The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions include steps to explore existing codebase files and configuration to inform the interview process, which constitutes an ingestion point for untrusted content if used on external or compromised codebases.
Ingestion points: The workflow explicitly reads existing implementation files, models, schemas, and configurations (SKILL.md).
Boundary markers: The instructions lack explicit boundary markers or directives to ignore instructions that might be embedded within the files being read.
Capability inventory: The skill has the capability to generate and update documentation like SPEC.md, product requirement documents (PRDs), and implementation plans.
Sanitization: There are no mentioned mechanisms for sanitizing or validating content retrieved from the codebase before it is used to generate new technical documents.
[SAFE]: The skill uses multi-platform tool mapping to interact with the user through official runtime capabilities (e.g., AskFollowupQuestion in Claude Code). This is a legitimate use of agent tools to facilitate user communication.
[SAFE]: All codebase exploration and file updates are local operations intended for architectural planning and development assistance, with no evidence of network exfiltration or unauthorized access to sensitive system directories.

interview-me