what
Use when a user explicitly wants a five-line recap: session, branch, ticket, next options, narrowed to any named subject.
- License 许可
- MIT
- 所属仓库
- tony/skills
- 最近更新
- 2026-11-08
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- what
- description
- Use when a user explicitly wants a five-line recap: session, branch, ticket, next options, narrowed to any named subject.
- allowed-tools
- [ "Bash", "Read", "Grep", "Glob" ]
- argument-hint
- [what you are confused about]
02 资源文件 1
- 参考文档agents/openai.yaml248 B
03 SKILL.md 正文
What
The reader is disoriented. End that in five lines.
Read ../../references/brief.md first — it defines the
five-line budget, what earns a line, the option lines, and the evidence
tiers that keep this cheap enough to ask casually.
User arguments: $ARGUMENTS
Procedure
1. Read what was actually asked
Empty $ARGUMENTS means the whole situation: what this session has been
doing, and where the repository stands.
An argument narrows it to that subject, and narrowing is the point —
/situate:what the test failure answers about the test failure and
spends no lines on branch position.
2. Answer from the session first
When the confusion is about what just happened (a command that ran, a decision taken, why the work moved), the answer is already in context and needs no tools.
When the session's memory is thin, gather only the cheap evidence relevant to the question, per the brief reference. For a named file, inspect its status and diff; Git records what changed, not why someone left it uncommitted. Do not infer that motive.
For the whole situation or a branch-position question, gather:
- Position: run
python3 ../../scripts/position.pyfor branch, trunk, ahead/behind, uncommitted work, and stashes.../../scripts/resolves from this skill's directory: write the absolute path literally in the command. Adotfile_repoor an empty repository (has_commitsfalse) gets one line saying so and no file list; the script prints nothing else then. - Commit subjects since the base:
git log --no-merges --format='%h %s' <trunk>..HEAD.
A null trunk or a failed command means that layer is unavailable; mention
it only when the answer depends on that layer. Measuring from trunk is approximate on a
branch stacked over another branch, because the range carries the
parent's commits; say so rather than re-resolving the base.
Go beyond that only for the paid tier in the brief reference.
3. Write the brief
Five lines is a ceiling; a narrow question usually needs two. Rank lines
as the brief reference sets out, then add numbered single-line options
only when there is a real fork; each option is a different action, never a
restatement of another. Stay read-only, per the contract in
../../references/situation-sweep.md, and write per
../../references/prose-rules.md.
Before sending, hold each line to ../../references/brief.md (its
Banned and Confidence sections), plus these:
- One sentence per line, at most about 20 words.
- A narrow question gets only the causal answer, with no branch or trunk name inside an option ("commit it", not "commit it on main").
- No absolute paths, and tool state (
ghsigned out) is never a line. - If the work has wandered from its purpose, offer
/situate:refocusas an option; do not diagnose it here.
Output
The body, then the options. Nothing else: no title, no framing sentence, no offer to go deeper.
Reply in this shape, as plain text outside any code fence, without bullets, bold, or local paths:
⚠ CI is red on PR #12: the type-check job fails on three files this branch added.
feat-brief-mode, 4 commits ahead of main, 2 files uncommitted.
This session rewrote the output contract and has not re-run the gates since.
The ticket asks for a terse mode; the terse mode works, the failing types do not.
1. Fix the three type errors and push
2. Show the failing job output
3. Stash the uncommitted work and check whether the branch alone is green
04 同仓库更多 Skills
serene-bliss
Use when asked to make a CLI --help, docs page, error message, or tooling UX more serene or calmer for developer experience; returns text, no edits.
review
Weave code review: run independent adversarial reviews in parallel, then synthesize findings
refine
Use when asked to refine a draft, iterate on code until cleaner, or polish a README through repeated critique and improvement.
prompt
Weave prompt: compare independent adversarial implementations, then adopt the single best one
plan
Weave planning: compare independent adversarial plans, then synthesize the best one
fix-review
Fix weave review findings: validate, add test coverage, fix, and commit each as atomic changes