apex-takeover

Warn

Audited by Gen Agent Trust Hub on May 7, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill explicitly directs the agent to access and analyze sensitive files containing credentials.
  • Evidence: "Warden: Security scan — secrets in code, vulnerable dependencies, exposed endpoints. Check .env files, hardcoded credentials, dependency audit." (SKILL.md)
  • [DATA_EXFILTRATION]: The skill instructs the agent to aggregate all findings, which includes the secrets gathered in the reconnaissance phase, and transmit them to a specific tool or endpoint.
  • Evidence: "Synthesize all findings, then route through atlas-report: ... Invoke /atlas-report with the full synthesized findings." (SKILL.md)
  • [PROMPT_INJECTION]: The skill acts as an audit tool for inherited or acquired codebases, which constitutes a large surface for indirect prompt injection from untrusted files.
  • Ingestion points: Full codebase contents, project manifests, configuration files, and CI/CD scripts (SKILL.md).
  • Boundary markers: None specified for the data ingestion phases.
  • Capability inventory: Access to tools like Read, Write, Edit, Bash, WebFetch, and WebSearch (SKILL.md).
  • Sanitization: No sanitization or validation of the inherited codebase content is described prior to processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 7, 2026, 01:39 PM