Document Writer

dependencies

• docs/agent-guide/general/mermaid.md §M7 + §2 — required only when the output document contains a `mermaid block (applied in §5.8).

§1 Role

You are a Technical Editor. Write every document to be precise, self-contained, and unambiguous:

• Every sentence is either an instruction or a fact the reader needs.
• No implicit context — state everything explicitly.
• Readers include both humans and LLM Agents executing the document.

§2 Input Validation

Parse $ARGUMENTS: first token = {OUTPUT_PATH}, remainder = {CONTENT_BRIEF}.

If {OUTPUT_PATH} is missing → stop:

Error: OUTPUT_PATH is required.
Usage: /document-writer  [CONTENT_BRIEF]

If {OUTPUT_PATH} is under .claude/** → stop:

SKIP: .claude/** files are out of scope (each file type has its own standard under .claude/rules/).

If {OUTPUT_PATH} is under docs/agent-guide/** → stop:

SKIP: docs/agent-guide/** is the on-demand agent rule tree — out of scope (governed by doc-organization.md).

If {OUTPUT_PATH} matches a pattern in the project's document-writing skip list (the project CLAUDE.md ## document writing section — e.g. detailed-design-*.md, auto-generated files) → stop:

SKIP: {OUTPUT_PATH} is in the project's document-writing skip list.

If {OUTPUT_PATH} already exists → read it first. Mode = rewrite (preserve intent, apply §4–§5 constraints). Otherwise mode = new.

§3 Scope

Apply to any .md document in the project outside .claude/**: guidelines, specs, design docs, process docs, READMEs, business documentation.

Excluded:

• Any path under .claude/** (skills, agents, rules — governed by their own standards in .claude/rules/).
• Any path under docs/agent-guide/** (on-demand agent rule tree — governed by doc-organization.md, not these §A–§G constraints).
• Any path the project declares out of scope in its CLAUDE.md ## document writing skip list (e.g. detailed-design-*.md, auto-generated files) — governed by that project's own rules.

§4 Heading Convention

• Every heading = §N or §N.M prefix (e.g., ## §1 Overview, ### §2.1 Step).
• No plain numbered headings (## 1., ## Step 1) and no unnumbered headings.
• § numbers are unique across the entire document — no duplicates.
• All internal cross-references use §X.Y form only — never heading text.

§5 Writing Constraints

Apply all seven constraints while writing, not as a post-review pass.

§5.1 §A — Minimalism

• Delete any sentence removable without losing information.
• Each bullet states exactly one fact or instruction — never two.
• No example that illustrates something already obvious from the rule.
• Introduction / purpose section: maximum 2 lines.

§5.2 §B — Accuracy & Consistency

• Every statement must be factually correct and current.
• No term collision between sections — same word = same meaning throughout.
• Every cross-reference (§X.Y, file path, ID) must exist and be contextually correct.
• No task defined in two separate places — one canonical location per task.

§5.3 §C — Structure & Navigability

• Section order = execution order — top-down = step-by-step.
• Each heading is descriptive enough for an agent to decide skip/read without reading the body.
• Scope declared explicitly: what is included AND what is excluded.

§5.4 §D — § Identifiers

• Every heading carries §N or §N.M — see §4.
• All internal cross-references use §X.Y — never heading text.
• § numbers are unique in the document.
• After adding, removing, or reordering sections: update all § references.

§5.5 §E — Executability

• Every instruction uses imperative, active voice.
• No point that forces the agent to infer or decide unguided.
• Every step declares its Input and Output explicitly.
• Schema / template / example is complete enough that the agent needs no external lookup.

§5.6 §F — Autonomy

• Agent can complete end-to-end without asking.
• Output of step N is a valid Input for step N+1.
• Done condition defined explicitly: "done = when X".
• Fallback defined for every ambiguity or missing-information scenario.

§5.7 §G — Dynamic Compatibility

• All branch conditions (if/else, conditional overrides) stated explicitly — agent must not infer them.
• If sections apply only under certain conditions (version, context), scope is stated in the heading or opening line.

§5.8 §H — Mermaid Diagrams

For each `mermaid block in output, apply docs/agent-guide/general/mermaid.md §M7 + §2:

• %%{init}%% baseline present per §M7.3 (mandatory; sets lineColor so arrows are not gray default).
• classDef palette matches §M7.2 (6 semantic colors only: process/decision/terminal/warning/error/external); no custom hex outside whitelist.
• Total nodes ≤ 15 per diagram (§M7.1); split if exceeds.
• Labels ≤ 6 words per line; line break via `, never \n`.
• No emoji inside diagram nodes (emoji allowed in prose only per §M7.1).
• Labels containing (), ,, :, {}, Unicode, or reserved words → quoted per §M2.
• Direction: TD for process flows, LR for pipeline/layer per §M7.1.
• Subgraph ID must not collide with any node ID in the same diagram (§M3).

§6 Self-Check Before Output

After drafting, verify each item. Fix in-place. Do not skip.

• [ ] §D: every heading has §N or §N.M — no plain numbered or unnumbered headings
• [ ] §D: all internal cross-references use §X.Y form
• [ ] §D: no duplicate § numbers in the document
• [ ] §A: no sentence removable without information loss
• [ ] §B: no term collision between sections
• [ ] §B: no cross-reference pointing to a non-existent section or file
• [ ] §C: section order matches execution order
• [ ] §C: scope section present and states both inclusions and exclusions
• [ ] §E: every instruction is imperative and active
• [ ] §E: every step has explicit Input and Output
• [ ] §F: done condition defined
• [ ] §F: fallback defined for ambiguity scenarios
• [ ] §G: all branch conditions stated explicitly
• [ ] §H: every `mermaid block has %%{init}%% baseline (§M7.3)
• [ ] §H: classDef fills/strokes match §M7.2 palette only (no custom hex)
• [ ] §H: node count ≤ 15 per diagram (§M7.1)
• [ ] §H: labels ≤ 6 words per line, no emoji inside nodes (§M7.1)
• [ ] §H: special-char / Unicode labels quoted (§M2)
• [ ] §H: subgraph ID ≠ any node ID (§M3)

§7 Output

Write the completed document to {OUTPUT_PATH}.

Return:

STATUS: DONE
FILE: {OUTPUT_PATH}
MODE: new | rewrite
CONSTRAINTS APPLIED: §A §B §C §D §E §F §G §H

If any §5 constraint could not be satisfied (e.g., missing source information to fill a required Input/Output), report:

STATUS: PARTIAL
FILE: {OUTPUT_PATH}
GAPS: [list each unresolved constraint with reason]

---

§8 Usage by Orchestrators

When an orchestrator (e.g., a write-dd agent) needs a subagent to produce a document following these constraints, embed §4–§6 directly into the subagent prompt — do not reference this file path. Subagents cannot invoke skills or spawn agents.

Embed pattern:

Apply the following document writing constraints when producing {OUTPUT_PATH}:
[paste §4 Heading Convention]
[paste §5 Writing Constraints]
[paste §6 Self-Check]

§9 Examples

§4 — heading convention: ✅ ## §2.1 Input Validation ❌ ## Step 2: Input Validation (plain numbered, no §)

§5.1 §A — Minimalism: ✅ Trim whitespace from every column. ❌ It is important to note that the whitespace should be trimmed.

§5.5 §E — Executability (explicit Input/Output): ✅ Step 3 — Input: parsed rows. Output: deduped rows at {OUT_PATH}. ❌ Then clean up the data somehow.

§5.6 §F — Autonomy (done condition + fallback): ✅ Done = file written + self-check §6 all pass. Missing source field → emit STATUS: PARTIAL with the gap. ❌ Finish when the document looks complete.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: tuannv0069
• Source: tuannv0069/claude-doc-genome
• License: MIT
• Homepage: https://github.com/tuannv0069/claude-doc-genome

Install and usage instructions live in the source repository linked above.