wiki-lint-semantic
Skill2 starsChanged 10 days agoMITView on GitHub
Run the judgement-requiring health checks on the KB — whether a comparison truly covers the key difference dimensions (L4), whether cross-page conclusions contradict, whether recent kb-save output adds real value (Q2) — emitting Review-Queue proposals only, never editing any wiki page. Use when the user says "do a semantic health check / check for contradictions / see whether the comparison pages are complete". Deterministic lint (L1/L2/L3/L5/L6/broken-links/duplicates) is handled by scripts/pipeline.py lint, not here.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- ingestSkill
- kb-postmortemSkill
- kb-qaSkill
- kb-querySkill
- kb-reviewSkill
- kb-saveSkill
- pipeline-doctorSkill
- skill-evolveSkill
- source-preflightSkill
- source-xraySkill
- ingestSkill
- kb-postmortemSkill
- kb-qaSkill
- kb-querySkill
- kb-reviewSkill
- kb-saveSkill
- pipeline-doctorSkill
- skill-evolveSkill
- source-preflightSkill
- source-xraySkill
- wiki-lint-semanticSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 33
Follow-up handling goes to kb-review: this skill only files proposals.
- Distinctive: line 68
No wiki content page edited directly.
≈ “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.
Files
wiki-lint-semantic — semantic lint (the half the finishing CLI cannot do)
Deterministic lint (L1/L2/L3/L5/L6/broken-links/duplicates) is handled by python scripts/pipeline.py lint;
this skill does only the judgement-requiring part. It emits Review-Queue proposals only and never directly
edits any wiki content page. The execution layer is scripts/pipeline.py; this skill only orchestrates.
1. Triggers / Non-triggers
- Triggers: "do a semantic health check", "check for contradictions", "see whether the comparison pages are complete", "check whether the kb-save output adds value".
- Non-triggers: broken links / missing sections / bare evidence IDs / orphan pages and other deterministic lint (use
python scripts/pipeline.py lint); read-only Q&A (usekb-query); handling an existing Review-Queue (usekb-review); coverage, Q-chain, formula/evidence spot-checks and other broad QA/audit requests belong tokb-qa— do not hijack its triggers.
2. Inputs
wiki/comparisons/**,wiki/topics/**,wiki/synthesis/**,wiki/overview.md, and the relevant concept/lesson/source pages.- Recent
kb-savequery-sessions (especiallydecision.md,candidate_write_set.json,evidence_refs.json). - The existing
wiki/Review-Queue/, to avoid duplicate proposals.
3. Outputs
wiki/Review-Queue/semantic-lint-<YYYY-MM-DD>.md.- Each proposal carries: page path, problem type (L4 / contradiction / Q2), evidence, suggested fix direction, whether a user decision is needed.
- It does not directly edit comparison/topic/synthesis/concept/lesson/source content pages.
4. Dependencies
- CLI: may run/reference
python scripts/pipeline.py lint --source <source_id>for the deterministic result, but does not re-implement L1/L2/L3/L5/L6/broken-link/duplicate rules. - Protocols:
docs/skill-runtime/schema.md(page roles / per-type frontmatter contract),save-back-policy.md(Q2 save-value judgement). - Follow-up handling goes to
kb-review: this skill only files proposals.
5. Persisted artifacts
wiki/Review-Queue/semantic-lint-<YYYY-MM-DD>.md: the semantic-lint proposals.- A chat summary: scope checked, number of findings, suggestion to handle via
kb-review.
6. CLI commands
python scripts/pipeline.py lint --source <source_id>
python scripts/pipeline.py statusThese only confirm the deterministic gate / state background; semantic judgements become Review-Queue proposals, never a direct promote or content edit.
7. Workflow
| Sub-unit | Input | Output | Acceptance | Persisted | Failure stop |
| L1 scope | request + vault index | page set to check | focus on comparison/contradiction/Q2, not broad QA | check list | vault missing |
| L2 comparison L4 | comparison pages | key-dimension gaps | covers assumptions/conditions/outcomes/costs | proposal draft | page evidence thin |
| L3 contradiction | related concept/topic/lesson | conflicting claim pairs | claims locatable under one concept/model | proposal draft | source not locatable |
| L4 Q2 | kb-save output + evidence | added-value judgement | distinguishes new synthesis from restating | proposal draft | evidence missing |
| L5 write proposals | proposal drafts | semantic-lint file | each has path/problem/evidence/fix direction | Review-Queue | duplicate proposal |
8. Failure stops / recovery
vault or index missing; deterministic lint already failed and blocks the semantic scope; a page lacks
evidence to judge; the finding already has an open proposal; the user asks to edit content pages directly
(switch to kb-review and wait for confirmation). Recovery: proposals are the durable output; hand off
to kb-review.
9. Acceptance criteria
- No wiki content page edited directly.
semantic-lint-<YYYY-MM-DD>.mdwritten to Review-Queue.- Each proposal has a page path, problem description, evidence, and a suggested fix direction.
- L4/Q2/contradiction judgements do not duplicate deterministic-lint responsibilities.
- The user was prompted to handle the proposals via
kb-review.
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
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.

