The Agent Skills Directory

[Ingestion of Untrusted Data]: The skill is designed to process external, potentially untrusted patch files and diffs. To manage this, the instructions explicitly command the agent to treat all subject text strictly as data and ignore any embedded instructions, which is a standard safety practice for preventing indirect prompt injection. Context: SKILL.md Workflow Step 2.
[Isolated Code Execution]: The instructions permit the agent to execute code from the analyzed patch for inspection purposes. To ensure safety, the skill mandates that such execution must be confined to a disposable, isolated workspace without credentials or network access. This prevents potential data exfiltration or system compromise from malicious patches. Context: SKILL.md Workflow Step 3.
[Dynamic Module Loading]: The provided Python validation script (validate_patch_risk_assessment.py) dynamically loads a local helper module (finalize_scan_contract.py) using importlib. This is an architectural pattern used for internal consistency checks within the plugin's file structure and does not involve downloading external, unverified code. Context: scripts/validate_patch_risk_assessment.py.
[External Tool Invocation]: The skill utilizes a Python interpreter to validate its own JSON output. This ensures that the generated risk assessments adhere to a specific schema and semantic rules before being finalized, adding a layer of structured validation to the agent's output. Context: SKILL.md Output section.

assess-patch-risk