestrate runs the main flow per issue from inside a Herdr pane in Claude Code, Codex, Cursor, or OMP, when Herdr can start that host's corresponding CLI. It reads GitHub issues or the configured local Markdown queue; local ticket

阅读完整来源 ↗