The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes shell and Python scripts, such as orchestrate.sh and search.py, by directly interpolating user-provided task descriptions and prompts into command-line arguments. This pattern is vulnerable to command injection if the underlying scripts do not properly sanitize these inputs.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its synthesis of external and local data.\n
Ingestion points: Data is ingested from user implementation requests, project state files (.octo/STATE.md), and synthesis output from external providers (Codex, Gemini) stored in ~/.claude-octopus/results/.\n
Boundary markers: The workflow does not utilize explicit delimiters or 'ignore' instructions when passing user-provided content to the orchestrate.sh script or other sub-processes.\n
Capability inventory: The skill has broad capabilities, including executing shell commands, reading/writing files, and orchestrating subagents via SendMessage.\n
Sanitization: There is no evidence of sanitization or escaping of external content before it is used for command execution or prompt construction.

flow-develop