eng-ghostwriter
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from repository source files and git history to generate social post drafts.
- Ingestion points: Repository files and git history are read in Step 1 of SKILL.md.
- Boundary markers: A "Privacy rule" is defined in the body of SKILL.md to restrict the agent's scope to the current repository, though explicit technical delimiters for processed data are not utilized.
- Capability inventory: The skill uses bash to execute git commands, create directories, and write markdown files locally within the project scope.
- Sanitization: No specific sanitization or filtering of repository content is performed before it is synthesized into drafts.
- [SAFE]: The skill explicitly includes instructions to protect user privacy, forbidding the agent from reading session transcripts, accessing sibling repositories, or performing network exfiltration and git push operations.
Audit Metadata