UP-15: Agent capabilities declaration with A2A-compatible skills

[bertheto]

Summary

• Add structured skills[] metadata to agent registration (A2A AgentCard-compatible), alongside existing capabilities string tags
• New agent_update MCP tool and PUT /api/agents/{id} REST endpoint for post-registration metadata updates (token-authenticated)
• New GET /api/agents/{id} REST endpoint for single-agent detail
• Expose capabilities and skills in agent_list MCP tool, GET /api/agents, and chat://agents/active resource
• Minimal UI: .skills-badge pill in agent status item when skills are declared

Skill Schema (A2A AgentCard compatible)

Each skill object requires id and name; description, tags, and examples are optional:

{
  "id": "code-review",
  "name": "Code Review",
  "description": "Reviews code for style, security, and best practices",
  "tags": ["review", "security"],
  "examples": ["Review this PR for security issues"]
}

Changes

Backend

• src/db/models.py: AgentInfo.skills field (optional, after existing optional fields — backward-compatible)
• src/db/database.py: safe migration ALTER TABLE agents ADD COLUMN skills TEXT
• src/db/crud.py: agent_register + skills, new agent_get, new agent_update
• src/mcp_server.py: agent_register schema extended, agent_update tool added, agent_list description updated, chat://agents/active resource includes skills
• src/tools/dispatch.py: handle_agent_register, handle_agent_list, new handle_agent_update
• src/main.py: AgentRegister + skills, new AgentUpdate model, GET /api/agents + skills, new GET /api/agents/{id}, new PUT /api/agents/{id}

Web UI

• acb-agent-status-item.js: .skills-badge pill (count badge, supports array or JSON string)
• main.css: .skills-badge styles (dark + light theme)

Tests

• tests/test_agent_capabilities.py: 19 pytest (11 unit in-memory + 8 HTTP integration)
• frontend/src/__tests__/agent-capabilities.test.js: 6 Vitest DOM tests

All tests pass: 73/73 pytest + 30/30 Vitest

Test plan

• agent_register with skills=[] → response includes parsed skills array
• agent_list → each agent includes capabilities and skills
• agent_update with valid token → skills updated, response ok: true
• agent_update with wrong token → error returned
• GET /api/agents/{id} → full agent detail with skills
• GET /api/agents/{id} for unknown ID → 404
• PUT /api/agents/{id} → updates skills, returns updated agent
• Agent with skills in web UI → badge N skills visible in agent status panel
• Agent without skills → no badge rendered