shipshape
AGENTS.md7 starsChanged 4 months ago0BSDView on GitHub
This repository contains the Shipshape skill package. These instructions are for agents editing this repository, not for projects using Shipshape. Agents using Shipshape should treat this entire repository as Captain/human-owned material, like assets/** in an adopting project: edit it only when explicitly asked, and do not treat it as disposable implementation or verification. Shipshape is distributed as skill files:
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
17 rules. About 0 are common, 10 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 36
Keep repository guidance separate from downstream Shipshape usage guidance.
- Distinctive: line 37
Put public positioning in README.md.
- Very specific: line 39
Put shared workflow rules in skills/shipshape/SKILL.md.
- Distinctive: line 40
Put role-specific rules in the matching role skill.
- Very specific: line 41
Keep all role skills consistent with the Articles of Agreement in skills/shipshape/SKILL.md.
- Very specific: line 48
Use Shipshape Controlled English in skill files: short sentences, precise subjects, RFC 2119 terms where useful, and Canadian spelling such …
- Very specific: line 50
Do not add project-specific assumptions from Jolly, Saleor, or other adopters to shared Shipshape rules.
- Distinctive: line 51
Do not create binding repository artifacts such as roadmaps, memory banks, decision logs, constitutions, or task lists unless the user …
- Distinctive: line 52
Do not update installed global skills or downstream project copies unless the user explicitly asks.
- Distinctive: line 53
Do not commit or push unless the user explicitly asks.
- Distinctive: line 67
Keep every plugin artifact vendor-neutral.
- Very specific: line 69
Do not add a vendor manifest at a .<vendor>-plugin/plugin.json path.
- Very specific: line 74
Do not add agent-specific authorship, committer identity, or Co-authored-by trailers for coding agents.
- Distinctive: line 78
Before reporting a documentation change as complete:
- Distinctive: line 80
run diagnostics for changed Markdown files when available.
- Distinctive: line 81
search for stale names or project-specific leakage when relevant.
- Distinctive: line 82
check git diff for accidental duplication or unrelated edits.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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
Agent Instructions for Shipshape
This repository contains the Shipshape skill package. These instructions are for agents editing this repository, not for projects using Shipshape.
Agents using Shipshape should treat this entire repository as Captain/human-owned material, like assets/** in an adopting project: edit it only when explicitly asked, and do not treat it as disposable implementation or verification.
Scope
Shipshape is distributed as skill files:
| Path | Purpose |
skills/shipshape/SKILL.md | Shared workflow rules and project setup templates |
skills/captain/SKILL.md | Captain role skill |
skills/qm/SKILL.md | Quartermaster role skill |
skills/crew/SKILL.md | Crew Mate role skill |
skills/boatswain/SKILL.md | Boatswain role skill |
skills/shipwright/SKILL.md | Shipwright role skill |
skills/shipwright/templates.md | Fitting-out scaffold templates: RIGGING.md, AGENTS.md, README block |
.plugin/plugin.json | Vendor-neutral plugin manifest |
agents/ | Role agents for context isolation (thin adapters over role skills) |
hooks/ | Custody enforcement hooks (mechanize Article text only) |
commands/ | Plugin slash commands (derived-state reporting and install diagnostics, no doctrine) |
rules/ | Open-plugin rule files (.mdc): cited role checklists restating skill text, no added doctrine |
CAPTAIN.md | Captain-only non-binding notes for this repository's own Shipshape work |
assets/logo.svg | Plugin logo |
tests/ | Hook behaviour, map drift, style, doctrine-home, and bulkhead conformance tests |
README.md | Public project positioning and usage overview |
llms.txt | Agent-facing index of canonical files (pointers only, no doctrine) |
shipshape.md | Structural orientation map (names, relations, pointers; non-normative) |
skills.sh.json | skills.sh registry metadata |
AGENTS.md | Repository-local agent instructions |
Editing rules
- Keep repository guidance separate from downstream Shipshape usage guidance.
- Put public positioning in
README.md. - The README slogan is "Specifications are durable. Code is disposable. Agents are replaceable." It is a slogan, not doctrine. Keep it punchy. Do not rewrite it to match skill wording; the skills carry the precise three-layer doctrine.
- Put shared workflow rules in
skills/shipshape/SKILL.md. - Put role-specific rules in the matching role skill.
- Keep all role skills consistent with the Articles of Agreement in
skills/shipshape/SKILL.md. - Role skills are structured as jobs. A role's work divides into jobs. A job is entered by dispatch or by completion of the preceding job, and discovers work through exactly one retrieval: one question asked of one source, yielding one worklist. A job polices the quality of its own retrieval and of its own writes. Proof runs are execution, not retrieval. Incidental findings surfaced by a job's own work route by the Blocker policy; a duty that hunts through a second retrieval is a defect.
- A role duty MUST belong to a job, name its trigger in the role's walkthrough, read only evidence its job's retrieval or execution puts in front of it, and load no more context than its decision needs. Mechanize, move, or delete a duty that fails any of these.
shipshape.mdis a structural map: names, relations, and pointers only. A normative sentence in it is a defect. Update it with any rename it references;tests/map.shchecks the names against the skills.- Bump the
versionin.plugin/plugin.jsonwith any doctrine change, skill content included, not only a plugin-layer change to.plugin/,agents/,hooks/,commands/,assets/, orshipshape.md. A plugin-channel install's update tooling was observed gating on this version field rather than skill content hash; an unbumped skills-only change risks silent staleness for every plugin-channel consumer. - Enforcement claims are per-runtime. Claim a mechanism for a runtime only after live-fire verification on that runtime; name unverified runtimes unsupported for enforcement.
- Skills alone MUST fully instruct agents. The plugin layer (
.plugin/,agents/,hooks/,rules/) only mechanizes or restates what skill text already states. Every plugin artifact cites the skill text it enforces. Adding behaviour to the plugin layer that the skills do not state is a violation; fix the skill first, then mechanize it. Deleting the plugin layer must lose nothing but enforcement. - Use Shipshape Controlled English in skill files: short sentences, precise subjects, RFC 2119 terms where useful, and Canadian spelling such as
behaviour. Useartifact, notartefact. - The README Ship of Theseus section is the only text with relaxed Controlled English rules, for literary effect. Every other file follows Controlled English in full, including the punctuation rules: no em dashes and no parenthetical asides.
- Do not add project-specific assumptions from Jolly, Saleor, or other adopters to shared Shipshape rules.
- Do not create binding repository artifacts such as roadmaps, memory banks, decision logs, constitutions, or task lists unless the user explicitly asks.
- Do not update installed global skills or downstream project copies unless the user explicitly asks.
- Do not commit or push unless the user explicitly asks.
Plugin format
The repository is vendor-neutral. It carries no runtime-specific support. Downstream runtimes and installers add their own vendor layers; Shipshape ships only the generic minimum.
The core is skill-only, which is the widely portable form. Install it with npx skills add, which works across most agent runtimes. The optional plugin layer follows the vendor-neutral open-plugin specification. That format is experimental and its adoption is uncertain, so treat it as early-adopter support and keep it generic.
- Docs: https://open-plugins.com/
- Spec: https://github.com/vercel-labs/open-plugin-spec
- Installer: https://www.npmjs.com/package/plugins, run as
npx plugins add dmytri/shipshape
Rules that follow:
- Keep every plugin artifact vendor-neutral. Use no runtime-specific variable, path, directory, or manifest anywhere in the repository.
- The manifest lives at
.plugin/plugin.json, the spec-required path and the installer's source of truth. Component directories such asskills/,agents/,hooks/, andcommands/live at the plugin root, never inside.plugin/. - Do not add a vendor manifest at a
.<vendor>-plugin/plugin.jsonpath. On install,npx plugins addcopies.plugin/to each vendor directory, generates the vendor marketplace file, and registers the plugin in the runtime settings. That translation is downstream, not in this repository. hooks/hooks.jsonuses the neutral${PLUGIN_ROOT}. The installer rewrites it to each runtime's own plugin-root variable at install time. Do not hardcode a runtime-specific variable.
Commit attribution
Do not add agent-specific authorship, committer identity, or Co-authored-by trailers for coding agents. Commits should use Dmytri Kleiner <dev@dmytri.to> unless the maintainer explicitly instructs otherwise.
Validation
Before reporting a documentation change as complete:
- run diagnostics for changed Markdown files when available;
- search for stale names or project-specific leakage when relevant;
- check
git difffor accidental duplication or unrelated edits.
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.

