The Agent Skills Directory

[COMMAND_EXECUTION]: The skill identifies and executes shell commands defined in the project's configuration files, such as package.json scripts (test, lint, build, typecheck), and runs standard development tools like pytest, cargo, and supabase.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and acts upon untrusted data from Linear issues, including acceptance criteria, descriptions, and external test plans.
Ingestion points: Acceptance criteria and descriptions in Linear issue bodies (Phase 1); test plans downloaded from issue attachments (Phase 0.1).
Boundary markers: No specific delimiters or safety instructions are used to isolate instructions within the issue content from the agent's core system prompts.
Capability inventory: The agent has broad shell execution capabilities (Phase 3), file system access (Phase 4d), and the ability to interact with external APIs and project databases.
Sanitization: The skill lacks explicit sanitization or validation logic for content retrieved from Linear before using it to define test scenarios and expected results.
[EXTERNAL_DOWNLOADS]: The skill uses curl to download a test plan from a URL provided within a Linear issue attachment (curl -L "<test_plan_attachment_url>" -o /tmp/qa-test-plan.md).
[DYNAMIC_EXECUTION]: The agent is instructed to dynamically generate and execute Node.js or Playwright helper scripts within the project repository to facilitate UI verification (Phase 4d).
[CREDENTIALS_UNSAFE]: The skill accesses and reuses sensitive authentication state files (e.g., Playwright storageState JSON) to perform authenticated UI testing (Phase 4c).

qa