The Agent Skills Directory

[SAFE]: The skill performs its primary logic using a local Python script (score_stack.py) which is restricted to the Python standard library and does not initiate any network connections.
[COMMAND_EXECUTION]: The skill executes a local script via the Bash tool. This operation is secured through a frontmatter policy that restricts the Bash tool to only execute the specific script located at ${CLAUDE_SKILL_DIR}/scripts/score_stack.py.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data from gtm/offer-stack.json. This attack surface is mitigated by strict validation logic within the Python script.
Ingestion points: The skill reads structured data from gtm/offer-stack.json using the Read tool.
Boundary markers: Data is strictly treated as a JSON object, preventing the interpretation of plain-text instructions as commands.
Capability inventory: The skill has access to file reading/writing and a restricted Bash environment for running its internal scoring logic.
Sanitization: The score_stack.py script performs manual type checking, JSON parsing, and specific phrase-filtering (e.g., banning filler words like 'best-in-class') to ensure data integrity.

offer-stack