inline-execution
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and execute instructions from external 'plan' and 'spec' files. If these files are sourced from an untrusted environment, they could contain malicious instructions or shell commands that the agent would execute during its automated task loop.
- Ingestion points: Plan files in
.godmode/runs/<plan-slug>/progress.md, individual task briefs, and external project specifications. - Boundary markers: The skill does not specify any delimiters or instructions to treat data within the plan as untrusted; it explicitly directs the agent to 'execute it exactly'.
- Capability inventory: The agent is authorized to write files, perform Git commits, and execute shell commands for testing purposes.
- Sanitization: There is no requirement to sanitize or validate the plan's content before implementation or execution.
- [COMMAND_EXECUTION]: The skill instructions mandate the execution of shell commands provided in the plan to perform test-driven development.
- Evidence: Step 4 and Step 6b of the protocol instruct the agent to run tests and compare output, effectively executing any command defined as a 'test' in the ingested plan.
Audit Metadata