The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses a 'Zero-Bailout Mandate' to automatically fetch and execute third-party code from remote registries if the required MCP tools are not found locally.
Evidence: Usage of npx -y chrome-devtools-mcp@latest and uvx devstudio-mcp for dynamic tool provisioning.
[COMMAND_EXECUTION]: The skill frequently executes shell commands to run diagnostic tools and manage local processes.
Evidence: Use of uv tool run windows-mcp serve, graphify query, and grep_search to perform audits.
[PRIVILEGE_ESCALATION]: The skill is instructed to interact with system-level OS components, potentially bypassing security boundaries of the browser or application under test.
Evidence: Using Windows-MCP to operate the native accessibility tree (UIA), Windows File Explorer dialogs, and system authentication modals.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external content (source code and PR files) while having significant capabilities (shell access, screen recording, network trace auditing).
Ingestion points: Ingests 'Generated Source Code Files' and 'Sprint Tickets & Acceptance Criteria' as primary inputs.
Boundary markers: None identified; the skill does not use specific delimiters or instructions to ignore embedded commands in the reviewed code.
Capability inventory: Access to shell commands (npx, uvx, graphify), UI automation (windows-mcp), screen recording (devstudio), and network auditing tools.
Sanitization: No evidence of input validation or sanitization before processing external code files.

developer-qa-reviewer