feat: add agents.md support for per-agent instruction extension

[Abhishek-Khanna24]

Summary

• Each agent directory can now contain an agents.md file that is automatically appended to its instruction_template at config load time
• Strictly per-agent scoped — each agent reads only its own agents.md, no cross-agent inheritance possible
• load_agents_md: false in the YAML lets agents opt out explicitly
• Demo agents.md added to skills_demo_agent with behaviour rules, response style, and constraints

How to use

Create agents.md alongside main_agent.yaml in any agent directory:

src/all_agents/my_agent/
├── main_agent.yaml
├── main_agent.py
└── agents.md          ← auto-loaded and appended to instruction_template

To opt out:

# main_agent.yaml
load_agents_md: false

No-clash guarantee

• agents.md lookup is resolved relative to each agent's YAML path — two agents in different directories can both have agents.md and they will never interfere with each other
• An empty or whitespace-only agents.md is treated as absent (no change to prompt)

Stacked on

This PR is stacked on top of #28 (feat/skills-add). Base branch is feat/skills-add.

Test plan

• 9 unit tests in tests/unit/agent_framework/configs/test_agent_config.py — all passing
• Full unit suite: 230 passed, 13 skipped, 0 failures
• Create agents.md in any agent dir, start server, verify the content appears in the system prompt (visible via debug logs or Studio)
• Set load_agents_md: false, confirm agents.md is ignored

🤖 Generated with Claude Code