The Agent Skills Directory

[COMMAND_EXECUTION]: The skill runs shell commands using python and python -m to invoke audit and collection logic. These commands take parameters like URLs and keywords from user input, which could be exploited for command injection if not correctly handled.\n- [EXTERNAL_DOWNLOADS]: The skill references the github.com/cgallic/kai-cmo-harness repository for its core harness and scripts. It also interacts with multiple third-party marketing services and APIs to retrieve data during the audit process.\n- [DYNAMIC_EXECUTION]: The skill dynamically resolves script paths within its installation environment and executes them as modules at runtime.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from websites and vendor APIs.\n
Ingestion points: Web pages and API data stored in audit-data.json via SKILL.md.\n
Boundary markers: Uses specific metadata fields for provenance but lacks explicit instructions to treat external text as data only.\n
Capability inventory: File writing in the workspace/ folder and execution of shell commands.\n
Sanitization: No explicit data sanitization is mentioned; it depends on the supporting Python scripts and the agent platform.

kai-audit