ask-harness
Find which skill or flow fits your situation, with a map of every harness-kit skill.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-13
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- ask-harness
- description
- Find which skill or flow fits your situation, with a map of every harness-kit skill.
- disable-model-invocation
- true
02 资源文件 1
- 参考文档PHASE-BOUNDARIES.md2.3 KB
03 SKILL.md 正文
Ask Harness
You don't need to remember every skill. This is the map: one main flow
most work travels, the on-ramps that feed it, the lifecycle around pull
requests and issues, and the standalone tools. A name with a slash is one you
type to start (/name on Claude Code, $name on Codex); a bare name is one
the agent also picks up on its own when the task fits, and you can still ask
for it by name.
The main flow: idea to merged change
- Align.
/grill-with-docsinterviews you until every branch of the design is settled, and records the project's terms inGLOSSARY.mdand hard-to-reverse decisions as ADRs. Use/grill-mewhen there's no repo to record into. Both run thegrillingskill underneath. - Settle what talk can't. When a question needs a runnable answer, build a
cheap throwaway artifact to react to with
prototype(a state-model demo or UI variants; in a bigger effort, a wayfinder prototype ticket); when it needs outside facts, theresearchskill reads primary sources and cites them. - Specify. For work that spans sessions,
/to-specturns the conversation into a spec issue, and/to-ticketssplits it into tracer-bullet sub-issues with blocking links. Small work skips straight to step 4. - Build.
/implementtakes one ticket test-first (tdd,minimal-change) to an open PR./implement-specbuilds a whole spec's tickets in parallel worktrees onto one integration branch. - Ship.
open-prsimplifies the diff, writes the body (pr-body), runs the standards and spec review passes (review-pr,code-reviewer), and waits for CI. - Improve the environment.
/harness retrolooks back over the session for missing checks, standards, pointers, and access, and sends each lesson to the project's facts, a project-local skill, or upstream, never into a vendored file./session-retrorecords what the work itself taught. Run them in the session they look back on, before you clear it.
Keep steps 1 to 3 in one context window so the spec builds on the actual
reasoning; each /implement can then start fresh from its ticket. If the
window fills before /to-tickets (reasoning gets noticeably weaker past
roughly 150k tokens on current models), compact at the nearest phase boundary
rather than pushing on.
On-ramps
- Issues and PRs piling up from others:
/triagemoves them through triage states to agent-ready briefs, which/implementlater picks up. Tickets from/to-ticketsare ready already. - Something's broken:
diagnosing-bugsbuilds a fast check that goes red on the bug before any theory, then fixes the root cause with a regression test. - The code has got hard to change:
/improve-codebase-architecturefinds shallow modules worth deepening, shows them in a visual report, and grills through the one you pick. - Too big and foggy for one session:
/wayfindercharts a map of decision tickets and settles them one per session; when the way is clear, it hands off to/to-spec.
Around pull requests and issues
| You want to | Use |
|---|---|
| Commit with the project's checks | commit |
| Open a PR and see it through review and CI | open-pr |
| Work through or consolidate several dependency PRs | open-pr |
| Review a PR, or answer its review comments | review-pr; /pr-comments for a reply-only pass |
| Bring a branch up to date | /rebase |
| File, update, or close an issue | issue-create, issue-update, issue-close |
| Split or merge issues | /issue-restructure |
| Decide what to work on next | /groom (it also surveys shortcut-ledger) |
| Recap your day | /standup |
| Coordinate several agents or people | agent-standup; on Claude Code, budget before a big fan-out |
Vocabulary underneath
domain-modeling: sharpen the project's terms and record them.codebase-design: the deep-module vocabulary (module, interface, depth, seam) for shaping a module.
Standalone
/to-questionnaire: when the answer is in someone else's head, write them a questionnaire./wait-what: the last message didn't land; re-explain it plainly./adhd-mode: action-first responses for the rest of the session (on Claude Code, the ADHD output style does it without invoking anything)./handoff: move the work to another host, directory, or person./teach: learn a topic over several sessions, with lessons and a record of what you've learned./loop-me(experimental): find recurring work worth delegating and spec it as workflows.ui-review: accessibility and UX audit of web UI.wizard: a script that walks a person through steps only they can do, such as creating accounts and copying API keys into.envand CI secrets.documentation-writerand/svg-logo-designer: docs and logos.skill-writer,harness,/harness-builder,harness-issue: maintain the harness itself. Theharnessskill audits, bootstraps, updates, and improves it, and migrates a copy-and-edit harness to a vendored snapshot; once vendored, updates arrive as pull requests from the project's scheduled update workflow.
Between phases
At the boundary between two chunks of work (grilling done, implementation done), choose how to carry on: continue, clear, hand off, send it to a subagent, or compact. PHASE-BOUNDARIES.md has the order to ask in and why. Mid-phase, keep going or split the rest into subagents.
04 同仓库更多 Skills
teach
Teach yourself a topic over several sessions, with lessons, references, and a record of what you've learned.
svg-logo-designer
Generate SVG logos with multiple concepts, layouts, and color variations.
documentation-writer
Writes and revises human-facing documentation — READMEs, contributor and setup guides, reference pages, architecture notes, CLAUDE.md, and changelog entries. Covers picking the right document type for the question being asked, structuring for skimming, writing examples that actually run, and keeping docs from going stale. Use when creating or rewriting docs, when a README has drifted from the code, or when deciding where a piece of information belongs. For SKILL.md and agent files, use /skill-writer instead.
adhd-mode
Shapes every response for a reader with ADHD: the next action first, numbered steps, progress restated each turn, tangents held back, concrete time estimates, wins made visible. Stays on for the rest of the session until the reader says "stop adhd mode". On Claude Code, the harness-kit:adhd output style does the same without invoking this skill.
wait-what
That last message didn't land: re-explain it with the missing context.
to-questionnaire
Turn a decision you can't make alone into a questionnaire for the one person who can answer it, filled in async or together in a meeting.