project-graveyard
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local git repositories, including commit messages and file paths, which are then interpolated into the AI agent's prompt context as "forensic evidence."
- Ingestion points: The
scripts/graveyard.pyscript reads commit messages and file names usinggit logandgit ls-files. - Boundary markers: The skill instructions (
SKILL.md) and the script output do not use specific delimiters or protective instructions to distinguish untrusted git metadata from authoritative skill content. - Capability inventory: AI agents executing this skill typically possess capabilities for filesystem access, shell execution, and network communication, which could be abused if an injection is successful.
- Sanitization: No sanitization, escaping, or filtering is applied to the ingested git metadata before it is presented to the LLM.
- [COMMAND_EXECUTION]: The skill executes local commands on the host system to gather repository metadata.
- Evidence: The
scripts/graveyard.pyscript uses thesubprocess.runfunction to invoke thegitbinary for operations such aslog,ls-files,remote,tag, andconfig. While the arguments are passed as a list (reducing shell injection risk), this remains a direct interaction with the host operating system.
Audit Metadata