The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The README mentions installation via npx skills add catena-labs/dev-skills. This is a reference to a skill repository owned by the vendor and is the standard installation method for this platform.\n- [PROMPT_INJECTION]: The skill instructions proactively mitigate risks from potentially malicious or outdated project files by instructing the agent to treat existing AGENTS.md and CLAUDE.md content as 'hypotheses' rather than 'ground truth'. It requires the agent to verify stack claims against authoritative sources like package.json, lockfiles, and the project source code.\n- [COMMAND_EXECUTION]: The skill guides the agent to perform standard file system operations, such as creating a symbolic link (ln -s) to unify instruction files. This is a common and legitimate practice for maintaining a single source of truth in software projects.\n- [DATA_EXFILTRATION]: While the skill reads project metadata (dependencies, scripts, and monorepo structure) to generate documentation, there are no network operations that would transmit this data to external or untrusted domains.

optimize-agents-md