The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes code-review findings from external sources. The descriptions in these findings are used to guide the agent's verification and fixing process, creating a surface for potential instruction injection.
Ingestion points: Findings are ingested via a JSON format defined in the Input section of SKILL.md.
Boundary markers: Findings use a structured JSON schema, but there are no specific markers or instructions to prevent the agent from following natural language commands embedded in the body field.
Capability inventory: The skill has the ability to execute shell commands, write files, and delegate to other high-privilege skills.
Sanitization: No sanitization or validation of the input finding descriptions is performed before they are used to guide actions.
[COMMAND_EXECUTION]: In Step 6, the skill identifies the project's toolchain and executes build/test commands such as go build, npm install, or cargo test. These commands are executed in the local shell environment and can run arbitrary code defined in the project's configuration files.
[DYNAMIC_EXECUTION]: Step 3 requires the agent to verify bug reports by running code paths or creating and executing new tests. This dynamic execution is essential for verifying the defect but involves running code that may not yet be fully audited.

fix