The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like find to discover its own sub-skills within standard local agent directories (~/.claude, ~/.agents). This is a standard discovery mechanism for the vendor's integrated tool suite.
[COMMAND_EXECUTION]: The orchestrator executes standard project maintenance and validation commands, including bun test, bun run build, and vue-tsc, to verify the integrity and quality of generated code.
[EXTERNAL_DOWNLOADS]: The skill integrates with Figma, a well-known design service, to retrieve UI design specifications. This process uses designated MCP tools and targets official figma.com domains.
[PROMPT_INJECTION]: The skill provides a natural language "prompt mode" for code modification, which presents an attack surface for indirect prompt injection. This risk is managed through structural delimiters and a mandatory multi-agent QA cycle (Ralph Loop) that verifies all changes using automated test suites.
Ingestion points: Natural language prompts processed in the prompt-mode.md workflow.
Boundary markers: Untrusted input is encapsulated within specific Markdown sections to isolate it from system instructions.
Capability inventory: The agents possess file modification (Write, Edit) and shell execution (Bash) capabilities to perform development tasks.
Sanitization: Outputs are sanitized via mandatory linting, type-checking, and unit testing performed by independent QA agents.

peach-team-dev