The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash to execute local Python scripts (python3 scripts/profile.py) for retrieving user profiles, saving audit results to a history log, and performing comparisons between audits. It also dynamically executes verification commands (referred to as MCP tools or API scripts) during the step-by-step walkthrough in Phase 6 based on the findings detected in the audit files.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes untrusted external data from JSON files that drive agent coaching and verification actions.
Ingestion points: Reads *-audit-results.json files from the current working directory and audit history from ~/.claude-ads/history/index.json (SKILL.md, Phase 1).
Boundary markers: There are no specified boundary markers or instructions to the agent to disregard potentially malicious instructions embedded within the JSON fields (such as recommendation or finding).
Capability inventory: The agent has the capability to execute shell commands via Bash for profile management and verification loops, and can perform file operations via Write (SKILL.md, Phases 2, 6, and 7).
Sanitization: While the skill validates the JSON structure and platform names, it does not perform sanitization on the descriptive text fields that are used to guide the user and trigger verification steps.

ads-next