system-check
Skill1 starsChanged 3 months agoMITView on GitHub
Mechanical checks on the second brain (broken links, orphan files, stale paths, non-conforming naming, macro/micro date discrepancies) plus, on a periodic run, the judgement layer (duplicated content, rules drifting from reality, repeated patterns in the feedback log). Use for "check the system", "run a cleanup", "check the vault".
2 other projects use a version of this.
Adopted by
Most recently taken up by
- wrg32786/aigent-ostheir own version, first seen Oct 13, 2026
- aribajahan/claude-skillstheir own version, first seen Sep 23, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- wrg32786/aigent-os★ 25same name
- aribajahan/claude-skills★ 0same name
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 repository7 other files it gives its agents.
- template/CLAUDE.mdCLAUDE.md
- knowledge-pillSkill
- librarianSkill
- onboardingSkill
- project-kickoffSkill
- session-logSkill
- specializeSkill
The file
Files
System check
v2.0
Goal: catch structural breakage with a repeatable check instead of discovering it by accident, plus the judgement the mechanical pass alone cannot see. It reports, it does not correct — fixes are decided with the user after the report.
Steps
Mechanical (1-7, deterministic — no interpretation, only comparison between files):
- Get the real date with a command — never infer it from the latest entries in the files.
- Map the structure — list (do not read) all
CLAUDE.md,memory.md,archive.md,index_*.mdand*_index.mdfiles in the vault, including every level of10_projects/. Structure only, not content: this is the list the checks below run against. - Broken or ambiguous links — for every wikilink
[[ ]]or path link in the mapped files: verify it resolves to exactly one file. Flag both links to nothing and ambiguous ones (several files with the same name). - Orphan files — files inside an indexed folder (
02_wiki/*,00_context/*,10_projects/*, etc.) that appear in no index of their category: unreachable through the cascade even though they exist on disk (Rule 1 ofCLAUDE.md). - Stale paths — references to paths that no longer exist on disk (folders renamed, moved, flattened). Compare the paths written in the files against the real tree.
- Non-conforming naming — files that break the current conventions (kebab-case for wiki notes, the two-level
*_index.md/index_<category>.mdscheme for folders — the full list lives in99_system/operating-manual.md). - Macro/micro date discrepancies — for every project in the "Active Projects" table of the root
memory.md: compare the "Last updated" date with the one in that project's ownmemory.md. Every mismatch is a candidate for missed propagation.
Judgement (8-10, these require reading and interpretation — only on the periodic run, not every execution):
- Duplicated content — the same fact written in more than one
memory.mdinstead of living in one place with a pointer elsewhere (Rule 4 ofCLAUDE.md). - Rules drifting from reality — compare the written rules (
CLAUDE.mdat every level,99_system/operating-manual.md) with the practice observed in00_context/dump/(recent period) and in thememory.mdfiles. Flag where a rule is no longer followed, or where practice has moved past the rule without the rule being updated. - Repeated patterns — read
00_context/dump/(recent period) and99_system/system-feedback.md: corrections or feedback that keep coming back to the same point are candidates to become a new rule inCLAUDE.md, instead of staying scattered reminders.
Final report — a list by category (broken links, orphans, stale paths, naming, date discrepancies, duplicated content, rules drifting from reality, repeated patterns), each with the file and line involved. No changes applied in this run.
Hard rules
- It reports, it does not correct. Fixes are decided with the user after the report, step by step or in one batch — never applied on the assistant's own initiative.
- On demand → steps 1-7 only. The judgement pass (8-10) belongs to a periodic run: it costs more and needs interpretation, not just comparison.
- Prerequisite: the whole vault root mounted, not a specific project. Steps 8-10 also need
00_context/dump/and99_system/system-feedback.mdreachable.
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.

