software-docs
Skill4 starsChanged 6 months agoMITView on GitHub
Expert skill for writing high-quality software documentation including PRDs, Design Docs, RFCs, ADRs, User Guides, Runbooks, Changelogs, Release Notes, and Postmortems. Use when user asks to write, create, or draft any technical or product document. Triggers: "write a PRD", "create design doc", "draft RFC", "write ADR", "create runbook", "write release notes", "create changelog", "write postmortem", "write user guide", "document this feature", "create incident report". Add --human / --humanise / --humanize anywhere in the prompt to auto-humanize the final doc after writing.
In this repository12 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- cavemanSkill
- drawio-diagramSkill
- forgeSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- humanizerSkill
- karpathy-codingSkill
- teachSkill
- write-a-skillSkill
- zoom-outSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 20
Check the user's prompt for --human, --humanise, or --humanize (case-insensitive, any position).
- Very specific: line 43
Ask: "What are you building/shipping/explaining?" then map to one of:
- Distinctive: line 59
Ask only what's missing. Standard questions per type are in each template.
- Very specific: line 64
Use the matching template.
- Distinctive: line 73
Wait for the user to answer all questions
- Very specific: line 74
Only then mark them resolved with [x] and include the rationale in the answer
- Very specific: line 76
Do NOT skip this step or leave open questions unresolved.
- Very specific: line 80
Update each resolved question with [x] + the decision + the reason (1–2 sentences).
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Software Docs Skill
Guiding Principles
- Max 1 page / ~6 min read — trim ruthlessly, link to details
- Audience-first — tune every doc to its primary reader (see AUDIENCES.md)
- Interview before writing — gather context, write rough draft, iterate
- No filler — every sentence earns its place
- Markdown only — use headers, tables, code blocks; no emojis unless audience is internal-casual
Humanization Flag
Check the user's prompt for --human, --humanise, or --humanize (case-insensitive, any position).
- If flag present: strip the flag from the prompt before processing. After the doc is written to file (Step 6), output exactly
Humanizing...on its own line, then use theSkilltool to invokehumanizerwith the written file path as context. The humanizer rewrites the file in place atfullintensity. The humanized version replaces the draft — do not show the pre-humanized output. - If flag absent: proceed normally, no humanization step.
Examples: write a PRD for X --humanise · --human write a user guide · create release notes --humanize
Workflow
1. Identify doc type → load template from templates/
2. Interview user (round 1, 3–5 targeted questions)
3. Write rough draft
4. Ask open questions BEFORE finalising
5. Resolve open questions with user
6. Iterate with user until approved
7. If --human/--humanise/--humanize flag present → humanize written file in placeStep 1 — Identify doc type
Ask: "What are you building/shipping/explaining?" then map to one of:
| Type | When to use |
| PRD | New feature or product initiative — defines what and why |
| Design Doc | Technical approach — defines how |
| RFC | Cross-team proposal needing community consensus |
| ADR | Record of a one-way architectural decision |
| User Guide | End-user instructions (internal or external) |
| Runbook | Ops procedures for recurring tasks or incidents |
| Release Notes | What changed in a version, for users |
| Changelog | Developer-facing version history |
| Postmortem | Incident retrospective — blameless, action-focused |
Step 2 — Interview (round 1)
Ask only what's missing. Standard questions per type are in each template. 3–5 questions max per round. Don't ask what you can infer.
Step 3 — Write rough draft
Use the matching template. Fill every section — write TBD rather than omitting.
Flag assumptions with > **Assumption:** ...
Step 4 — Ask open questions BEFORE finalising
MANDATORY: Before presenting the doc as complete, surface every open question to the user and get answers.
- Identify all
[ ]open questions in the draft - Present each question with a brief explanation of why it matters and what the tradeoffs are
- Wait for the user to answer all questions
- Only then mark them resolved with
[x]and include the rationale in the answer
Do NOT skip this step or leave open questions unresolved. A doc with unresolved [ ] items is not done.
Step 5 — Resolve and finalise
Update each resolved question with [x] + the decision + the reason (1–2 sentences). This gives future readers the context behind the decision, not just the outcome.
Step 6 — Iterate
Present diff of changes. Stop when user says "looks good" or "ship it".
Templates
| Type | File |
| PRD | templates/PRD.md |
| Design Doc | templates/DESIGN-DOC.md |
| RFC | templates/RFC.md |
| ADR | templates/ADR.md |
| User Guide | templates/USER-GUIDE.md |
| Runbook | templates/RUNBOOK.md |
| Release Notes | templates/RELEASE-NOTES.md |
| Changelog | templates/CHANGELOG.md |
| Postmortem | templates/POSTMORTEM.md |
Audience Tone Guide
See AUDIENCES.md for full guide. Quick reference:
| Reader | Tone | What they care about |
| Eng (internal) | Technical, precise, terse | Architecture, edge cases, tradeoffs |
| Data Analyst | Plain language, examples | What changed, how it affects their work |
| Exec | Business framing, top-line only | Impact, cost, risk, timeline |
| External user | Clear, friendly, task-focused | How to use it, what to do when it breaks |
Quality Checklist (run before marking done)
- Under 6 min read (~1200 words)
- Audience is explicit in doc header
- No section is vacuous ("we will monitor the situation")
- Every action item has an owner + date
- No jargon without definition for non-eng audience
- Links to supporting docs rather than inlining
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

