The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on an 'Interview' process to gather project details, which constitutes an ingestion point for untrusted user data. Maliciously crafted input could influence the agent's strategic advice or subsequent actions.\n
Ingestion points: User responses to the 11-question interview in the 'Interview' section of SKILL.md.\n
Boundary markers: There are no explicit instructions or delimiters provided to the agent to differentiate between its core instructions and the potentially adversarial user content.\n
Capability inventory: The skill explicitly directs the agent to execute code derived from the user's repository (Step 2).\n
Sanitization: No input validation or sanitization is defined for the user-provided data.\n- [COMMAND_EXECUTION]: In Step 2 ('Run the readiness gate'), the skill instructs the agent to perform a 'cold run' by executing installation commands directly from a repository's README. While this is an intended functional capability to verify repository quality, it creates a significant risk of remote code execution if the repository being analyzed contains malicious installation scripts. Evidence: SKILL.md, Step 2 ('If you can execute commands, do the cold run yourself in a fresh directory').

oss-launch