geronimo-iia/llm-wiki-skills/review

Claude Code and agent skill by geronimo-iia/llm-wiki-skills

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Build a prioritized review queue from lint findings, draft pages, and low-confidence knowledge; process items one at a time with guided decisions.

Install

npx vskill@latest install geronimo-iia/llm-wiki-skills/review

SKILL.md

Review

Process wiki pages that need human attention: structural errors, unverified drafts, and low-confidence active pages. The queue is rebuilt from existing fields each session — no dedicated state required.

Step 1 — Discover the wiki root

wiki_spaces_list()

Step 2 — Build the review queue

Assemble from three sources, then merge and deduplicate:

Source 1 — lint errors and warnings (highest priority):

wiki_lint()

Pages with Error findings (broken links, missing fields, unknown type) rank first — they are structurally broken, not just unreviewed.

Source 2 — draft pages:

wiki_list(status: "draft")

Source 3 — low-confidence active pages:

wiki_list(status: "active")

Filter to pages where confidence < 0.4.

Priority order:

PriorityCondition
1 — Errorwiki_lint Error finding
2 — Warningwiki_lint Warning finding
3 — Draftstatus: draft, no lint finding
4 — Low confidencestatus: active, confidence < 0.4

Within each tier: sort by confidence ascending (least certain first), then last_updated ascending (oldest first).

If the user provided a max-items argument, cap the queue at that count.

Step 3 — Review loop

Process items one at a time. For each page:

3a. Read the page
wiki_content_read(uri: "<slug>", backlinks: true)
wiki_history(slug: "<slug>", limit: 3)
3b. Show context

Present:

  • What triggered the flag (lint finding type, draft status, or confidence value)
  • Backlinks — who links to this page (load-bearing vs. orphaned)
  • Last 3 commits — how recently and how often this page was touched
3c. Guide the review decision
DecisionWhenAction
PromoteContent is correct and completeSet status: active, raise confidence to 0.7–0.9, commit
UpdateContent is partially correctEdit claims, update confidence, keep status: draft if still incomplete, commit
Resolve contradictionPage has open question from ingestVerify conflicting claims, update claims[], remove open question, adjust confidence, commit
DeferCannot verify nowLeave as-is, add a dated note in body; skip for this session
Flag for deletionPage is redundant or wrongSet status: archived, lower confidence to 0.1, add superseded_by if applicable, commit
3d. Resolve contradictions (when applicable)

When a page has an open question flagged during ingest:

  1. Read the source page cited in the open question (wiki_content_read)
  2. Compare the conflicting claims side by side
  3. Determine which is more authoritative (recency, source type, confidence)
  4. Update claims[] — add an entry for the resolved claim with the winning source
  5. Remove the open question from the body
  6. Set confidence to reflect the resolved state (raise if resolved, leave low if uncertain)
  7. Write and commit
3e. Write and commit
wiki_content_write(slug: "<slug>", content: "<updated content>")
wiki_ingest(path: "<slug-path>")
3f. Move to the next item

Step 4 — Report

After the session ends (queue exhausted or user stops):

Review session complete.
Processed: N pages
Promoted (draft → active): N
Updated (confidence raised): N
Contradictions resolved: N
Deferred: N
Flagged for deletion: N
Remaining in queue: N pages
Errors: N
Warnings: N
Drafts: N
Low confidence: N
Suggested next session: start with the N Error findings.

Key rules

  1. Never delete — use status: archived + superseded_by; deletion

removes the page from backlinks and breaks graph edges

  1. Preserve claim history — add to claims[], do not replace existing

entries; both the old and new versions should be visible

  1. Defer is valid — add a dated reason, move on; the queue rebuilds next

session from current field values

  1. One page at a time — finish each review decision before starting the

next; do not batch writes

  1. Confirm borderline promotions — if confidence would jump from < 0.4

to ≥ 0.7, confirm with the user before writing

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/geronimo-iia/llm-wiki-skills/review/badge)](https://verified-skill.com/skills/geronimo-iia/llm-wiki-skills/review)

Something wrong with this skill or its page? Tell us.