diagnostic-queue-runner-acrazie
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes diagnostic output from external compiler and linter tools, which could potentially contain adversarial instructions or malformed data.
- Ingestion points: Diagnostic tool outputs (stdout/stderr) are captured and parsed in SKILL.md and references/queue-guardrails.md.
- Boundary markers: The skill implements a 'Project loading boundary' requiring explicit activation and a 'Contract and permissions' section requiring scoped user approval for all actions.
- Capability inventory: The skill possesses capabilities for scoped file writes, repository reading, and the execution of build/analysis tools (e.g., tsc, cargo, go build).
- Sanitization: Instructions mandate reporting parse coverage and blocking any assignments based on unrecognized or multiline data to prevent manipulation.
- [COMMAND_EXECUTION]: The skill facilitates the execution of various system commands for diagnostic ingestion, including tsc, cargo, pyright, eslint, go build, and clang, based on the project's supported toolchain.
- [EXTERNAL_DOWNLOADS]: The skill references npx for executing JavaScript-based tools like tsc and eslint. Although npx can download packages from the npm registry, the skill explicitly warns that npx is not pre-authorized and requires the agent to verify project-supported versions and obtain user permission.
- [DYNAMIC_EXECUTION]: The skill includes a Python script (scripts/test_engineering_migration_contract.py) that uses the standard library unittest framework to perform static consistency checks on the skill's own documentation files.
Audit Metadata