The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Linear issue descriptions and comments, creating a potential surface for indirect prompt injection. However, the risk is mitigated by the use of structured data validation.
Ingestion points: Data enters the context through linear-status which utilizes get_issue and list_comments tools from the Linear MCP.
Boundary markers: The skill uses explicit HTML comment delimiters (e.g., ) to encapsulate and verify its own state data within issue descriptions.
Capability inventory: The agent has capabilities including Git operations, shell execution for verification commands, and Linear API write access.
Sanitization: All structured state blocks are validated against JSON schemas using the ajv library and local validation scripts (scripts/validate_marked_comments.ts).
[COMMAND_EXECUTION]: During the implementation phase, the skill executes package manager commands and verification scripts within the local workspace as defined by the implementation plan. This is a standard operational requirement for its delivery workflow.
[EXTERNAL_DOWNLOADS]: The skill retrieves metadata and issue state from the Linear platform. This communication is conducted through an authorized Model Context Protocol (MCP) server, which is the intended mechanism for this skill.

linear-deliver-feature