The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically loads and executes logic from the local filesystem. Specifically, it searches for and follows instructions in project-local skills matching the pattern verify-* (e.g., in .cursor/skills/verify-*/SKILL.md). It also executes multiple Javascript files (init-evidence.mjs, run-capture-command.mjs, write-report.mjs, and verify-evidence.mjs) from a relative path (../build/scripts/user-acceptance/).
[REMOTE_CODE_EXECUTION]: The skill specifies a runtime dependency on agent-reviews, which is executed using npx agent-reviews. This involves downloading and running code from an external source at runtime.
[EXTERNAL_DOWNLOADS]: The skill references the pbakaus/agent-reviews repository on GitHub as a required runtime dependency.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically acceptance criteria from Linear issues and comments from GitHub Pull Requests.
Ingestion points: Linear issue descriptions (get_issue) and PR comments (list_comments, gh pr view).
Boundary markers: None explicitly defined in instructions to delimit untrusted data.
Capability inventory: The skill can execute shell commands via scripts, write to Linear issues via MCP, and interact with GitHub via the gh CLI.
Sanitization: No explicit sanitization or validation of the ingested acceptance criteria or comments is documented before they are used to drive verification actions.

verify