The Agent Skills Directory

[SAFE]: A detailed audit of all skill files, including instructions and reference materials, indicates that the logic is strictly devoted to its documented purpose of strategy analysis. No signs of prompt injection targeting the platform, credential theft, or unauthorized data access were found.\n- [COMMAND_EXECUTION]: The skill instructs the agent to perform standard file system operations, such as reading input documents and writing reports or state files (e.g., in the .beagle/ directory). These actions are transparently documented as part of the 'Durable review state' workflow and do not involve arbitrary shell commands or privilege escalation.\n- [PROMPT_INJECTION]: The skill's primary function is to analyze untrusted user-provided content, establishing an indirect prompt injection surface. Ingestion points: Strategy drafts, notes, and briefs ingested in Step 1. Boundary markers: Absent; the instructions do not specify the use of delimiters or 'ignore-instructions' markers for external content. Capability inventory: File reading and writing capabilities (SKILL.md Step 5 and Durable review state). Sanitization: No input validation or sanitization methods are described. The potential impact of this surface is minimal because the skill is not granted high-privilege capabilities like network communication or executable code generation.

strategy-review