SKILL: Markdown Writer

Purpose: help agents reliably author and review Markdown with correct structure, links, and minimal ambiguity. This skill is intended to be used by any agent that creates/edits/reviews .md content.

When to Load This Skill

Load this skill when you:

• Create or edit any *.md file
• Write documentation changes (README, guides, design docs)
• Create or update ADRs in .github/decisions/
• Maintain repo context files such as AGENTS.md, llms.txt, or copilot-instructions.md
• Create or update SKILL.md packages or .agent.md prompts (Markdown structure still applies)
• Review documentation as a critic (links, headings, formatting)

Requirements (Must-Haves)

Agents writing or editing Markdown in this repo must satisfy all of the following:

R1 — Language and permanence

• All committed Markdown must be English.
• Avoid time-sensitive phrasing (“today/now”) unless a date is explicitly stated.

R2 — Document structure

• Exactly one top-level title: a single # ... heading.
• Headings are hierarchical and ordered: # → ## → ### (no skipped levels).
• Headings are unique (avoid duplicates that may break anchors).

R3 — Links

• Prefer relative links for repo files.
• Do not introduce broken links; if you add/change a link, verify the target exists in the workspace.
• Use / as a path separator (never \).
• Use descriptive link text (avoid dumping raw URLs unless the URL itself is the reference).

R4 — Anchors

• Avoid deep anchor links unless the section is stable.
• Keep heading titles simple to reduce renderer-specific anchor differences.

R5 — Code blocks

• Fenced code blocks must be balanced (opening/closing fences match).
• Add a language tag when possible (e.g. yaml, json, bash, text).

R6 — Frontmatter

• Only use YAML frontmatter when the file format expects it.
• If frontmatter exists, it must be valid YAML wrapped in --- lines.

R7 — Safety and hygiene

• Do not include secrets, tokens, credentials, or sensitive logs.
• Keep edits minimal: do not rewrite unrelated sections.

R8 — Lists and readability

• Use - for unordered lists.
• Keep bullets short and scannable; avoid deep nesting unless it improves clarity.

R9 — Tables

• Prefer tables only when they improve comparison; avoid very wide tables (use bullets instead).

Common Markdown Pitfalls (Avoid)

• Broken relative links after renames/moves (always re-check targets).
• Inconsistent heading casing that creates near-duplicates (hurts search and navigation).
• Over-wide tables (hard to read in GitHub; prefer bullets if the table exceeds typical width).
• Copy-pasted logs containing secrets, tokens, or internal URLs.

Templates (Use When Applicable)

Documentation page template

# <Title>

## Purpose
<What this document is for, in 1–3 sentences>

## Audience
<Who should read this>

## Workflow / Process
1. <Step>
2. <Step>

## References
- <links>

ADR template (minimal)

# ADR-XXXX — <Title>

## Status
Proposed | Accepted | Superseded by ADR-YYYY

## Context
<Why we needed to decide>

## Decision
<What we decided, one sentence>

## Consequences
<What changes; what becomes easier/harder>

Review Checklist (For Critics and Self-Review)

• One # title at top; headings are unique and ordered.
• No broken links to repo files; link paths use /.
• Fenced code blocks are balanced; language tags are present where relevant.
• The doc is skimmable: short sections, short bullets, minimal redundancy.
• No secrets, tokens, or sensitive logs embedded.
• Changes are minimal and do not rewrite unrelated sections.

Standard Checks (Optional, Recommended)

Use these when you want a repeatable “doc QA” pass. If the repo does not have these tools configured, treat this section as guidance (do a manual check instead).

• Markdown lint: catch common Markdown style issues (headings, fences, spacing).
• Spell check: catch typos and inconsistent terminology (especially in headings).
• Link check: validate external URLs (best-effort) and internal relative links.

Suggested tooling (choose one set; do not require all):

• markdownlint-cli2 (Markdown lint)
• cspell (spell check)
• lychee (link check)

Minimum manual equivalent if you cannot run tools:

• Re-scan headings: one #, no skipped levels, no duplicates.
• Re-scan all new/edited links: open targets; fix path separators and casing.
• Re-scan code fences: ensure opening/closing fences match; add language tags.

References

• Agent file and repo context standards: framework/spec/appendices/01-appendix-a1-ai-and-llm-standards.md (A1.1)
• VS Code Copilot customization overview: https://code.visualstudio.com/docs/copilot/customization/overview

Source: AlexandrSurkov/ForgentFramework — distributed by TomeVault.