The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external sources such as Linear issues, comments, documents, and project-specific LINEAR.md files. This represents a vulnerability surface where instructions in an issue or comment could potentially influence agent behavior.
Ingestion points: The agent reads data via the Linear MCP server and local filesystem (LINEAR.md) in SKILL.md and references/artifacts.md.
Boundary markers: The instructions do not define specific delimiters or explicit instructions to ignore embedded prompts in processed content, although role-based logic acts as a partial mitigation.
Capability inventory: The skill allows for Linear metadata updates, comment/document publication, and git operations such as branch creation and merging.
Sanitization: No explicit sanitization or escaping of external content is mentioned before use in prompts or durable artifacts.
[COMMAND_EXECUTION]: The documentation includes example shell commands for users to configure the environment and mint OAuth tokens. These are provided as setup guidance for the user and are not automatically executed by the skill.
Evidence: Shell command blocks in references/setup.md demonstrating curl and pi/claude CLI usage.
[EXTERNAL_DOWNLOADS]: The setup instructions reference fetching data from well-known services for configuration.
Evidence: curl example targeting https://api.linear.app/oauth/token for token acquisition in references/setup.md.

linear