agentleFS
anh-chuanh-chu/llm-wiki-pm/skills/llm-wiki-review/SKILL.md

llm-wiki-review

Skill7 starsChanged 9 days agoMITView on GitHub

Independent review of every answer or page built from the wiki, before it reaches the user. A separate reviewer (Claude Opus by default, another Claude model, Codex CLI, or a command the user sets) checks each statement against its sources, searches for what the answer left out, and checks the reasoning; then you revise. Run it before reporting any wiki-based answer, page write, brief or analysis.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write your draft answer or page, exactly as you would send it, to a temp file outside the wiki:

    line 14
  2. Very specific:

    Put the user's question on the first line as Question: ..., then the draft.

    line 20
  3. Distinctive:

    Run the script next to this file (its path is this skill's base directory).

    line 24
  4. Very specific:

    Add a second argument only when the user named a reviewer in this request ("review with codex" → "$DRAFT" codex).

    line 30
  5. Very specific:

    If the script fails, say so and run it again with claude:opus.

    line 43
  6. Distinctive:

    If you change a statement, the source must support the new text.

    line 49
  7. Distinctive:

    Delete the draft file.

    line 59

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

LLM Wiki Review

Sub-skill of llm-wiki-pm. A wrong fact in a PM answer is expensive: "customer A asked for X in Q4" gets repeated in roadmaps and customer calls. Tests on product memos showed a separate reviewer followed by a revision cut errors by about three quarters; a self-check by the author gained less. So the reviewer is always a separate context, never you re-reading your own draft.

1. Write the draft down

Write your draft answer or page, exactly as you would send it, to a temp file outside the wiki:

DRAFT=$(mktemp /tmp/wiki-review-draft.XXXXXX.md)

Put the user's question on the first line as Question: ..., then the draft. Keep every [source: ...] marker and [[wikilink]]: the reviewer checks against them. A page write is reviewed before you report it; if you already wrote it, the review may still change it.

2. Run the review script

Run the script next to this file (its path is this skill's base directory). Copy this line exactly; the script picks the configured reviewer itself:

WIKI_REVIEWER_OPTION='${user_config.reviewer}' "<this skill's directory>/run-review.sh" "$DRAFT"

Add a second argument only when the user named a reviewer in this request ("review with codex" → "$DRAFT" codex). Never pass a reviewer otherwise, not even claude.

The script resolves the wiki, fills reviewer-prompt.md, runs the reviewer and prints the review. Run it once per turn: after you revise, do not run it again unless the user asks or the revision brought in sources the first review never saw. Always use it: do not review in your own context or start your own subagent instead, and do not choose a different model. The review-gate hook only accepts a turn that ran this script.

Reviewer values (the user's choice for this turn, else WIKI_REVIEWER, else the plugin option reviewer, else claude:opus):

ValueReviewer
claudea separate read-only Claude process, account default model
claude:haiku, claude:sonnet, claude:opusthe same on that model (claude:opus is the default)
codexCodex CLI in a read-only sandbox
anything elsea shell command that reads the prompt on stdin and prints the review

If the script fails, say so and run it again with claude:opus; never skip the review silently.

3. Revise

For each point the reviewer raises, decide: accept (fix the draft) or reject (the reviewer is wrong; say why in one line). Do not accept a point just because a reviewer made it: check it against the source yourself. Reviewers are usually right, but not always. If a point is about something missing and the missing fact is real, add it with its source.

If you change a statement, the source must support the new text. Never resolve a point by softening wording until it is unfalsifiable.

4. Report

Send the revised answer. End it with one line:

Reviewed by <reviewer>: <n> points, <a> fixed, <r> rejected. followed by anything still open (an unverified claim, a source the reviewer could not reach, a gap only the user can fill).

For a page write, also append to log.md: ## [YYYY-MM-DD] review | <page> | <reviewer>: <n> points, <a> fixed, <r> rejected.

Delete the draft file.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.