file-organization
Skill34 starsChanged 37 days agoMITView on GitHub
Use WHENEVER creating, saving, or generating a new file (task output, research artifact, report, chart, log, dataset, brief, dispatch) inside a shared agent/human workspace. Decides the correct sub-folder + naming so files are organized for fast AI grep AND human browsing — never dumped loose at a directory root. Also governs how tooling must glob so the structure can deepen without breaking.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- neilghosh/skillstheir own version, first seen Nov 6, 2026
- nimadorostkar/Claude-Skills-collectiontheir own version, first seen Oct 13, 2026
- z-Zihan/Evirtheir own version, first seen Oct 13, 2026
- nbdesai1992/goalposttheir own version, first seen Jul 26, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- nimadorostkar/Claude-Skills-collection★ 33same name
- z-Zihan/Evir★ 11same name
- nbdesai1992/goalpost★ 0same name
- neilghosh/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 repository31 other files it gives its agents.
- .github/AGENTS.mdAGENTS.md
- tui/AGENTS.mdAGENTS.md
- actionability-passSkill
- agent-memory-opsSkill
- blocked-page-retrievalSkill
- brain-consistency-auditorSkill
- brainstorm-panelSkill
- concurrent-push-integrationSkill
- curation-auditSkill
- decomposed-local-researchSkill
- dual-model-reconciliationSkill
- eval-integritySkill
- fleet-model-routingSkill
- fleet-operational-runbookSkill
- generate-review-fix-loopSkill
- guard-target-correctnessSkill
- honesty-stop-gateSkill
- large-file-cartographerSkill
- local-lane-build-loopSkill
- local-model-onboardingSkill
- model-routing-tableSkill
- multi-agent-code-workflowSkill
- peer-drop-watcherSkill
- protected-function-guardSkill
- research-dispatchSkill
- research-verificationSkill
- research-workflowSkill
- shared-brain-bookkeepingSkill
- should-weSkill
- task-dependency-sequencingSkill
- verify-running-buildSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
1
Distinctive
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
- Distinctive: line 12
If both pull in different directions, keep the location predictable (AI) and the *name* descriptive (human).
- Very specific: line 15
Read the map first: Check the nearest AGENTS.md (or equivalent context-map doc) and any INDEX.md for the directory's structure.
“Common” is measured: kept by at least 6 in 10 of the 4 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
File Organization — where every new file goes
Loose files at a directory root are forbidden. Every file you create lands in a typed sub-folder with a clear name. The structure serves two consumers at once:
- AI grep: PREDICTABLE locations (a
FINAL_*report is always underresearch/finals/) → an agent greps the right place first try. - Human browse: clear names, shallow trees, related things together → a person finds a thread by eye.
If both pull in different directions, keep the location predictable (AI) and the name descriptive (human). Never sacrifice predictability for cleverness.
Decision procedure (before writing any file)
- Read the map first. Check the nearest
AGENTS.md(or equivalent context-map doc) and anyINDEX.mdfor the directory's structure. Place into an existing folder if one fits. - Classify by pipeline stage / type, then drop it in the matching folder (example scheme below).
- Name it descriptively:
STAGE_<slug>_<who-or-model>_<YYYY-MM-DD>.extwhere it helps grep (e.g.RESULT_quant-test_<model>_2026-06-28.md). Keep the stage prefix — it's critical for routing. - If no folder fits, create a new typed sub-folder (don't invent a one-off at root) and update the map (
AGENTS.md+INDEX.md) so the new category is discoverable. Updating the map is part of creating the category, not optional. - Heavy/cruft/superseded (multi-MB logs, big zips,
*.bak, raw machine dumps) → off the live tree to an archive location outside it, never left at a root.
Example routing scheme (stage-based — adapt to your pipeline)
This is one working scheme for a research-workspace tree; treat it as a template, not a mandate.
| New file | Folder |
RESULT_* / FIX_* (dispatch results) | research/results/ |
outgoing dispatch briefs (BRIEF_*, DISPATCH_*, per-model dispatches) | research/dispatches/ |
RECONCILED_* | research/reconciled/ |
FINAL_* (finalized reports) | research/finals/ |
AUDIT_* | research/audits/ |
VERIFY_* | research/verify/ |
transcript* | research/transcripts/ |
SWEEP_* (experiment sweeps) | research/sweeps/ |
specs/plans/notes .md/.txt (no stage prefix) | research/notes/ |
*.html / *.svg (visual reports, cards) | reports/ |
*.json data | data/ |
*.log run logs | logs/ |
STAY AT ROOT (live anchors — never move): AGENTS.md, INDEX.md, and any live ledger/digest files that other tooling addresses by fixed path | (root) |
| KEEP AS-IS: already-organized working dirs and repos | (don't reshuffle) |
A second tree used for human↔agent coordination can route by kind instead of stage — e.g. coordination/ for handoffs/instructions/requests, research-briefs/ for briefs, setup-docs/ for setup documentation — with the same rule that loose files at the root are forbidden and designated anchor files stay.
Exception classes — the ONLY files allowed loose at a root
The no-loose-files rule has exactly three exception classes, and each must stay checkable:
- Tool-required anchors — files a tool, contract, or ecosystem addresses at a fixed root path
(
pyproject.toml,package.json,LICENSE,README.md, VCS dotfiles). This class is the ECOSYSTEM's list, not any one repository's root listing. A repository that has its own fixed-path contract file DECLARES it, with a reason, inguard/org_anchors.txt— and each declaration is checked, so the extension point cannot quietly become a blanket allow-list. - Runnable entry points named in the README — a root script is legitimate only while the
README names it; discoverability is the licence, and a root entry point the README never
mentions is a stray whatever it does. "Named" means a whole token on a line that is not a
prohibition: a substring test licenses any file whose name is a suffix of a name the README
does use (
leg_contract.pyriding oncheck_leg_contract.py), and it reads "never commitscratch_dump.log" as an endorsement — which turns the README into an allow-list inverter. - Directories — their internal organization is governed by the rules above, not the root rule.
Everything else at a root is a stray. Arm: guard/org_lint.py goes red on a stray root file
(gate: guard/tests/test_org_lint.py); run it in CI so an exception class cannot widen silently.
TOOLING RULE (so the structure can deepen without breaking)
Any script that READS artifacts by pattern must use a recursive glob — glob.glob(f"{BASE}/**/FINAL_*.md", recursive=True), not {BASE}/FINAL_*.md. This is what lets files live in sub-folders (or deeper, later) without breaking readers. (Learned during a reorg: several reader scripts had to be patched exactly this way.) When you MOVE files, run a path-chain audit: grep tooling for hardcoded old paths; fix any specific-file ref; verify generated HTML asset links still resolve.
Propagation
Agents that can't auto-retrieve skills (e.g. small local models dispatched as sub-tasks) need the relevant routing rows PREPENDED into their task brief when the task produces files.
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.

