The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external commands via a Python wrapper script (scripts/meta_ads_agent.py) that interacts with the Meta Ads CLI. The script uses subprocess.run with arguments passed as a list and shlex.split, which follows security best practices to prevent shell injection attacks.
[DATA_EXFILTRATION]: The wrapper script (scripts/meta_ads_agent.py) contains specific logic (redact_token, redact_command, and scrub_env) designed to detect and redact sensitive information like access tokens, app secrets, and cookies from execution logs and command outputs. This serves as a significant security control to prevent credential exposure.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks common to agents that process external tool outputs.
Ingestion points: Untrusted data from the Meta Ads CLI is ingested and parsed by the Python wrapper in scripts/meta_ads_agent.py.
Boundary markers: The instructions advocate for structured JSON output (--output json) to minimize parsing errors, though they do not employ explicit security delimiters for raw CLI data.
Capability inventory: The skill possesses the capability to modify campaign states (create, update, delete) via the meta ads tool.
Sanitization: The skill mitigates risks by requiring explicit human approval strings (--approved) for any state-changing command and through a multi-tier risk classification system.
[PROMPT_INJECTION]: A metadata discrepancy is present where the author is listed as 'OpenAI' in the SKILL.md frontmatter, which contradicts the provided developer context ('tristanmanchester'). This is likely a template artifact but constitutes metadata poisoning.

meta-ads-cli