The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads and processes source code from the local environment, specifically targeting markers like TODO, FIXME, and HACK. It then uses the surrounding context to perform automated code modifications using the Write tool. This creates an attack surface where malicious instructions embedded in code comments could potentially mislead the agent into performing harmful refactoring or bypassing safety logic.
Ingestion points: The skill utilizes Read, Glob, and Grep to ingest content from the src/ directory, focusing on application logic and developer comments.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat external code content as untrusted data or to ignore embedded instructions.
Capability inventory: The skill is authorized to modify files via the Write tool and execute shell commands via the Shell tool.
Sanitization: The skill lacks sanitization or validation mechanisms for the input data extracted from the source files before it influences the agent's decision-making process.
[COMMAND_EXECUTION]: The skill executes shell commands to perform searches and manage project dependencies. While these actions are consistent with the skill's refactoring purpose, the use of pnpm and grep involves direct interaction with the system environment.
Evidence: The skill utilizes grep for locating patterns in the codebase and executes pnpm add react-error-boundary to install a well-known dependency.

correctness-and-error-handling