Agent-Skills.md

Agent Skills: Write Product Documentation

Create fact-based product documentation from code and release evidence. Use for manuals, operations, installation, specifications, acceptance, release notes, FAQs, or claim audits.

UncategorizedID: aAAaqwq/claude-code-skills/write-product-documentation

Repository

aAAaqwq/claude-code-skills
aAAaqwqLicense: MIT
8821

Install this agent skill to your local

pnpm dlx add-skill https://github.com/aAAaqwq/AGI-Super-Team/tree/HEAD/skills/write-product-documentation

Skill Files

Browse the full folder contents for write-product-documentation.

Download Skill

Loading file tree…

skills/write-product-documentation/SKILL.md

Skill Metadata

Name
write-product-documentation
Description
Create fact-based product documentation from code and release evidence. Use for manuals, operations, installation, specifications, acceptance, release notes, FAQs, or claim audits.

Write Product Documentation

Write from verified product behavior, not aspiration. Give each document one audience and one job, while preserving a shared fact base.

Workflow

  1. Freeze scope. Record product version, commit/tag, platform, deployment mode, audience, and date. Do not mix future features or another branch into current instructions.
  2. Build a fact matrix. Map each claim to code, configuration, test, screenshot, runtime evidence, and known limitation. Mark facts as verified, inferred, planned, or unknown.
  3. Choose the document role. Use a product description for value and boundaries, a full user manual for end-to-end operation, a daily guide for routine use, a technical specification for implementation, and an acceptance document for testable outcomes.
  4. Design the user journey. Lead from prerequisites to first success, normal operation, exceptions, recovery, data handling, and support. Explain parameters in business terms.
  5. Write exact side-effect semantics. State what preview/dry-run does for each operation. Distinguish “processed,” “request sent,” “artifact received,” “verified,” and “persisted.”
  6. Separate platforms and versions. Put Windows/macOS differences where users encounter them. Never let an old platform prerequisite leak into a new implementation.
  7. Use honest product language. Explain value first, but avoid unsupported detection, compliance, ROI, security, or availability guarantees. State human decision points and external dependencies.
  8. Create acceptance traceability. Turn every important promise into a precondition, action, expected result, evidence, and failure criterion.
  9. Validate the deliverable. Check links, headings, terminology, screenshots, version strings, commands, privacy, accessibility, rendering, and consistency against the fact matrix.
  10. Retire stale documents. Archive superseded evidence with a replacement link; do not leave multiple files claiming to be the current source of truth.

Writing rules

  • Use the reader's language before internal implementation names.
  • Define one term once and reuse it consistently.
  • Give copyable commands only after verifying them on the scoped version.
  • Keep screenshots synthetic or redacted and bind each screenshot to the documented version/platform.
  • Distinguish product boundary, known limitation, expected degradation, and defect.
  • Do not hide real side effects under “preview” wording.
  • Keep legal/privacy/security claims aligned with the current architecture and policy review.

References

  • Read document-system.md to select document types, structures, and validation gates.
  • Read project-evidence.md for examples and contradictions found in this repository's documentation set.