The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to construct shell commands using variables extracted from local configuration files (package.json) via shell pipelines. Specifically, the repository URL is extracted using grep, cut, and sed, then passed directly into a pnpm exec tsx command. This pattern is vulnerable to command injection if the repository URL field contains shell metacharacters like semicolons or backticks.
[PROMPT_INJECTION]: The skill facilitates the processing of untrusted data from external sources, leading to an indirect prompt injection risk.
Ingestion points: The agent is instructed to read, analyze, and apply changes from git diff files stored in the .diffs/ directory, which are fetched from external GitHub repositories.
Capability inventory: The agent has the authority to edit local source code, execute terminal commands (test and lint), and interact with the Pal MCP refactor tool.
Boundary markers: No delimiters or safety instructions are provided to distinguish between legitimate code changes and potential malicious instructions embedded in the diff content.
Sanitization: The skill lacks any mechanism to validate or sanitize the content of the diffs before the agent processes them or uses them to generate implementation plans.
[EXTERNAL_DOWNLOADS]: The skill automates the download of content from external GitHub repositories. While the documentation provides examples from a specific user, the underlying logic allows fetching data from any repository specified in a provider's package.json file.

sync-provider