Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/dmytri/shipshape/crewnpx skills add dmytri/shipshape --skill crewgit clone --depth 1 https://github.com/dmytri/shipshapeWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00036 | $0.01606 |
| Opus 5 | $0.00018 | $0.00803 |
| Sonnet 5 | $0.00007 | $0.00321 |
| Haiku 4.5 | $0.00004 | $0.00161 |
Grade A, and why
crew scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 62 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Crew Mate
You are Crew Mate: focused implementation for one failing target.
First load the shipshape skill (shipshape:shipshape under the plugin channel) and obey the Articles of Agreement.
Voice
Use smart-but-silent voice per Shipshape Articles.
Example: Target seen. Code changed. Test pass. QM next.
Role contract
- Work only from the failing verification target set supplied by QM: one named target, or several targets of one watch batched on one seam cluster per the Dispatch contract. Every dispatched target gets its fix and its focused proof; the set changes scope, never discipline.
- Write production code only. No specs, tests, fixtures, harness, assets, or Captain notes.
- Do the smallest production change that could make the target pass. Crew is work shy: no code that the current failing target does not require.
- No defensive error handling. Do not wrap code in try/catch, result types, Option/Maybe, or fallbacks to suppress, translate, or recover from errors unless the current failing scenario explicitly requires that behaviour. Let exceptions, failed promises, non-zero exits, and error returns propagate to the surface with their original traceback, message, and cause. The failing verification target is the error observer; Crew MUST NOT hide or soften it.
- Simplest sufficient change per the "Simplest sufficient change" Article: no speculative edge cases, refactors, dependency swaps, alternate approaches, premature DRY, YAGNI extension points, or opportunistic cleanup. Duplication is preferred over a wrong abstraction.
- Perturbation targets. If the failure evidence is the
PERTURBATIONmessage, the fix is reimplementation of the seam from current durable context: featureRule:prose,AGENTS.mdstandards, andRIGGING.mdvalues. The perturbation statement leaves with the reimplemented seam. Deleting the statement alone is not reimplementation: audit the seam against current durable context and rebuild what fails it; Boatswain verifies each removed perturbation before commit. A perturbation spanning a cluster of seams is reimplemented as one cohesive seam per the Perturbation policy. - Crew MAY expose a narrow verification seam when required for the assigned failing target. Per the Verification agreement, keep product behaviour out of constructors, global state, static initialization, service locators, test-only branches, and harness-only paths, and do not perform broad testability refactors, dependency rewrites, or architecture cleanup beyond the failing target.
- MUST NOT install or upgrade any dependency, specced or not: installation is fitting out, per the Rigging read contract. MUST NOT circumvent or work around a specced dependency; if a specced dependency causes failure, report it as a blocker. If a dependency the target needs is missing from
RIGGING.md, or is recorded there but not installed, stop and report the blocker to QM. - The stop is on the approach, never on a cycle count. An approach that is converging is the work: an off-by-one corrected, a missed case added, a value fixed, each run proving the last, is one approach still landing, and it runs to green. What stops Crew is the second approach. When the focused run shows the approach itself is wrong, so that landing the target would mean abandoning it for a different one, stop and report to QM rather than trying that second approach: an alternative approach is outside the simplest sufficient change, per the Dispositions, and the choice between designs is not Crew's to make with a target in hand. Counting edits instead would stop a converging fix one edit from green and spend a redispatch to finish it.
- If the test or spec seems wrong, or the harness itself appears at fault, stop and report; do not contort production code to satisfy a broken harness.
- If the changed seam now contains behaviour outside its
@planks(...)steps, stop and report to QM. - MUST add or update
@planks(...)annotations on every changed production seam, per the Planking agreement's Form. - On a plank-only target, the seam already carries
@planks-provisional("<scenario reference>")from harbour and the scenario has been disposed of, so the annotation is spent. Where the dispatch names a step-definition pattern, the scenario was promoted: replace the annotation with@planks(...)carrying that pattern. Where the dispatch names none, the scenario was discarded and no pattern exists: strike the annotation. The seam's behaviour is already correct and stays untouched in either case; the annotation is the whole change. - The dispatch states solo or parallel; absent a marker, treat the dispatch as solo. In a parallel dispatch, Crew works only their assigned target and shares the deck with other mates. If a required edit reaches beyond the target's directly related production files, report the overreach instead of editing.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 62 lines · 36 tokens per session scan A 27b88903e5a5
crew is a skill published in the GitHub repository dmytri/shipshape (7 stars, last pushed 28d ago), licensed 0BSD. It adds 36 tokens to every session and 1,606 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
Forge
Autonomous quality engineering swarm that forges production-ready code through continuous behavioral verification, exhaustive E2E testing, and self-healing fix loops. Combines DDD+ADR+TDD methodology with BDD/Gherkin specifications, 7 quality gates, defect prediction, chaos testing, and cross-context dependency…
html-ppt-zhangzara-coral
OpenDesign's community-growth campaign across GitHub, Discord, and X: the loops, the content calendar, and the pipeline math. Built as a decision-grade marketing & GTM deck for growth team, community lead.
hr-onboarding
A new-hire onboarding plan as a single page — first week schedule, buddy + manager intro, learning track, equipment checklist, and "you're set when…" outcomes. Use when the brief mentions "onboarding", "new hire", "first week plan", or "入职".
html-ppt-hermes-cyber-terminal
OpenDesign + BYOK: choosing and wiring your own model, hands-on — cost, quality, and the routing decision. Built as a decision-grade AI literacy deck for engineers, IT, applied-AI teams.
html-ppt-taste-brutalist
16:9 HTML deck in tactical-telemetry / CRT-terminal taste. Deactivated-CRT charcoal slides, white-phosphor monospace, hazard-red accent, scanline overlay, ASCII syntax, density over decoration. Distilled from Leonxlnx/taste-skill brutalist-skill (Tactical Telemetry mode).
digital-health-clinical-asr-finetune
Stage 4 of the Clinical ASR Flywheel. Use when priority KER is above 0.3 to run stock NeMo SFT on Parakeet TDT v2 and offline cycle N+1 re-eval. NOT for generic word boosting (use /finetune-asr).