feature-flow
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content (code, instructions, and context) which could contain malicious instructions designed to influence the agents' behavior during the planning or implementation phases.
- Ingestion points: The main agent gathers context from the repository in
SKILL.md(Step 1). ThePlannerreads supplied context and repository instructions inreferences/planner.md. TheCoderreads repository instructions and existing code inreferences/coder.md. - Boundary markers: No explicit delimiters or instructions to ignore embedded commands within processed files are present.
- Capability inventory: The
Coderagent inreferences/coder.mdcan modify the working tree. TheUser Flow Testeragent inreferences/user-flow-tester.mdcan run CLI commands, call APIs, drive a browser, and invoke other skills. - Sanitization: The skill does not specify sanitization or validation steps for the content ingested from the repository.
- [DYNAMIC_EXECUTION]: The core multi-agent workflow involves generating new code and subsequently executing it to verify requirements.
- The
Coderagent implements the plan in the working tree, and theUser Flow Testeragent executes that code via the project's entry points (CLI, API, UI) to validate acceptance criteria. - [COMMAND_EXECUTION]: The testing agents are instructed to execute external commands as part of the verification process.
- Evidence:
references/user-flow-tester.mdcontains instructions for the agent to "run the CLI" and "invoke the skill" to exercise the real entry point of the software under test.
Audit Metadata