The Agent Skills Directory

[SAFE]: The skill functions as a controlled workflow runner for manual QA testing, following a strict interactive process with human-in-the-loop verification.
[SAFE]: It implements defensive constraints by explicitly prohibiting the use of dangerous capabilities such as terminal execution (no terminal), database access (no DB), or GraphQL queries, even if the user is a developer.
[SAFE]: No external network dependencies, obfuscation, or remote code execution patterns were detected. The skill operates exclusively on local project files.
[SAFE]: The skill handles data exposure risks appropriately by restricting its file operations to specific, project-related naming patterns (qa-plan-*.md and qa-results-*.md) rather than accessing sensitive system or credential files.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it reads and executes instructions from external qa-plan-*.md files. This is consistent with its primary purpose as a QA runner.
Ingestion points: Reads data from qa-plan-*.md files in Step 1 and Step 2.
Boundary markers: Absent; the skill directly presents plan steps for the agent or user to perform.
Capability inventory: Browser CDP (Chrome DevTools Protocol) for UI interaction and file writing to the qa-results-*.md file.
Sanitization: Absent; the plan content is treated as authoritative testing instructions.

qa-run