commit-context
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The
TROUBLESHOOTING.mdfile instructs the agent to read sensitive credential files (~/.agentmemory/.envand~/.agentmemory/secret) to facilitate a REST API fallback when MCP tools are unavailable. While it includes a warning not to send local secrets to remote URLs, the practice of reading and using local credentials based on instructions in a skill constitutes a sensitive data access pattern. - [COMMAND_EXECUTION]: The skill relies on executing shell commands (
git blame,git log) using user-provided arguments ($ARGUMENTS). While these are standard development tools, they represent the skill's primary capability for interacting with the local filesystem and repository history. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data (git output) to extract commit SHAs and then passes this data into subsequent tool calls (
memory_commit_lookup). This creates a standard ingestion surface where malformed or malicious content in a git commit message or diff could theoretically influence the agent's behavior, although the risk is mitigated by the structured nature of the expected input (SHAs).
Audit Metadata