shipyard
Installation
SKILL.md
Shipyard
The main conversation owns the user's goal, priorities, integration, and final acceptance. Handle small tasks directly; introduce workers or a PM only when their independent work saves more time than the handoff costs.
If the text after the skill name (/shipyard, /itamaker-skills:shipyard in the Claude plugin, or $shipyard) begins with --, follow commands.md first and pass the literal arguments to the flat-parameter helper. Do not reinterpret malformed flags as natural language or interpolate them into shell code. With no parameters, continue the normal workflow; natural-language requests remain available.
Load only your path
- Role personality: for normal delivery, load the current role's working style once at entry using personas.md. The primary conversation uses
chief-of-staff. Resolve the assigned role again when starting a new invocation. Persona commands use the flat-parameter path above. - Configure models: for
/shipyardor$shipyardrequests to change/show/reset model settings, read configuration.md first. Map intent to task, one-shot, project, or personal scope; persist only an explicit save request. - Coordinator / PM: read coordination.md for dispatch, exclusive ownership, milestones, and recovery. Read verification.md when defining checks or accepting a revision.
- Ask-mode replacement: when the Chief of Staff proposes a staffing change in ask mode, use the timed notice and handoff procedure.
- Model / effort selection: before starting a subagent, follow models.md. Select by task and risk from the central policy, honor explicit overrides, and check the actual host's override support.
- Implementation worker: read implementation.md. Run only tests and checks related to the ticket's behavior and affected dependencies; full-suite execution belongs to the coordinator's CI/integration stage.
- Advisor: answer the bounded question using current evidence. Remain read-only, return actionable findings, and finish without spawning helpers. Read verification.md only when test/acceptance policy is the question.
- Merge / cleanup: read delivery.md before publishing, merging, or deleting a task branch.
- Team design / CI placement: read staffing.md only when choosing roles or changing the pipeline.
- Client setup: read clients.md when installing the bundled native role assets or mapping role names. Installation is an explicit setup step; loading the skill does not register roles or start a background scheduler.