The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data from Linear issues, repository files, and git history, which creates a surface for indirect prompt injection attacks.
Ingestion points: METHODOLOGY.md §1a and §4 specify reading existing issues, milestones, documents, commit messages, and repository files (such as docs/*.md).
Boundary markers: METHODOLOGY.md §0 includes a dedicated "Trust boundary" section that explicitly instructs the agent to treat this content as "untrusted project data" and to ignore any instructions embedded within it.
Capability inventory: The skill is designed to perform write operations via the Linear MCP, including creating or updating issues, comments, status updates, and attachments.
Sanitization: The instructions in §0 require the agent to verify that all proposed changes follow the confirmed project scope and methodology gates before execution.
[EXTERNAL_DOWNLOADS]: The skill requires the use of the Linear MCP (Model Context Protocol) server for its core functionality.
Evidence: AUTH.md provide detailed setup instructions for connecting to the official Linear MCP integration within supported agent harnesses like Cursor and Claude Code. Linear is a well-known project management service.
[PERSISTENCE]: The skill provides instructions for setting up scheduled monitoring tasks within the user's environment.
Evidence: AUTOMATION.md outlines a process for creating a weekly health check automation using the platform's native features. This is a functional requirement for the skill's monitoring capabilities.
[COMMAND_EXECUTION]: The methodology documents the use of command-line tools for specific technical operations.
Evidence: METHODOLOGY.md §10 describes the use of curl for performing raw byte PUT operations to signed URLs as part of the Linear attachment upload flow. This is a standard procedure for interacting with the service's API.

linear-sync