The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions (in SKILL.md and references/exceptions.md) direct the agent to install missing libraries using npm install or pip install based on the 'Technical Details' section of a Linear issue. This creates a mechanism where the agent might execute commands to install untrusted or malicious packages if the issue description is manipulated by an external actor.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and parses data from external Linear issues using Linear:get_issue. The content in sections like 'Task Description' and 'Technical Details' is used as the primary source of instructions for code implementation and environment configuration.
Ingestion points: Linear issue descriptions are fetched and parsed in Step 1 and Step 3a of SKILL.md.
Boundary markers: The skill does not define explicit delimiters or 'ignore' instructions to distinguish between data and commands within the retrieved issue text.
Capability inventory: The agent has the authority to modify the local filesystem, post to the Linear API, and execute package managers (npm, pip).
Sanitization: No evidence of input validation or content filtering was detected to prevent the agent from following malicious instructions embedded in the issue metadata or body.

linear-issue-worker