commit-archaeologist
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches skill components from the author's repository during installation as described in the README.
- [COMMAND_EXECUTION]: Executes local
gitcommands through a Python script. The implementation uses list-based arguments forsubprocess.runto prevent shell injection and includes path validation to ensure operations are confined to the repository usingos.path.commonpathto mitigate path traversal risks. - [INDIRECT_PROMPT_INJECTION]: Processes untrusted data from git commit messages. An attacker could embed malicious instructions in the history. The skill mitigates this through explicit evidence rules and interpretation guidelines that instruct the agent to maintain low confidence in inferred intent.
- Ingestion points: Commit subjects and bodies are read in
scripts/archaeologist.pyviagit logandgit showcommands. - Boundary markers: Guidance in
SKILL.mdandreferences/reading-git-history.mdhelps the agent process data safely by labeling conclusions as inferences. - Capability inventory: Limited to read-only local
gitcommand execution with no file-write or network capabilities. - Sanitization: Commit messages are returned in a structured JSON format without specific sanitization of the message text content.
Audit Metadata