The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses directive language in the 'Skill Isolation' section to override the agent's control flow. Specifically, it commands the agent to be the 'sole active authority' and to 'suppress' any other skills that might activate. This technique is used to bypass platform orchestration and ignore potentially safety-critical instructions.
[DATA_EXFILTRATION]: The instructions direct the agent to perform 'read-only exploration' of the existing codebase and compile the results into a technical_requirements.md file. This creates a data exposure risk where sensitive information, such as configuration details or hardcoded secrets encountered during exploration, could be moved from protected files into a more accessible requirements document.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it reads potentially untrusted content from the codebase and processes user input without explicitly defined boundary markers or sanitization steps.
Ingestion points: User input ('Interrogate me') and codebase exploration.
Boundary markers: Absent.
Capability inventory: File read (codebase exploration) and file write (technical_requirements.md).
Sanitization: Absent.

break-req