CtrlK
BlogDocsLog inGet started
Tessl Logo

document

Write structured Markdown documentation. Use when: creating a README, ADR, design doc, RFC, runbook, playbook, release notes, incident report, API doc, troubleshooting guide, FAQ, onboarding guide, contributing guide, changelog, user guide, meeting notes, Confluence page, JIRA ticket, or any technical document. Handles tables of contents, summaries, conclusions, comparison tables, decision matrices, glossaries, and diagram placeholders.

90

Quality

88%

Does it follow best practices?

Impact

No eval scenarios have been run

SecuritybySnyk

Passed

No known issues

SKILL.md
Quality
Evals
Security

Document

Produces well-structured Markdown documentation. Output is one .md file or a directory of .md files, paste-compatible with Confluence, JIRA, GitHub, Dropbox Paper, etc.

The skill is format-neutral — domain-specific links (internal wikis, design systems, vendor docs) are added based on what the agent finds in the workspace and what the user provides.

Mandatory workflow

Run every step in order; do not skip or collapse steps even if the user's prompt seems to make one unnecessary. If a step was missed, stop and go back.

  1. Preflight — context-readiness question; loop until user confirms "All provided"
  2. Pass 1 — 4 universal questions (document-type, topic, audience, length); wait for answers
  3. Pass 2 — applicable conditional questions (subtype, special-sections, TOC, diagrams); wait for answers
  4. Step 1 — load type-specific references and templates
  5. Step 2 — discover existing documentation in the workspace
  6. Step 3 — draft skeleton (headings + 1-line section summaries; no body)
  7. Step 4 — ask: single file or directory output
  8. Step 5 — present skeleton for mid-review; loop until approved
  9. Step 6 — fill in content (only after every step above is complete)
  10. Step 7 — self-review checklist; fix and re-validate on failure
  11. Step 8 — write the output

Planning phase

Run Preflight, Pass 1, and Pass 2 as separate vscode_askQuestions calls; wait for answers between each. See planning-questions.md for full payloads, conditional rules, and the diagram routing table.

Generation procedure

Step 1 — Load references

Always load:

Then load the type-specific reference and template:

Document typeReference fileTemplate
READMEreadme-patterns.mdreadme.md
ADRadr-patterns.mdadr.md
Runbookrunbook-patterns.mdrunbook.md
Playbookrunbook-patterns.mdplaybook.md
Release notesrelease-notes-patterns.mdrelease-notes.md
Incident reportincident-report-patterns.mdincident-report.md
API documentationapi-doc-patterns.mdapi-doc.md
Confluence pageconfluence-patterns.mdconfluence.md
JIRA ticketjira-patterns.mdjira.md
FAQ / Changelog / Meeting notes / Troubleshooting / Onboarding / Contributing / User guide / Design doc / RFC / Othershort-doc-patterns.mdone of: faq.md, changelog.md, meeting-notes.md, generic.md

Step 2 — Discover existing documentation

Apply linking-strategy.md: search for README.md, *.md, docs/ folders, ADRs, related vendor docs and standards. Note them for inline linking in Step 6.

Step 3 — Draft skeleton only

Produce H1 + H2 headers + a single sentence per section describing what it will contain. No body content.

If the content naturally splits into multiple files (e.g. an epic with sub-tickets, a multi-chapter guide), note this in the skeleton — Step 4 lets the user decide.

Step 4 — Choose output format

Ask the output-format question (see planning-questions.md).

If Directory: include the proposed structure (folder name, file names, one-line purpose each) in the Step 3 skeleton. Use kebab-case file names that match the subject (e.g. overview.md, spike-ai-limitations.md). No mandatory index.md. Each file follows the same formatting rules; use relative links between files.

Step 5 — Mid-review

Present the skeleton via the structure-review question (see planning-questions.md). Apply requested changes and re-present until approved. If the user adds more context, revise the structure before re-presenting. See structure-review.md for examples.

Step 6 — Fill in content

Verify every prior step is complete; if anything is missing, stop and complete it. Then with the approved structure:

  1. Fill each section based on the user's topic and audience
  2. Apply formatting-rules.md strictly
  3. Add only the special sections selected in Pass 2 Q6
  4. Insert TOC only if Q7 = Yes
  5. Handle diagrams per the routing table in planning-questions.md
  6. Embed Step 2 links inline at the first relevant mention

Step 7 — Self-review

  • Exactly one H1 per file
  • No H4 (####) — replace with bold + content
  • If TOC present, every H2 appears in it (and vice versa)
  • No unfilled {{placeholders}}
  • No relative links point to non-existent paths
  • Special sections match user's selection — no extras or omissions
  • Length matches user's preference (concise / standard / detailed)
  • If directory output: all inter-file relative links resolve; each file has exactly one H1

If any check fails, fix and re-validate before writing.

Step 8 — Write the output

Single file: write one .md with a kebab-case filename based on topic and type (e.g. adr-001-caching-strategy.md, runbook-rotate-database-credentials.md).

Directory: create a kebab-case folder (e.g. epic-adopt-ai-copilot/) and write each .md inside. Verify all relative links resolve.

Hard rules

These are enforced by the Step 7 checklist; full details in formatting-rules.md:

  • Exactly one H1 per file; never use H4 (replace with bold + content or a table)
  • TOC contains every H2 and only H2s
  • Inline links on first mention; specify the language on every fenced code block
Repository
kvokov/ai-toolkit
Last updated
Created

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.