caveman-commit
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection due to how it handles external data and persistent file storage.
- Ingestion points: The skill ingest code diffs and staged changes (file:
SKILL.md) to generate content for commit messages and subsequent reports. - Boundary markers: The instructions do not specify any delimiters or instructions to treat data within the diffs as untrusted, which could allow an attacker to inject instructions via a code change.
- Capability inventory: The skill is configured to perform file-write operations, specifically creating new files in
llmwiki/wiki/draft/cave/and appending data tollmwiki/wiki/index.mdandllmwiki/wiki/log.md. - Sanitization: There are no sanitization or escaping protocols defined for the content extracted from the diffs before it is written to the wiki files.
- [COMMAND_EXECUTION]: The 'Output Report' section requires the agent to perform file system operations, including file creation and modification (appending rows to index files), to maintain the wiki state.
Audit Metadata