The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and acting upon untrusted data from both user-provided project plans and mission reports.
Ingestion points: Tools such as plan_project and create_mission ingest natural language prompts. The get_report tool processes external mission summaries including changed files and error logs from the execution environment.
Boundary markers: The instructions lack explicit delimiters or instructions for the agent to disregard potential instructions embedded within the processed data or mission reports.
Capability inventory: The skill manages agents running in git worktrees with 'full tooling', implying capabilities for file system modification, command execution, and automated git merges.
Sanitization: There is no documentation regarding the sanitization or validation of mission outputs before they are processed by the orchestrator.

claude-devfleet