The Agent Skills Directory

[SAFE]: The skill instructs the agent to use the mushi CLI tool and various MCP tools (e.g., get_report_detail, run_qa_story) to manage a testing pipeline. These operations are consistent with the skill's stated purpose of performing integration and smoke tests for the Mushi platform.
[SAFE]: Stage 3 involves crawling a live URL to map user stories. While this introduces an external data ingestion surface that could theoretically contain indirect prompt injections, it is a primary feature of the tool and is gated by explicit user instructions and a manual review/approval process for resulting proposals.
[SAFE]: Stage 4 and 5 involve the generation and execution of Playwright tests. Security risks associated with dynamic code generation are mitigated by the inclusion of an approval queue (--mode review) and a manual approval command (mushi tdd approve), ensuring human oversight before code execution.
[SAFE]: The skill references standard secret management practices by advising the user to have active BYOK keys for Anthropic and Firecrawl, without hardcoding or requesting the exposure of these credentials within the prompt or scripts.

mushi-integration