The Agent Skills Directory

[COMMAND_EXECUTION]: The methodology instructions in METHODOLOGY.md direct the agent to use the curl command to perform a PUT request for uploading file attachments. This process uses a dynamically generated signed URL provided by the Linear MCP tool prepare_attachment_upload. Although this is a standard workflow for the Linear API, the execution of shell commands with externally provided URLs represents a potential risk if the signed URL or metadata are manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and analyzing data from untrusted sources within the repository and the Linear workspace.
Ingestion points: Repository documentation (docs/ folder, README.md, MANIFESTO.md), CLAUDE.md, AGENTS.md, and Linear issue metadata (titles, descriptions, and comments).
Boundary markers: The instructions do not define specific delimiters for processed data, but the methodology explicitly requires the agent to confirm actions with the user before performing writes on existing projects.
Capability inventory: The skill uses Linear MCP tools to create and modify issues, projects, milestones, and labels, and utilizes shell execution (curl) for media attachments.
Sanitization: The methodology includes a specific rule to corroborate claims found in documentation against the user's own knowledge before acting on them, providing a human-in-the-loop safeguard.
[DATA_EXFILTRATION]: The skill is designed to read information from the local repository (such as roadmap documents and manifestos) and synchronize this data to the Linear platform. While this is the intended functionality, it establishes a pathway for moving local project data to an external service (Linear). The skill relies on the Linear MCP for these operations.

linear-sync