The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the dynamic context injection syntax (!) in SKILL.md to execute a local Python script (retrieve.py) for context bootstrapping at load time. It also contains several bash scripts (e.g., move.sh, build-queue.sh, heuristic.sh) and instructions for Git operations related to repository maintenance and methodology refactoring. These operations are scoped to the local project environment and are consistent with the skill's stated purpose of assisting a solopreneur developer.
[PROMPT_INJECTION]: Several methodology files, including agent-patterns, agent-replay-harness-cookbook, and agent-shape-decision-frame, explicitly define 'instruction-shaped content' in user inputs as a failure mode and instruct the agent to ignore such content to prevent behavior overrides. This is a strong defensive pattern against indirect prompt injection.
[DATA_EXFILTRATION]: No exfiltration patterns were identified. While the retrieval mechanisms access platform-specific directories like ~/.claude/projects/ to read session context, this behavior is localized to the tool's execution and no unauthorized external network transmissions were found.

faion