The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow involves executing a local Python script: python3 scripts/validate_engagement_artifacts.py <client_dir>. This is used to validate engagement workpapers.
[DATA_EXFILTRATION]: The skill accesses the file system to search for engagement_state.json files within the current directory and its subdirectories (cwd/clients/*/). It also accepts and processes user-provided file paths.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by processing data from engagement_state.json, which determines the agent's next steps and status.
Ingestion points: engagement_state.json file content, specifically fields like legal_name, operator, engagement_type, and current_stage.
Boundary markers: None identified; data from the state file is integrated directly into the status board and runtime logic.
Capability inventory: Subprocess execution (python3), file read/write operations, and dynamic loading of subsequent skill files.
Sanitization: There is no explicit sanitization or validation of the JSON state content before it is used to identify the next SKILL.md to load.

resume-engagement