agentleFS
dev-vpandeydev-vpandey/vpandey-skills/software-docs/SKILL.md

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.

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
  1. Very specific:

    Check the user's prompt for --human, --humanise, or --humanize (case-insensitive, any position).

    line 20
  2. Very specific:

    Ask: "What are you building/shipping/explaining?" then map to one of:

    line 43
  3. Distinctive:

    Ask only what's missing. Standard questions per type are in each template.

    line 59
  4. Very specific:

    Use the matching template.

    line 64
  5. Distinctive:

    Wait for the user to answer all questions

    line 73
  6. Very specific:

    Only then mark them resolved with [x] and include the rationale in the answer

    line 74
  7. Very specific:

    Do NOT skip this step or leave open questions unresolved.

    line 76
  8. Very specific:

    Update each resolved question with [x] + the decision + the reason (1–2 sentences).

    line 80

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

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 the Skill tool to invoke humanizer with the written file path as context. The humanizer rewrites the file in place at full intensity. 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 place

Step 1 — Identify doc type

Ask: "What are you building/shipping/explaining?" then map to one of:

TypeWhen to use
PRDNew feature or product initiative — defines what and why
Design DocTechnical approach — defines how
RFCCross-team proposal needing community consensus
ADRRecord of a one-way architectural decision
User GuideEnd-user instructions (internal or external)
RunbookOps procedures for recurring tasks or incidents
Release NotesWhat changed in a version, for users
ChangelogDeveloper-facing version history
PostmortemIncident 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

TypeFile
PRDtemplates/PRD.md
Design Doctemplates/DESIGN-DOC.md
RFCtemplates/RFC.md
ADRtemplates/ADR.md
User Guidetemplates/USER-GUIDE.md
Runbooktemplates/RUNBOOK.md
Release Notestemplates/RELEASE-NOTES.md
Changelogtemplates/CHANGELOG.md
Postmortemtemplates/POSTMORTEM.md

Audience Tone Guide

See AUDIENCES.md for full guide. Quick reference:

ReaderToneWhat they care about
Eng (internal)Technical, precise, terseArchitecture, edge cases, tradeoffs
Data AnalystPlain language, examplesWhat changed, how it affects their work
ExecBusiness framing, top-line onlyImpact, cost, risk, timeline
External userClear, friendly, task-focusedHow 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

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.