spec-writing

Pass

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of Azure DevOps handoff JSON, which includes user-generated work item descriptions and pull request comments. This creates a surface for malicious actors to attempt to influence the agent's logic or override instructions during the spec-authoring process.\n
  • Ingestion points: The skill accepts an azure-devops handoff JSON object containing work_item details and pr_comments (documented in SKILL.md and rules/handoff-consumption.md).\n
  • Boundary markers: Instructions mandate segregating PR comments into a 'Review Context' section and providing clear output schemas to maintain structural integrity (references/workflow.md).\n
  • Capability inventory: The skill performs complex text generation and can execute local Python scripts for validation (scripts/validate-output.py).\n
  • Sanitization: Explicit rules in rules/handoff-consumption.md forbid the promotion of PR comments to requirements and require validation of the producer field and JSON structure before processing.\n- [DYNAMIC_EXECUTION]: The skill's test utility utilizes dynamic module loading to import logic from external scripts within the skill directory, which is a common but sensitive programming pattern.\n
  • Evidence: tests/run-validation.py uses importlib.util.spec_from_file_location to load scripts/validate-output.py.\n- [COMMAND_EXECUTION]: The skill is designed to run local Python scripts to validate the integrity and quality of its own output, providing the agent with executable capabilities over local assets.\n
  • Evidence: References to scripts/validate-output.py and tests/run-validation.py in SKILL.md and the skill's assets definition.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 29, 2026, 03:18 PM