sg-code-audit
Pass
Audited by Gen Agent Trust Hub on Apr 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted content from the repository and passing it to AI agents. \n
- Ingestion points:
CLAUDE.md(Phase 2),.shipguard/learnings.yaml(Phase 3), and repository source files (Phase 4). \n - Boundary markers: The skill uses markdown headers like 'Project Rules' to delimit ingested content but lacks explicit 'ignore' instructions for embedded commands. \n
- Capability inventory: Dispatched agents can execute shell commands (
git,node,pytest,npx,go build), write files, and commit changes. \n - Sanitization: No sanitization or escaping of the ingested content is performed before interpolation into agent prompts.\n- [COMMAND_EXECUTION]: The skill executes multiple shell commands to discover the project stack, run syntax checks, and perform tests. \n
- Evidence: Uses
git,find,node,python3,npx tsc,go build,pytest, andnpx jest. \n - Context: While these commands are standard for auditing and testing, they execute content from the repository, which poses a risk if auditing untrusted or malicious codebases.
Audit Metadata