The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes its execution context by re-deriving facts from external tickets and pull requests, which are potentially attacker-controlled ingestion points. Maliciously crafted tickets could attempt to influence the agent's behavior during the implementation or review phases. • Ingestion points: The skill reads ticket/PR data as described in lifecycle Phase 0 and the /trackers lookup. • Boundary markers: It implements 'Active Rules' and 'Locked decisions' in the chat context, though these provide only partial protection against sophisticated injections. • Capability inventory: The skill can dispatch task workers for code writing, execute lint/test commands, and open pull requests via pr-ship.md. • Sanitization: No explicit content sanitization or instruction filtering for external tracker data is specified.
[COMMAND_EXECUTION]: The skill requires the execution of repository-specific scripts, including 'lint' and 'test' (or 'test:quality'), as part of its pre-commit gate in the reference.md ship questions section. These commands are executed locally based on the repository's configuration.

task