The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands such as git, gh, and local TypeScript scripts (scripts/verify_closeout.ts, scripts/validate_marked_comments.ts) to verify implementation evidence, check mainline integrity, and validate workflow states. These actions are aligned with the skill's purpose and use safe execution methods (execFile) that avoid shell interpolation.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for using standard package managers (npm, pnpm, bun) to install necessary dependencies for its helper scripts. These references point to well-known registries and represent standard setup procedures for developer tools.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes user-generated content from Linear issues and comments to drive automation.
Ingestion points: External data is ingested via get_issue and list_comments tools from the Linear MCP server.
Boundary markers: The system uses specific comment markers (e.g., ) to delimit blocks intended for machine processing.
Capability inventory: Capabilities include updating Linear issue status (save_issue), posting comments (save_comment), and executing local Git/GitHub commands.
Sanitization: All ingested blocks are subjected to mandatory schema validation using ajv in the validate_marked_comments.ts script, which enforces a strict YAML structure and prevents the agent from acting on arbitrary text.
[SAFE]: The skill includes explicit security controls, such as requireRelativeRepoPath in verify_closeout.ts, which prevents path traversal attacks by validating that file paths do not contain absolute paths or directory traversal sequences (..).

linear-close