workflow-implement-spec
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions to 'read that file in full yourself (every line, with no line limit)', which is an attempt to override default agent behavior and platform constraints regarding context management and file reading.
- [COMMAND_EXECUTION]: The skill instructs the agent to 'Regenerate this block with node .claude/scripts/lib/workflow-skill-contract.cjs --write', which involves executing a local script file with write permissions.
- [DYNAMIC_EXECUTION]: The workflow relies on the runtime execution of a local JavaScript utility (
workflow-skill-contract.cjs) to update its own contract and documentation blocks. - [INDIRECT_PROMPT_INJECTION]: The skill processes external 'canonical specs' and 'test-case sets' to drive code implementation, which creates an attack surface for malicious instructions embedded in those documents.
- Ingestion points: External canonical specifications, test-case sets, and user-provided prompts (SKILL.md).
- Boundary markers: None explicitly mentioned to separate untrusted spec data from instructions.
- Capability inventory: Writing code and tests (via $plan), spawning sub-agents (spawn_agent), and executing shell commands (node).
- Sanitization: No evidence of sanitization or validation for the content of the specifications being implemented.
Audit Metadata