document-authoring
Author a real document — a report, proposal, slide deck, spec, or paper — by settling audience, claim, and structure before drafting prose, then writing to that structure. Use when asked to write or draft a document rather than answer a question.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-05
- Stars
- ★ 212
- Forks
- 33
01 前言 Front Matter
- name
- document-authoring
- always
- false
- triggers
- write a document, draft a document, draft a report, write a report, slide deck, build a deck, presentation outline, write a proposal, write a spec, whitepaper, outline a paper, structured document, executive summary, author a doc, one pager, !commit message, !write a test, !write a script
- description
- Author a real document — a report, proposal, slide deck, spec, or paper — by settling audience, claim, and structure before drafting prose, then writing to that structure. Use when asked to write or draft a document rather than answer a question.
03 SKILL.md 正文
Document authoring
A document that starts as prose becomes a document nobody can edit, because the structure is buried in the sentences. Settle the frame first; the prose is then mechanical.
Settle four things before writing a sentence
- Audience and their decision. Who reads this, and what do they do differently afterwards? "Inform the team" is not a decision. "Approve the migration budget" is. If there is no decision, the document is reference material — say so, and structure it for lookup instead of persuasion.
- The claim. One sentence that could be wrong. If your thesis cannot be disagreed with, you have a topic, not a claim.
- The shape. Pick one and commit:
- Decision memo — claim, options, recommendation, risks, ask.
- Report — question, method, findings, limitations, implications.
- Proposal — problem, cost of inaction, approach, plan, resources.
- Spec — contract, behaviour, edge cases, out of scope.
- Deck — one idea per slide, headline is the takeaway not the topic.
- The evidence you actually have. List it before drafting. A structure your evidence cannot fill is where invented specifics come from.
Then draft
- Front-load. Conclusion first. A reader who stops after the first paragraph should still have the answer.
- Headline the takeaway. "Latency regressed 40% after the cache change" beats "Latency analysis". This applies to every section heading and every slide title.
- One idea per unit — per paragraph, per slide, per bullet.
- Name the limitations in the document, not in a follow-up conversation. An unstated limitation reads as a claim you did not make.
- Cut the frame. "This document will discuss…" and "In conclusion…" are scaffolding. Remove them once the structure holds on its own.
Slides differ from prose
A slide is a visual argument. If a slide needs a paragraph, it is two slides or it belongs in an appendix. Speaker notes carry the nuance; the slide carries the claim and the one piece of evidence that supports it.
Worked example
Asked for "a document about our test flakiness".
Settling it first:
| Audience / decision | Eng leads — decide whether to fund a two-week stabilization sprint |
| Claim | Flaky tests cost us more engineering time than the sprint would, so the sprint pays back inside a quarter |
| Shape | Decision memo |
| Evidence I have | 6 weeks of CI reruns, 3 named flaky suites, the rerun-minutes total. No per-engineer interruption data — so I will not claim a focus cost |
The resulting skeleton:
Recommendation: fund the two-week stabilization sprint (the ask, first)
What it costs us now: 41 rerun-hours over 6 weeks, concentrated in 3 suites
Why it compounds: a suite people rerun by reflex stops being a signal
The plan: quarantine, root-cause the 3, add a vacuity check to each rail
What I cannot show: per-engineer interruption cost — no data collected
Risk: two weeks not spent on the roadmap; mitigated by taking it after the release
Only now write prose, and only to fill that skeleton. Note the fourth row of the table doing real work: it forced a limitation into the document instead of an overclaim into the argument.
04 同仓库更多 Skills
web-verify
Verify a built web page or app actually renders and works — drive it in a real browser, check the console and network, confirm the thing you changed is the thing being served, and capture evidence. Use before claiming a frontend change works.
visual-output
Emit rich visual output in chat via <widget> tags — styled HTML cards, charts, tables, interactive tools, and drawn SVG/canvas diagrams, illustrations, and animations — theme-aware, centered, and responsive instead of clashing or cramped.
task-and-project
Track work as Tasks (the /api/tasks CRUD + native task provider), and decompose larger projects into independent, verifiable units — fanning parallel investigation out to subagents while keeping one coherent plan; hand open-ended goal-driven work to a goal loop.
research-campaign
Run a multi-cycle research campaign — grill the question, decompose it into answerable sub-questions, investigate each against sources, then synthesize with confidence and gaps stated. Use for open questions that need real investigation rather than one lookup.
memory-discipline
When and how to persist durable memory — save genuine user preferences, facts, and corrections; NEVER persist transient environment failures (a failed command, a flaky network call, a one-off tool error) as lessons; recall before re-asking.
loop-worker
Per-cycle protocol for an autonomous goal loop worker — the unified autonomous goal engine that runs one self-directed cycle per turn until its goal is met, then self-retires. Drives one cycle against a loop's file interface (status.json gate, brief.md goal, guidance.txt nudges, findings/cycle_NNN.json, FINDINGS.md, the type-appropriate deliverable, STOP sentinel). The worker produces work and reports evidence; a deterministic check or a separate judge — never the worker — decides done-ness.