The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external codebases, which constitute untrusted data. This creates an attack surface where an attacker could place malicious instructions inside code comments, documentation, or string literals to manipulate the agent's analysis or subsequent actions.
Ingestion points: The agent is guided by SKILL.md to use tools like Read, Grep, Glob, and LSP to scan and parse project files, including database schemas, service layers, and configuration files.
Boundary markers: The skill does not provide specific boundary markers or instructions to help the agent distinguish between code content and potential injection attempts during the analysis process.
Capability inventory: The tools listed in the frontmatter (Read, Grep, Glob, LSP) are focused on discovery and analysis. The skill itself does not specify tools for file modification or network communication, which limits the potential damage from an injection attack.
Sanitization: There are no instructions for sanitizing or filtering the content found within the target codebase before the agent processes it.

codebase-analysis