# Jk Docs Review

> Verify project documentation against reality and report findings for review

- Skill: `janklan/jk-docs-review` (Agent Skill)
- Install (CLI): `npx skillmds@latest add janklan/jk-docs-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/janklan/jk-docs-review/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Docs & Writing
- Author: janklan (https://skillmd.com/u/janklan)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/janklan/jk-docs-review

---


Scope: every documentation file in this project (README, CLAUDE.md, docs/, other Markdown)
plus code comments that state facts about behaviour. Open the artefact with the list of
files reviewed so omissions are visible.

Method: read each documentation file in full and extract every falsifiable claim. Verify
the claims in parallel with subagents: read the referenced code verbatim; where reading
cannot settle a claim, run the relevant command. Never rely on conversation history or
prior knowledge. A claim you cannot settle gets verdict UNVERIFIED with the reason - never
a guess. While you're at it, remove all mannered prose. Instead of "a parameter worth varying,"
the mannered writer produces "a dial worth turning." Instead of "this point still matters,"
they write "this point earns its keep." The phrases exist to display the writer, not to convey 
the idea, and readers can tell. That is why mannered prose irritates: it makes the reader work
harder so the writer can perform. It is also imprecise. Metaphors drag in connotations the writer
did not choose and cannot control. The fix is to say what you mean. When a literal phrase is 
available, use it.

Findings: false or outdated statements, ambiguities, verbose prose, LLM slop. Every finding
must quote the claim, cite its file and line, and cite the evidence (code location or
command output). A finding without evidence must not appear.

Identifiers: letter = category (F false, O outdated, A ambiguous, V verbose/slop), number =
sequence. Number proposed solutions within their finding (F3.1, F3.2).

Output: a reviewable HTML artefact, not a terminal response. One row per finding: id,
location, quoted claim, evidence, proposed solutions. Factual findings first, style after.
End with the UNVERIFIED list.