Lambenthan/empiricalwiki / i18n/en/skills/check
check
VerifiedScan the full wiki to detect health issues and produce a tiered fix-recommendation report (covers all 9 entity types + graph consistency)
/check
Scans the full wiki to detect structural, link, field, and graph health issues, and generates a tiered fix-recommendation report. Covers all 9 entity types, including claims confidence plausibility, idea failure-reason completeness, experiment-claim link validity, and graph edge consistency.
Inputs
- Full wiki directory (default
wiki/) - Optional:
--jsonflag (output JSON format via tools/lint.py) - Optional:
--fixflag (auto-fix deterministic issues) - Optional:
--fix --dry-run(preview fixes without applying them) - Optional:
--suggestflag (show recommendations for issues that cannot be auto-fixed)
Outputs
- Lint report (reported directly to the user)
- Optional file write:
wiki/outputs/lint-report-{date}.md
Wiki Interaction
Reads
wiki/papers/*.md— paper page fields and linkswiki/concepts/*.md— concept page fields and linkswiki/topics/*.md— topic page fields and linkswiki/people/*.md— people page fields and linkswiki/ideas/*.md— idea status, failure_reason, origin_gapswiki/experiments/*.md— experiment status, target_claim, outcomewiki/claims/*.md— claim confidence, status, evidence, source_paperswiki/Summary/*.md— survey page fieldswiki/graph/edges.jsonl— semantic graph edge consistency checkwiki/graph/citations.jsonl— bibliographic citation consistency checkwiki/index.md— cross-check page completeness
Writes
- Does not directly modify wiki content (reports only, no fixes)
wiki/log.md— records lint result summary viatools/research_wiki.py log
Workflow
Pre-conditions: confirm the working directory is the wiki project root (directory containing wiki/, raw/, tools/).
Set WIKI_ROOT=wiki/.
Step 1: Run the Automated Lint Tool
Default mode (report only):
python3 tools/lint.py --wiki-dir wiki/ --json
Auto-fix mode (when user specifies --fix):
python3 tools/lint.py --wiki-dir wiki/ --fix --json
Auto-fixes deterministic issues (xref reverse-link completion, missing fields filled with default values) and outputs a fix report.
Preview mode (when user specifies --fix --dry-run):
python3 tools/lint.py --wiki-dir wiki/ --fix --dry-run --json
Previews what would be fixed without applying any changes.
Parse the JSON output to obtain all automatically detected issues (and fix results).
Step 2: Structural Completeness (automated coverage)
The automated tool checks:
- Broken wikilinks:
[[slug]]target file does not exist - Orphan pages: pages with no incoming links
- Missing required fields (all 9 entity types):
- papers: title, slug, tags, importance
- concepts: title, tags, maturity, key_papers
- topics: title, tags
- people: name, tags
- Summary: title, scope, key_topics
- ideas: title, slug, status, origin, tags, priority
- experiments: title, slug, status, target_claim, hypothesis, tags
- claims: title, slug, status, confidence, tags, source_papers, evidence
Step 3: Field Value Validation (automated coverage)
- Enum value checks:
- papers.importance ∈ {1,2,3,4,5}
- concepts.maturity ∈ {stable, active, emerging, deprecated}
- ideas.status ∈ {proposed, in_progress, tested, validated, failed}
- ideas.priority ∈ {1,2,3,4,5}
- experiments.status ∈ {planned, running, completed, abandoned}
- experiments.outcome ∈ {succeeded, failed, inconclusive}
- claims.status ∈ {proposed, weakly_supported, supported, challenged, deprecated}
- Claim confidence ∈ [0.0, 1.0]
- Idea failure_reason: must be non-empty when status=failed (anti-repetition memory)
- Experiment target_claim: the referenced claim must exist
Step 4: Cross Reference Symmetry (automated coverage)
Check all bidirectional link rules defined in CLAUDE.md:
| Forward link | Reverse link checked |
|---|---|
| concepts.key_papers → papers | papers.Related contains concept link |
| papers → people (wikilink) | people.Key papers contains paper |
| claims.source_papers → papers | papers.Related contains claim link |
| ideas.origin_gaps → claims | claims.Linked ideas contains idea |
| experiments.target_claim → claims | claims.evidence contains experiment |
Step 5: Graph Edge Consistency (automated coverage)
- JSON format validity: every line is valid JSON
- Required fields: each edge has from, to, type
- Edge type validity: semantic edges use the current endpoint-aware type sets; legacy paper-paper / paper-concept types produce migration warnings
- Edge confidence:
/ingestpaper-paper and paper-concept semantic edges useconfidence: high|medium|low - Citation layer:
graph/citations.jsonlrows usetype: cites, valid source/date, paper endpoints, and no confidence field - Dangling nodes: wiki pages referenced by from/to must exist
Step 6: Content Quality (LLM-assisted)
Items detectable by the automated tool:
- Papers with importance=5 have no concept page referencing them
- Concepts with maturity=stable have only 1 key_paper
- Topics have empty Open problems sections
Additional LLM judgments (requires reading content):
- Concept near-duplicate detection: scan all concept page titles + aliases and assess whether any pairs are semantically identical or highly similar (e.g. "attention mechanism" and "self-attention"). Output merge recommendations for suspected duplicates.
- Contradictory statement detection (inconsistent descriptions of the same fact across different pages)
- SOTA records not updated in over 6 months
- people Recent work not updated in over 6 months
- Claim confidence inconsistent with evidence quantity/strength
- High-priority idea stuck in proposed status for a long time
Step 7: Generate Report
Output sorted by priority:
## Lint Report — YYYY-MM-DD
**Summary**: N 🔴, M 🟡, K 🔵
### 🔴 Fix Immediately
1. [file] — {issue description}
### 🟡 Recommended Fixes
1. [file] — {issue description}
### 🔵 Optional Improvements
1. [file] — {issue description}
Classification:
- 🔴 Fix Immediately: broken links, missing required fields, invalid enum values, failed idea without failure_reason, invalid JSON in edges, confidence out of range
- 🟡 Recommended Fixes: xref asymmetry, dangling graph edges, broken claim references, unknown edge types
- 🔵 Optional Improvements: orphan pages, quality suggestions, empty sections
Append log:
python3 tools/research_wiki.py log wiki/ "check | report: N 🔴, M 🟡, K 🔵"
Constraints
- Report-only by default: without
--fix, only reports, no modifications --fixonly repairs deterministic issues: xref reverse-link completion, missing fields filled with safe default values. Non-deterministic issues output recommendations (--suggest) for user approval- raw/ is read-only: do not modify files under
raw/ - graph/ is read-only: lint does not modify graph files, checks consistency only
- LLM judgments labeled by source: automated checks and LLM judgments are clearly distinguished in the report
- Idempotent: running multiple times produces the same result (unless wiki content changes)
Error Handling
- wiki/ does not exist: report error and suggest running
/init - graph files do not exist: skip the missing graph-file checks, note in report
- Partial directory missing: skip checks for missing directories, list missing directories in report
Dependencies
Tools(via Bash)
python3 tools/lint.py --wiki-dir wiki/ [--json] [--fix] [--dry-run] [--suggest]— automated structural check + fix (core dependency)python3 tools/research_wiki.py log wiki/ "<message>"— append logpython3 tools/research_wiki.py stats wiki/— get statistics (optional, for the report)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |