The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill analyzes untrusted code from the target repository, which represents a surface for indirect prompt injection where malicious instructions in the source code could attempt to influence subagent behavior. Ingestion points: Target codebase files processed during discovery and deep review phases. Boundary markers: The skill uses coverage ledgers (such as work_ledger.jsonl) and completion receipts to track and verify subagent work. Capability inventory: Subagents are authorized to perform code analysis, write finding artifacts, and validate vulnerabilities. Sanitization: The workflow relies on multi-agent validation and structured reporting to verify the accuracy of findings.
[Local Script Execution]: To manage the scanning process, the skill executes several Python scripts located within its local plugin directory. These scripts perform deterministic tasks like generating file lists, ranking candidates, and final report assembly. Examples include generate_rank_input.py and finalize_scan_contract.py.
[Subagent Orchestration]: The skill uses a modular architecture where specific tasks are delegated to subagents. This provides a security boundary as each subagent has a restricted scope of work and is guided by explicit handoff instructions, preventing individual agents from inheriting excessive context.

security-scan