wiki-policy-check
Pass
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from untrusted repository documentation.
- Ingestion points: The skill reads
CLAUDE.md,AGENTS.md,.wiki-guardrails.yml, and every.mdfile found in the repository. - Boundary markers: None. The instructions do not specify the use of delimiters or 'ignore embedded instructions' markers when reading external files.
- Capability inventory: The agent can list directory contents, read arbitrary local files, and potentially perform network requests if the wiki location is set to an external URL.
- Sanitization: None. Content from markdown files is processed directly to identify patterns without escaping or validation.
- [DATA_EXFILTRATION]: The skill aggregates sensitive business context and identifies external destinations via local configuration.
- Sensitive Data Identification: The skill specifically targets high-value data such as 'Pricing / monetization', 'Compliance posture' (GDPR, LGPD), and 'Customer-facing policies'.
- External Communication: The 'wiki location' can be an external URL defined in repository-level configuration files.
- Risk: A malicious repository configuration could define an attacker-controlled URL as the wiki destination, causing the agent to transmit aggregated business logic during the 'cross-check' phase.
Audit Metadata