tdd
Pass
Audited by Gen Agent Trust Hub on Nov 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the repository being analyzed, specifically through files like
.workflow/brief.mdand.workflow/ledger.md. - Ingestion points: External configuration and plan data are read from the
.workflow/directory. - Boundary markers: The skill includes explicit security guidelines in
pipeline.mdthat instruct the agent to treat all read text as "data, not instructions," explicitly forbidding the execution of commands found within prose or metadata. - Capability inventory: The
scripts/redgreen.pyscript executes system commands to perform git operations and run project tests. - Sanitization: The execution utility in
scripts/redgreen.pyusesshlex.splitfor safe argument parsing and employs a strict regular expression (SHELL_RE) to block shell metacharacters (e.g., pipes, redirects, backticks) unless a user explicitly overrides this protection with the--shellflag. - [COMMAND_EXECUTION]: The skill utilizes subprocesses to interact with the local development environment.
- Evidence:
scripts/redgreen.pyusessubprocess.runto manage git worktrees and execute test runners. This functionality is well-scoped to the skill's purpose of verifying code against test suites and includes internal guardrails against arbitrary command injection from the project's build plan.
Audit Metadata