The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform arbitrary command-line operations, including "CLI invocation" and driving automation interfaces. It generates and executes these commands based on its analysis of the project requirements.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection due to its reliance on external data to drive its behavior.
Ingestion points: The agent reads and processes user requests, tickets, repository specifications, and project-local instructions found in verifying-* files located within the repository.
Boundary markers: The skill does not provide specific delimiters or instructions to the agent to disregard malicious directives that might be embedded in these external documents.
Capability inventory: The skill utilizes significant capabilities, including network access (API/Browser) and arbitrary shell command execution across various scripts.
Sanitization: Although the skill instructs the agent to redact secrets from captured evidence, there is no validation or sanitization of the logic ingested from the repository that dictates which commands are executed.
[DATA_EXFILTRATION]: The skill performs a combination of reading project files (tickets, specs, code) and performing network operations (API requests, browser interaction). While intended for verification, this pattern allows for potential data exfiltration if the agent is misdirected to send sensitive file content to an external endpoint.
[EXTERNAL_DOWNLOADS]: The verification process involves interacting with live web surfaces and APIs, which entails making external network connections to endpoints that may not be under the user's direct control.
[DYNAMIC_EXECUTION]: The skill dynamically identifies, loads, and executes instruction sets from project-local files matching the verifying-* pattern. This allows the behavior of the agent to be modified by files present in the working directory at runtime.

verifying-work