The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes data from Linear issues and comments, which represents a potential surface for indirect prompt injection.
Ingestion points: External data is ingested through Linear MCP tools (get_issue, list_comments) and read into the scripts/validate_marked_comments.ts script via the file system.
Boundary markers: Data blocks are delimited by specific HTML-style comment tags, such as  and , as defined in the main instructions and the validation script.
Capability inventory: The agent can read Linear data, perform file system operations, and execute shell commands to run validation scripts.
Sanitization: The skill implements strong sanitization by extracting content from delimited blocks, parsing it as YAML, and validating it against strict JSON schemas located in the schemas/ directory using the ajv library.
[COMMAND_EXECUTION]: The skill requires the execution of a local TypeScript script for state validation.
Evidence: SKILL.md instructs the agent to run scripts/validate_marked_comments.ts using either Bun or a Node.js TypeScript runner like tsx. The script performs file reads and data validation but does not exhibit dangerous side effects such as arbitrary command execution or network calls.

linear-status