multica-ops
CLAUDE.md11 starsChanged 49 days agoApache-2.0View on GitHub
Development of the skill, not use of it. Reading this from inside another project means the routing went wrong: a workspace built with the skill carries its own generated guide, and that guide governs there — this file governs only work on this source tree. Runtimes load a plugin's skills, never its root guide, so the two cannot meet by accident. The full contract is AGENTS.md — read it before changing anything of consequence, and GLOSSARY.md plus PATTERNS.md before writing…
The file
We split this CLAUDE.md into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 27
Not optional for a minor or major.
- Distinctive: line 28
Showcase: every new mechanic where a reader will meet it: a diagram where the shape is the explanation (we draw in mermaid), a situation in …
- Distinctive: line 69
Memory: update the project memory file with what shipped and what is owed.
- Distinctive: line 80
Evidence moves without a tag; a rule moves with one: Run records, eval verdicts and dated measurements are plain commits.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md 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
This is the skill's own repository
Development of the skill, not use of it. Reading this from inside another project means the routing went wrong: a workspace built with the skill carries its own generated guide, and that guide governs there — this file governs only work on this source tree. Runtimes load a plugin's skills, never its root guide, so the two cannot meet by accident.
The full contract is AGENTS.md — read it before changing anything of consequence, and GLOSSARY.md plus PATTERNS.md before writing prose others will read. What follows is the loop this file exists to stop anyone re-deriving once a session.
This repository is not a company Mops operates. It ships the skill; nothing here is a
workspace under migration, so there is no UPGRADES.md and none is owed.
The session loop
- Change — one home per rule, cited from everywhere else, never restated. When a rule turns out not to hold, the repair is a form, not a stronger sentence — measured here more than once. A capability ships with four things or it ships incomplete: a form where the rule can fail, a mutation test asserting what that form refuses, a scenario with the fixture it needs, and a door — AGENTS.md → What a capability owes before it ships, which carries the measurement behind each one. Missing one is a sentence in the commit message saying which and why, never a silence.
- Lenses on anything of consequence — deletion · adversarial · contradiction · cold-read — by a reader who did not write it, each reporting even when it found nothing. Not optional for a minor or major; a patch may skip them.
- Showcase every new mechanic where a reader will meet it: a diagram where the shape
is the explanation (we draw in mermaid), a situation in USE-CASES.md,
and a fact in sources/SOURCES.md when the claim is about the
outside world. A wording-only change owes none of these — say so rather than inventing one.
The entry's
**Trio:**line is measured, not trusted:scripts/check-trio.pycounts what the range since the last tag added — mermaid blocks, USE-CASES rows, register entries — and preflight §5j-bis fails a line that disagrees or that it cannot read. 0.4.18's line said situations 5 over a range that added 7, and nothing compared the two until the owner asked (2026-09-25). - Checks, exit codes captured first —
cmd > /tmp/out 2>&1; rc=$?— then read the tail:bash scripts/preflight.sh·python3 scripts/verify.py --live·python3 scripts/fetch-source.py --verifythen--verify-citations·python3 scripts/tests/test_issues_helpers.py· and every shipped guard's mutation suite —test-migration-hook.sh·test-migrate-layout.sh·test-outward-gate.sh·test-rule-home.sh·test-map-blocks.sh·test-dispatch-nudge.sh·test-preflight-checks.sh·test-native-register.sh·test-recheck-trigger.sh·test-link-names.sh·test-link-ids.sh·test-deps-report.sh·test-check-releases.sh·test-check-shelf-links.sh·test-check-trio.sh·test-verify-reads.sh. Each one's load-bearing assertion is about what its holder refuses, so a suite left out of this list is a guard nobody re-checks. Green is evidence about the corpus, never about behaviour — behaviour is the eval suite's job, and a minor or major is not tagged withoutevals/runs/<version>.md. Thenpython3 scripts/coverage-map.pyand readevals/COVERAGE.md: a holder with no test and a version with no run record both show up there first. - Changelog entry (capability first — it is the migration map
/multica-ops:upgradereads) → bump every manifest (the sweep runs inside preflight, so a straggler fails rather than ships) → README roadmap pruned of what landed. - Tag — only on the owner's word (see Versioning), then the GitHub Release whose notes are the entry whole with its heading collapsed to a bare italic date.
- Site:
cd ~/Dev/ai && python3 scripts/generate.py ~/Dev/multica-ops— the generator takes the source repo as its one argument and dies onIndexErrorwithout it — then commit and push that repo too. Then watch the deploy land: it deploys continuously, so a skipped regen silently ships the previous pages against the new tag — and continuously is a habit of the integration, not a guarantee. Measured 2026-08-08: the 0.4.1 regen (e84cb2c) reached GitHub at 04:20 and Vercel never built it; production stayed on the 03:32 build until 0.4.2 pushed past it, so the site served 0.4.0 for two releases while both repos looked clean. Confirm a deployment exists for that SHA and reachedREADY, not that the push succeeded. A new page-worthy file needs a route in the generator first. - Installs on this machine:
bash scripts/find-installs.sh, follow each row's route, run it again, and read theRESULT:line — that line is the canon, not a remembered count. Machines differ, installs come and go, and a runtime the script does not know yet gets added to it when met. - Memory: update the project memory file with what shipped and what is owed.
Versioning
The tag waits for the owner — every release, its own word. Everything before it — the
entry, the bump, the checks — is preparation and may land in main; the tag, the GitHub
Release, the site push and the machine re-sync happen only after an explicit yes, and an
earlier yes does not roll forward to the next version. This is the skill's own law about
outward acts — deploy and announce are owner-confirmed every time — applied to the one
repository where it is easiest to forget.
Evidence moves without a tag; a rule moves with one. Run records, eval verdicts and dated measurements are plain commits. Anything that changes behaviour or format is a release, however small, so nothing accumulates outside versions.
Batch, don't drip — pool small fixes into one patch (AGENTS.md → When to cut a version). The exception is an urgent fix: a real block, a security or data issue ships alone.
Machine notes
- A pipe eats the exit code.
preflight.sh | tailgates nothing — capture the code first, then read the output. Measured next door: a red preflight rode a green pipeline intomain. - CI runs on macOS on purpose — the same tools the release ritual runs on. A badge green on a platform nobody releases from is decoration.
- GitHub issues land as triage: read, classify, fix or decline with a reason, and close with the reasoning in a comment.
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.

