The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bun run to execute task management and verification scripts (e.g., bun run agent:task init). This is the primary intended function of the skill for local development workflows and follows standard TypeScript/Bun development practices.
[DATA_EXPOSURE]: The instructions explicitly mandate that task data and command output remain under .artifacts/ which is ignored by version control. It specifically warns: 'never attach them wholesale to a PR', demonstrating a proactive stance against accidental data exposure.
[EXTERNAL_DOWNLOADS]: The skill mentions fetching state from 'external issue, PR and CI', but these are typical developer workflow operations. The example provided uses a 'composiohq' GitHub URL (https://github.com/ComposioHQ/composio/issues/123), which is a trusted vendor resource for this skill.
[PRIVILEGE_ESCALATION]: The skill explicitly states that it 'does not grant commit, push, merge or release authority' and that the authority field is 'never permission to exceed the current user request', preventing the agent from assuming unauthorized powers.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (issue/PR URLs), which is a potential injection surface. However, it lacks high-risk capabilities like network exfiltration or shell execution of external content, and it focuses on local status updates and verification, placing it in the lowest risk tier for this category.

verify-change