The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a bash script intended for execution by the agent to implement a 'freeze guard'. This script resolves a module directory to its physical path and writes it to a session-specific temporary file in /tmp/octopus-freeze-${CLAUDE_SESSION_ID:-$$}.txt. This is used to track and restrict file edits to user-approved areas.\n- [INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to read and apply logic from external markdown files (e.g., skills/blocks/engineering-method-selection.md and skills/blocks/debug-feedback-loop.md). Evidence chain: 1. Ingestion points: Local markdown files in the skill's plugin root and user-provided symptom descriptions. 2. Boundary markers: Absent. 3. Capability inventory: Shell command execution (bash) and file read/write operations. 4. Sanitization: Absent. While these files are part of the skill's structure, they represent an ingestion surface for instructions.\n- [DATA_EXPOSURE]: The skill accesses a tool-specific configuration file located at ~/.claude-octopus/loop-config.conf to retrieve default values for its self-regulation logic. This is a local configuration access for state management.

skill-debug