The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files (CLAUDE.md, README.md, CONTEXT.md) to determine the project's document language and configuration.
Ingestion points: The skill reads the documentLanguage key from CONTEXT.md and performs content inspection on CLAUDE.md and README.md to infer language settings.
Boundary markers: No explicit boundary markers or "ignore instructions" delimiters are specified when reading the content of these repository files.
Capability inventory: The skill has extensive file system write access for scaffolding and can execute shell commands via the audit script and git hooks.
Sanitization: Language tags are normalized to BCP-47 style, but the broader content used for inference lacks specific sanitization.
[COMMAND_EXECUTION]: The skill references the execution of a Node.js script and the configuration of git hooks that trigger shell commands.
Evidence: The --check mode documentation specifies the execution of node scripts/shipyard-audit.mjs to perform repository audits.
[DYNAMIC_EXECUTION]: The skill generates hook presets and configuration files that dynamically trigger shell commands.
Evidence: The "Destructive-operation guardrail preset" seeds a PreToolUse hook with a command (e.g., <confirm-before-destructive-git>) that is executed by the environment.

drydock