The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses a sensitive local configuration file containing API credentials and exposes a specific customer account identifier.\n
Evidence: src/google.py defines yaml_path as ~/google-ads.yaml and uses GoogleAdsClient.load_from_storage to read it. This file is the standard location for sensitive Google Ads API credentials, including OAuth2 tokens.\n
Evidence: The skill hardcodes a specific Google Ads Login Customer ID (6445466801) as a default value in src/google.py and SKILL.md, which identifies the author's account.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through unvalidated CLI inputs that are interpolated into API queries.\n
Ingestion points: CLI options such as --campaign-id, --targeting, and --objective in cli.py accept external data that is passed directly to the unified client.\n
Boundary markers: There are no markers or instructions used to isolate untrusted data from functional API parameters or query logic.\n
Capability inventory: The skill has permissions to create, modify, and pause campaigns, as well as fetch financial and performance data across multiple platforms.\n
Sanitization: Minimal sanitization is applied (e.g., strip(), replace()), but there is no escaping or parameterization for the Google Ads Query Language (GAQL) used in src/google.py.\n
Evidence: In src/google.py, the campaign_id is directly interpolated into a GAQL query: f"SELECT ... WHERE campaign.id = {campaign_id_str}". This allows for query manipulation where an attacker could potentially inject GAQL clauses to access or modify unauthorized campaign information.

ads-cli