The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Step 4 instructions direct the agent to search for and install external skills using the find-skills tool if a domain is not covered by the current local environment.
[REMOTE_CODE_EXECUTION]: The instruction to install and subsequently invoke these newly discovered skills constitutes a download-then-execute pattern. Since the source and integrity of these skills are not restricted to trusted vendors, this allows for the execution of arbitrary remote code.
[DYNAMIC_EXECUTION]: The skill implements dynamic capability extension by fetching and executing external code at runtime based on the perceived domain requirements of the current task.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection.
Ingestion points: The skill reads brain/principles.md (and all linked files), explores the entire local codebase via subagents (Step 3), and ingests output from dynamically installed external skills.
Boundary markers: No boundary markers or instructions to ignore embedded commands are present when processing codebase content or principle files.
Capability inventory: The skill can spawn subagents (Task), write files to the filesystem (brain/plans/), and install/delete project-local skills.
Sanitization: No sanitization or validation logic is specified for the data ingested from the codebase or principle files before it influences the planning logic or the decision to install external skills.

plan