serp-keyword-harvester

Pass

Audited by Gen Agent Trust Hub on Jul 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill interacts with external search APIs (OpenSERP). It correctly classifies these operations as 'external-network' and mandates explicit user approval before any network operations are performed, mitigating unauthorized data transmission risks.
  • [COMMAND_EXECUTION]: The instructions include validation steps that execute local Python scripts using system-defined paths (e.g., python3 $CODEX_HOME/skills/...). These commands are intended for development-time integrity checks.
  • [PROMPT_INJECTION]: The skill ingests untrusted external data such as CSV exports and website text, creating a surface for indirect prompt injection attacks where malicious instructions could be embedded in the harvested keywords.
  • Ingestion points: SKILL.md (Processing CSVs, exported site text, and local reports)
  • Boundary markers: Absent
  • Capability inventory: Network requests (OpenSERP), file writes (reports/keyword-harvest.json), and subprocess execution (validation scripts in SKILL.md)
  • Sanitization: Not documented in the provided instructions
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 29, 2026, 07:31 AM
Security Audit — agent-trust-hub — serp-keyword-harvester