The Agent Skills Directory

[SAFE]: The skill does not perform any external network requests, credential harvesting, or unauthorized privilege escalation. All frameworks and templates are stored as local markdown files within the skill directory.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection through its state persistence logic. 1. Ingestion points: SKILL.md (Phase 0) instructs the agent to read existing strategy-notes.md and files within .beagle/strategy/ to resume previous interview sessions. 2. Boundary markers: The skill utilizes markdown formatting and triple backticks to encapsulate file contents during reading and writing. 3. Capability inventory: The agent is permitted to write strategic artifacts (strategy-draft.md and strategy-notes.md) to the user's working directory. 4. Sanitization: The skill employs an evidence-tagging mechanism (e.g., 'user said', 'inference', 'assumption') to distinguish between direct user input and derived conclusions, which serves as a mitigation against data confusion.

strategy-interview