The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection during the 'Story Mapping' stage. The agent is directed to crawl a live URL and pass the resulting data to an LLM for story drafting. Malicious content on the crawled page could potentially influence the agent's logic.
Ingestion points: Data ingested via mushi stories map --url <URL> in Stage 3.
Boundary markers: Absent; the skill does not specify delimiters or instructions to ignore embedded commands within the crawled content.
Capability inventory: The agent can execute shell commands (mushi CLI), perform SQL queries via MCP, and trigger test generation/execution.
Sanitization: No sanitization or validation of the crawled content is mentioned before it is processed by the AI model.
[COMMAND_EXECUTION]: The skill relies on executing a series of local CLI commands (e.g., mushi test, mushi reports, mushi stories) to verify the integration pipeline. These are expected within the scope of a testing skill.
[REMOTE_CODE_EXECUTION]: The skill generates and executes Playwright tests dynamically (mushi tdd gen and mushi tdd run). While this is a core feature of the Mushi platform for TDD, it involves the runtime execution of newly created code strings.

mushi-integration