[zimeg]

Summary

This pull request adds a Pydantic AI implementation of the coding agent in pydantic-ai/, ported from openai-agents-sdk/ with the starter agent's Pydantic AI app as the framework reference.

• Matches the OpenAI Agents SDK version's experience: the same 16 tools with the same names, arguments, and descriptions, the same prompts, and the same approvals, stop, and archive behavior. listeners/ and the framework-free agent/ files are copied unchanged, so only agent/agent.py, agent/guardrails.py, and agent/tools/ use Pydantic AI.
• Picks the model when a run starts with a SelectModel capability. Anthropic is preferred when ANTHROPIC_API_KEY is set: claude-sonnet-5-5 codes and claude-haiku-4-5 routes. Otherwise it uses OpenAI with the same gpt-6.1-sol and gpt-6-luna as the OpenAI version.
• Command approvals use deferred tools. A before_tool_execute hook raises ApprovalRequired, the run ends with DeferredToolRequests, and a click resumes it with deferred_tool_results. A dropped approval answers its call as not run, so the next turn's history stays valid.
• create_code_channel and archive_channel are output functions, so a successful call ends the run with no extra model call. A Slack error in either goes back to the model as a retry.
• The stop button skips every tool call with SkipToolExecution, a raising tool reaches the model as a failed tool call, and running out of turns ends with a message and keeps that turn's history.
• Long sessions compact old conversation with AnthropicCompaction and OpenAICompaction, in place of the OpenAI version's truncation="auto".
• Keeps both ANTHROPIC_ and OPENAI_ variables out of a command's environment, since this implementation can use either provider.
• Adds pydantic-ai/ to the lint and test matrices in ci.yml, dependabot, the root README, and AGENTS.md.

Deviations from the OpenAI version, all noted in pydantic-ai/AGENTS.md:

• If one step asks for more than one command that needs approval, the first is decided and the rest are answered as not run.
• Requires Python 3.11, as pydantic-ai-slim does. ruff ignores UP041 so the shared workspace.py keeps asyncio.TimeoutError.

Testing

• Copy pydantic-ai/.env.sample to .env, set ANTHROPIC_API_KEY, and start the app with slack run from pydantic-ai/. Repeat the checks below with only OPENAI_API_KEY set.
• Mention the agent in a thread with a question and confirm it answers in place, then with a task on a throwaway repo and confirm it links a new code channel.
• In the code channel, approve a test or build command and confirm it runs, then deny one and confirm the agent carries on without running it.
• Send a new message while an approval is pending and confirm the buttons show it was skipped and the agent picks up the new message.
• Press stop mid-turn and confirm the agent says it stopped and posts nothing more from that turn.
• Say the work is done and confirm the summary is shared back to your original message and the channel is archived.

🤖 Generated with Claude Code

[zimeg]

👾 Before merging a few comments for an agent nearest

[zimeg]

🗣️ note: Tests are planned to align after adjacent implementations land.

[zimeg]

📚 note: To self - will align later.

[zimeg]

🏁 Will merge so we can land more structure of related samples now.