The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script node scripts/shipyard-audit.mjs [repoRoot] to perform repository health checks and 'drydock' audits. This is a standard project-specific verification step and does not involve remote code execution or unknown binaries.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from 'mission briefs' or 'existing specs', which creates a potential surface for indirect prompt injection.
Ingestion points: External inputs are received at the entry phase (Phase 0) via command arguments and the 'fog gate' evaluation.
Boundary markers: The skill mitigates risks through a 'Comprehension reset' protocol (Phase 1) where decisions are restated to the user for approval, and five distinct human checkpoints (C1-C5) that gate critical transitions.
Capability inventory: The skill has the capability to write to project files (e.g., CONTEXT.md, docs/adr/, .omc/specs/), execute local Node.js scripts, and call other system skills using the Skill tool.
Sanitization: The workflow includes 'Agent-side mechanical validation' of decomposed tickets to ensure they meet technical standards (independence and context fit) before human review.

launch