The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates new executable skill files under the .agents/skills/ directory by analyzing the repository's startup commands and surfaces. This process uses repository data to construct new instruction sets, which is a meta-programming pattern that can be exploited if the source data is malicious.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository's documentation and commands during the 'Interview' phase to populate the generated recipe. 1. Ingestion points: Repository content analyzed in SKILL.md. 2. Boundary markers: None. 3. Capability inventory: The generated recipes can execute shell commands and drive user surfaces. 4. Sanitization: None documented.
[COMMAND_EXECUTION]: The skill includes a step to 'Demonstrate once' by executing commands discovered within the repository. Executing unverified commands found in project documentation poses a risk if the repository has been compromised to include malicious instructions.

project-verification