ppc-reporting
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill core functionality involves ingesting untrusted external data, such as campaign names, conversion labels, and CSV exports from ad platforms. Evidence found in SKILL.md and references/agent-security.md.
- Ingestion points: Data enters through pasted exports, account labels, and fetched website content.
- Boundary markers: The skill includes a dedicated references/agent-security.md file that explicitly instructs the agent to treat all retrieved content as data, never instructions.
- Capability inventory: The skill is authorized to read local files and write context data to .agents/product-context.md.
- Sanitization: references/agent-security.md mandates HTML-escaping for all interpolated values and requires validation of URLs to use HTTPS.
- [PROMPT_INJECTION]: Static analysis flagged instructions in SKILL.md regarding the suppression of drafts with warnings. This is evaluated as a benign safety protocol designed to prevent user reliance on incomplete data rather than an attempt to bypass safety constraints.
- [DATA_EXFILTRATION]: While the skill handles sensitive advertising data, it includes explicit safeguards in references/agent-security.md Section 2. These rules strictly prohibit the persistence or echoing of secrets like API keys and mandate their redaction if detected in input sources.
- [EXTERNAL_DOWNLOADS]: The skill is instructed to fetch content from user-provided URLs to build product context. Safety is managed via explicit instructions in references/agent-security.md to treat such content only as data.
Audit Metadata