The Agent Skills Directory

Potential Indirect Prompt Injection Surface: The methodology instructs the agent to ingest and analyze untrusted external data, including design documents and source code from the workspace (Steps 1 and 3). This creates a surface where instructions embedded within these external files could potentially influence agent behavior.
Ingestion points: Instructions in SKILL.md to read design documents and referenced files (Step 1) and use exploration tools on the codebase (Step 3).
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded commands within the processed files.
Capability inventory: The agent uses read_file, grep, and LSP for file access, alongside delegate, load_skill, and recipes for task orchestration.
Sanitization: No specific filtering or sanitization steps are defined for the content read from the codebase or design documents.
Codebase Interaction: The skill directs the agent to use tools like read_file, grep, and LSP functions (findReferences, incomingCalls) to examine project source code. While these capabilities are necessary for validating design implementations, they facilitate broad read access to the project environment.
Agent Orchestration: The skill uses orchestration patterns such as recipes, delegate, and load_skill to coordinate analysis across specialized components. These are standard platform features used here to manage complex, multi-stage workflows and delegate domain-specific analysis.

systems-design-review-methodology