spec-driven-dev
Pass
Audited by Gen Agent Trust Hub on May 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a POSIX-compliant shell script (
pipeline.sh) to manage its state machine, perform Git operations (branching, merging, pushing), and track pipeline progress. The agent is additionally instructed to discover and run local verification commands (test, build, lint) from the project's build orchestrators as part of the implementation and review phases. - [PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection as it processes project files during analysis. 1. Ingestion points:
Makefile,package.json,.spec/config.yaml, and source code files are read in the Explore and Implementation phases. 2. Boundary markers: Templates do not use explicit markers for untrusted code analysis. 3. Capability inventory: Execution of project-defined build/test commands via the agent's tool environment. 4. Sanitization: The state machine shell script implements robust escaping and validation for its internal KV store, although project-provided commands are executed verbatim as intended for development automation. - [SAFE]: The skill's behavior is transparent, restricted to the local project environment, and lacks any patterns of obfuscation, malicious exfiltration, or unauthorized privilege escalation.
Audit Metadata