The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to manage files and check for external configurations. Specifically, it uses cp to initialize templates and ls to check for the existence of overlay directories.
[DYNAMIC_EXECUTION]: The skill implements an "Overlay" mechanism that allows instructions to be loaded from an external path defined by the $POWERSUIT_OVERLAY environment variable. The skill explicitly directs the agent to "follow the overlay" if it disagrees with the primary instructions, which allows for arbitrary overriding of safety and operational rules via external files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources (interviewers) in the form of scorecards and evidence. There is a risk that an interviewer could include malicious instructions within their feedback to influence the AI's facilitation of the debrief or the final recorded decision.
Ingestion points: Scorecard collection in Step 1 (SKILL.md).
Boundary markers: None. The skill does not define delimiters or instruct the agent to ignore instructions embedded within the scorecards.
Capability inventory: Shell command execution (bash) and file system write access.
Sanitization: None. The skill asks the agent to verify that evidence matches rubric anchors, but does not sanitize the text for prompt injection patterns.
[DATA_EXPOSURE]: The skill handles highly sensitive HR data, including candidate IDs, interviewer evaluations, and hiring decisions. While the skill includes rules against storing data in public repositories, the execution of shell commands and the dynamic loading of external instructions could be leveraged to exfiltrate this sensitive information.

hiring-debrief