The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the target directory, creating a vulnerability surface where malicious instructions embedded in the codebase (e.g., in README files, documentation, or code comments) could influence the behavior of the discovery, tracing, or validation agents.
Ingestion points: The skill reads source files, documentation, manifests, and configuration files from the user-provided target-directory as defined in the Discovery Agent Instructions and Tracing Agent Instructions in references/agent-instructions.md.
Boundary markers: The instructions lack explicit delimiters or safety prompts to ensure agents ignore instructions found within the analyzed codebase, relying on the agent to distinguish between 'data' (the code) and 'instructions'.
Capability inventory: The skill performs file reading, file writing (creating the walkthrough/ directory and various markdown reports), and orchestrates multiple parallel sub-agents to process information.
Sanitization: There is no evidence of sanitization or filtering logic present in SKILL.md or references/agent-instructions.md to identify or neutralize potential prompt injection attacks hidden in the repository content.

linear-walkthrough