Teach builder to author DeepAgents skills

README.md

@@ -7,10 +7,24 @@ a2a-pack agents from a natural-language description.
 
 The single `build(name, prompt)` skill:
 
-1. Builds an inner `deepagents` LangGraph configured with five tools:
+1. Builds an inner `deepagents` LangGraph configured with project tools and
+   packaged builder skills:
+   - `deepagent-agent-design` — when to use DeepAgents skills, subagents,
+     workspace backends, and deterministic tools
+   - `a2apack-agent-authoring` — how to shape `agent.py`, public `@skill`
+     schemas, runtime metadata, pricing, resources, and auth
+   - `deepagents-implementation-patterns` — concrete `create_deep_agent`
+     wiring for caller LLMs, skills, tools, and subagents
+   - `workspace-artifact-safety` — workspace grants, artifacts, bounded
+     subprocesses, and scope expansion
+   - `agent-quality-review` — pre-sandbox and pre-deploy checks for generated
+     agents
    - `list_agent_files(name)` / `write_agent_file(name, path, content)`
      / `read_agent_file(name, path)` — boto3 → MinIO, scoped to
      `agents/<name>/` in the caller's workspace
+   - `write_agent_skill(...)` — creates valid DeepAgents
+     `skills/<skill-name>/SKILL.md` bundles for generated agents, so rich
+     behavior lives in progressive-disclosure skills instead of fake tools
    - `test_agent_in_sandbox(name)` — tarballs the workspace project,
      spins a microVM with the public `a2a-pack` wheel installed, runs
      `a2a card` to verify the scaffold is valid