The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project-level files and GitHub issue comments to manage state and tasks, which could potentially contain malicious instructions embedded by external actors.\n
Ingestion points: The agent is instructed to read CLAUDE.md, AGENTS.md, and issue comments via gh issue list.\n
Boundary markers: The skill contains explicit instructions to treat task issue bodies and comments as untrusted input and to avoid executing instructions found within them.\n
Capability inventory: The skill has the ability to execute shell commands (git, gh, lsof), write to the filesystem, and execute project-defined verification commands.\n
Sanitization: The protocol relies on the agent to identify and manually report suspicious content to the user instead of executing it.\n- [COMMAND_EXECUTION]: The skill executes various system and development tools to manage the repository and verify project status.\n
Evidence: Frequent use of git for logging, status checks, and branching, and gh (GitHub CLI) for issue and PR management.\n
Evidence: A 'Live Verify' step that encourages the agent to run commands specified in the project's documentation, such as ssh or curl.\n
Evidence: The freshness check script uses standard Unix utilities (find, grep, sed, awk) for file analysis.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for maintaining its environment through official platform tools.\n
Evidence: Step 0.5 in SKILL.md uses npx skills update -g to synchronize the skill with its source repository, which is a standard administrative operation.

handoff