lint
Skill5 starsChanged 35 days agoMITView on GitHub
Run harness, privacy, structural, and staleness checks.
82 other projects use a version of this.
Adopted by
Most recently taken up by
- microsoft/coreai-design-systemtheir own version, first seen Nov 15, 2026
- jakubtobiasz/second-brain-templatetheir own version, first seen Nov 15, 2026
- HurricaHjz/second-yourselftheir own version, first seen Nov 9, 2026
- PMDevSolutions/Flaviantheir own version, first seen Nov 8, 2026
- aiagentkit-labs/agentic-dev-kittheir own version, first seen Nov 8, 2026
- levi-qiao/obsidian-llm-wikitheir own version, first seen Nov 8, 2026
Used with
Projects with it also use query, ingest and 7 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ArtemXTechSkillquery
Query data from this vault. USE WHEN user asks about projects, clients, tasks, daily notes. Use grep to extract frontmatter - do NOT read full files.
Also in 17 projects that use this
JaimeGBASkillingest
Digest an external source (article, paper, transcript, URL, book chapter) and integrate its content into the knowledge base
Also in 26 projects that use this
PydanticSkillcoverage
Fetch coverage diff from Codecov for the current branch or a specific PR. Shows uncovered lines, patch coverage, and overall coverage change.
Also in 4 projects that use this
seongjaeryuSkillsave
Complete and save your current thinking to Bindly
Also in 5 projects that use this
BugRogerSkillproject
Project lifecycle management. Dispatches to Ada.
Also in 4 projects that use this
AWSSkillissue
Implement a GitHub issue end-to-end — branch, code, test, docs, and commit.
Also in 4 projects that use this
Other versions82 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 82
- ray-project/ray★ 44ksame name
- FalkorDB/FalkorDB★ 11ksame name
- FastLED/FastLED★ 7.5ksame name
- fleetdm/fleet★ 7.2ksame name
- aaddrick/claude-desktop-debian★ 5.5ksame name
- cloudposse/atmos★ 1.4ksame name
- ethereum/execution-specs★ 1.2ksame name
- adobe/S3Mock★ 1.2ksame name
- maxGraph/maxGraph★ 1.2ksame name
- jason-effi-lab/karpathy-llm-wiki-vault★ 718same name
72 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository38 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- atelierSkill
- autoevo-reviewSkill
- civSkill
- curateSkill
- daily-reflectionSkill
- decisionSkill
- digestSkill
- dineSkill
- hiSkill
- introspectSkill
- lintSkill
- prmSkill
- pushSkill
- readSkill
- reflectSkill
- reviewSkill
- triageSkill
- weeklySkill
- captureSkill
- readingSkill
- autoevo-reviewSkill
- civSkill
- curateSkill
- daily-reflectionSkill
- decisionSkill
- digestSkill
- dineSkill
- hiSkill
- introspectSkill
- prmSkill
- pushSkill
- readSkill
- reflectSkill
- reviewSkill
- triageSkill
- weeklySkill
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 2 are distinctive, and 8 are very specific to this project.
0
Common
Most of the 68 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 45
Read counts and findings (severity, code, where, message).
- Very specific: line 59
Read categories and total.
- Very specific: line 74
Add private names, places, program labels, and preference phrases that cannot be inferred from vault titles to gitignored …
- Very specific: line 88
Read wiki_dir, counts, and findings using the Phase 0 finding fields.
- Very specific: line 96
Read thresholds, counts, and notes (path, staleness, category).
- Distinctive: line 114
If all notes are active, say so in one line and move on.
- Very specific: line 130
If any user-driven fixes were made, suggest rerunning /lint manually to confirm the state.
- Very specific: line 134
Never hand-check structure: Always call scripts/lint.py.
- Very specific: line 135
Ask before fixing: All fixes route back to the user.
- Distinctive: line 136
ERROR findings block further wiki work: If a wiki entry has a parse error, do not cite it from new entries.
“Common” is measured: kept by at least 6 in 10 of the 68 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/lint — Structural + corpus-level checks over <paths.wiki>/
Also reachable via
/hi <natural language>(e.g.,/hi lint the wiki,/hi wiki audit,/hi wiki orphans). Seeharness/intents.toml[intents.lint]for the row's example phrases. Both paths execute this same procedure.
Deterministic Python pass. The LLM never hand-checks structure — scripts/lint.py is the single source of truth, mirroring the scripts/trust.py pattern.
Scope: Three passes. (0) Harness portability, $OV ingestion hygiene, and privacy checks. (1) Structural: everything under <paths.wiki>/. (2) Staleness: L2 working-layer directories (<paths.agent_findings>/, <paths.wip>/, <paths.gtd>/, <paths.preprints>/, <paths.reflections>/, <paths.research>/). Structural lint enforces the wiki schema; staleness lint surfaces L2 notes that need attention (archival, compaction, or promotion to L4).
What gets checked:
| Check | Severity | Source |
Per-note parse errors: schema items 1-10, malformed citation metadata, or dangling targets (parse-error) | ERROR | scripts/trust.py parser + resolver |
Disputed claim or fully expired evidence (claim-attention) | WARN | scripts/wiki_review.py; reasons are in autoevo-applied-* receipts. Route to the user. |
| Duplicate titles across wiki entries (breaks citation resolution) | ERROR | scripts/lint.py |
| Slug ↔ title alignment (filename stem matches H1 title) | WARN | scripts/lint.py |
| Orphan entry: no inbound citation from another wiki entry | WARN | scripts/lint.py graph topology |
| No outbound citation to another wiki entry | INFO | scripts/lint.py graph topology |
| Shared anchor without a citation edge between entries | INFO | scripts/lint.py graph topology |
url: or gist: anchor missing readwise: field (readwise-missing) | WARN | scripts/lint.py — save to Readwise with anchor-evidence tag and backfill the document ID; fix via uv run scripts/snapshot_anchors.py --apply --note "<paths.wiki>/<Title>.md" |
Technical term in claim body not in vocabulary allowlist and not matching any wiki entry title (unfounded-term) | INFO | scripts/lint.py — add term to scripts/wiki_vocabulary.txt if common knowledge, or add a wiki entry, or add a parenthetical definition inline |
Localized shadow missing for a configured language (shadow-missing) | WARN | scripts/lint.py — regenerate the shadow from the English entry. |
Localized shadow older than English source (shadow-stale) | WARN | scripts/lint.py — re-translate the localized shadow to match the updated English source |
Public configuration schema and Claude/Codex harness alignment (registry-schema, registry-read, registry-validator, source/reference/edge findings) | ERROR/WARN/INFO | harness/registry.schema.json + scripts/harness_lint.py |
$OV ingestion hygiene (missing READMEs, raw-without-digest, archive↔working-tier overlap, root-level orphans, empty .md files, suspicious top-level dirs, vault layout, duplicate Reflect titles) | INFO (advisory) | scripts/zk_audit.py — see protocols/drive-zk-ingestion.md § Post-ingestion verification |
Not checked: cross-note @anchor date consistency. Per protocols/wiki-schema.md, valid_at is the day the marker was added to its home note, so the same source being anchored from two notes on different days is the normal case.
Exit code: 0 if no ERROR-level findings, 1 otherwise. WARN and INFO never fail the run.
Process
Phase 0: Harness health
Bash: python3 scripts/harness_lint.py --jsonRead counts and findings (severity, code, where, message). Any ERROR
blocks the run until fixed; WARN and INFO are advisory. This pass owns the
AGENTS.md size and bold-marker checks; do not repeat them manually.
registry-schema identifies the source file and failing field. A
registry-validator setup error requires installing the pinned dependencies
with uv sync --locked before retrying; lint itself never installs them.
Phase 0b: $OV ingestion hygiene audit
Bash: uv run scripts/zk_audit.py --jsonRead categories and total; category rows identify where and detail.
Advisory only: never blocks the run. Exit code 0 unless $OV is missing (exit 2). Surface a one-line summary per non-empty category. Detailed listings are read on demand via uv run scripts/zk_audit.py (no --json). Source of truth: protocols/drive-zk-ingestion.md § Post-ingestion verification.
Phase 0c: Privacy leak scan
Bash: uv run scripts/privacy_check.py --json--json mode emits a document on every run regardless of exit code. Route on its action field: "proceed" → pass (WARN first on any coverage_warnings); "soft_skip" → note "privacy gate skipped (<reason>)" and continue; "abort" → ERROR, block and present each hits entry verbatim. No JSON / exit ≥ 2 without JSON → real script error: surface stderr, soft-skip.
Any non-empty hits array is an ERROR: each entry is a private identifier (filename stem, wikilink target, slug from profile/private_slugs.txt, or exact term from profile/private_terms.txt) that appears in a public-bound pathname, worktree file, or staged blob. Present each hit verbatim with its file, source, and line number. Remediation:
- Replace the private title with a generic placeholder (e.g.,
Sample Wiki Entry,Topic A). - Add private names, places, program labels, and preference phrases that cannot be inferred from vault titles to gitignored
profile/private_terms.txt, one exact term per line. - Or, if the exposure is deliberate (e.g., the title is fully public and appears as an illustrative example), add the stem to
scripts/privacy_allowlist.txtand document the rationale in the commit message.
The check is a blocking quality gate for any commit that touches tracked files when the gate ran meaningfully (no skip flag). Do not proceed to structural lint if Phase 0c returns hits. An absent exact-term sidecar is a coverage warning, not proof of a leak. Keep the semantic privacy round enabled and populate the gitignored sidecar before claiming exact identity or preference coverage.
Phase 1a: Structural lint
Bash: uv run scripts/lint.py --jsonRead wiki_dir, counts, and findings using the Phase 0 finding fields.
Phase 1b: Staleness lint
Bash: python3 scripts/staleness.py --jsonRead thresholds, counts, and notes (path, staleness, category).
Staleness findings are always advisory (no ERROR level). They surface L2 notes that have gone cold, using the formula days_since_modified / (1 + log(1 + reference_count)). Notes referenced from wiki entries or recent reflections decay slower.
Phase 2: Present
Group findings by severity. If the corpus is clean, say so and stop.
For each ERROR-level finding: show the code, file path, and message verbatim. Do not rephrase the message — scripts/trust.py and scripts/lint.py emit precise line numbers that the user needs.
For WARN-level findings: show them but mark them as non-blocking.
Summarize INFO findings (e.g., "4 entries lack outbound citations") unless the user asks for the full list.
Staleness section (from Phase 1b): present after the structural findings, under a separate heading. Group by category:
- stale notes: list paths, suggest archiving to
<paths.archive>/ - dormant notes: list paths, suggest review or compaction
- promote candidates: list paths; nightly Autoevo grows wiki entries from them
- If all notes are active, say so in one line and move on.
Phase 3: Offer fixes
For each fixable category, ask the user before acting:
| Finding code | Fix | How |
slug-mismatch | Rename file or edit H1 | Ask the user which side to change. Never rename without confirmation — downstream @cite targets key off the title. |
parse-error (e.g., missing valid_at, duplicate claim IDs, invalid ranges or dangling citations) | Edit the wiki entry | Route to the user; do not auto-edit wiki entries. |
duplicate-title | Edit one of the H1 titles | Ask the user which note keeps the title. |
orphan-entry | Add inline citation references inside related claims | Suggest supported claim links using protocols/wiki-schema.md; shared-anchor findings can identify candidates. |
shared-anchor-no-cite | Add a supported citation reference | Show the shared anchor and propose the citing claim and target. A shared source alone does not establish dependence. |
Phase 4: Rerun (if fixes applied)
If any user-driven fixes were made, suggest rerunning /lint manually to confirm the state.
Rules
- Never hand-check structure. Always call
scripts/lint.py. If the script is missing or errors, fix the script — do not re-derive the checks in the LLM. - Ask before fixing. All fixes route back to the user;
scripts/lint.pyis advisory-only. - ERROR findings block further wiki work. If a wiki entry has a parse error, do not cite it from new entries; its claims need work until it is fixed.
- WARN findings are advisory. The system keeps working; the user decides whether to act.
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.

