refactor(server): migrate to Agents SDK backend

[matheusgalvao1]

Purpose

Migrate server-side agent execution to OpenAI Agents SDK as the single backend path and remove legacy Responses adapter code.

What changed

• remove legacy OpenAI Responses adapter service and server backend-switching logic
• keep one WorkflowLLM implementation backed by OpenAI Agents SDK
• set explicit maxTurns to 20 for bounded agent-loop iterations
• update server dependency graph and lockfile for @openai/agents
• update AGENTS.md, CLAUDE.md, and README guidance to match the new backend and worktree submodule initialization requirements

Risk / Impact

• agent node execution behavior now depends exclusively on @openai/agents
• long-running or tool-heavy prompts now fail earlier with MaxTurnsExceededError at 20 turns
• no API contract changes to workflow routes or node schema

Linked issue

• none

Validation

• npm --workspace apps/server run typecheck
• npx eslint apps/server/src/index.ts apps/server/src/services/openai-agents-llm.ts
• npm run typecheck

[coderabbitai]

📝 Walkthrough

Walkthrough

The pull request migrates the LLM backend from the OpenAI Responses API to the OpenAI Agents SDK. Changes include replacing OpenAILLMService with a new OpenAIAgentsLLMService that dynamically loads the agents SDK and executes agents with a capped turn limit. The package.json dependency is updated from "openai" to "@openai/agents". The main application entry point is modified to instantiate the new service. Documentation files are updated to reflect the architectural shift and include guidance on initializing git submodules.

Suggested reviewers

• maria-codesignal

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: migrating from the legacy OpenAI Responses adapter to the OpenAI Agents SDK backend, which is the primary objective of the entire PR.
Description check	✅ Passed	The description clearly relates to the changeset, providing purpose, scope, risk assessment, and validation steps that align with the file changes and PR objectives.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

apps/server/src/services/openai-agents-llm.ts (1)

74-74: Add a timeout via AbortSignal to bound wall-clock runtime for sdk.run().

maxTurns alone doesn't prevent slow network or tool steps from extending request duration indefinitely. Use AbortSignal.timeout() to enforce a timeout:

Example implementation

const result = await sdk.run(agent, invocation.userContent, {
  maxTurns: MAX_AGENT_TURNS,
  signal: AbortSignal.timeout(10_000), // 10s timeout
});

Adjust the timeout value based on expected runtime for your agents.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@apps/server/src/services/openai-agents-llm.ts` at line 74, The call to
sdk.run(agent, invocation.userContent, { maxTurns: MAX_AGENT_TURNS }) needs a
wall-clock timeout; update the options passed to sdk.run (in the same scope
where agent, invocation.userContent and MAX_AGENT_TURNS are used) to include a
signal created via AbortSignal.timeout(<ms>) (e.g., 10_000 ms) so the request is
aborted after the desired duration; ensure any surrounding try/catch around
sdk.run handles the abort error appropriately.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@apps/server/src/services/openai-agents-llm.ts`:
- Line 74: The call to sdk.run(agent, invocation.userContent, { maxTurns:
MAX_AGENT_TURNS }) needs a wall-clock timeout; update the options passed to
sdk.run (in the same scope where agent, invocation.userContent and
MAX_AGENT_TURNS are used) to include a signal created via
AbortSignal.timeout(<ms>) (e.g., 10_000 ms) so the request is aborted after the
desired duration; ensure any surrounding try/catch around sdk.run handles the
abort error appropriately.

---

ℹ️ Review info

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 51b1681 and 0b188dd.

⛔ Files ignored due to path filters (1)

• package-lock.json is excluded by !**/package-lock.json

📒 Files selected for processing (7)

• AGENTS.md
• CLAUDE.md
• README.md
• apps/server/package.json
• apps/server/src/index.ts
• apps/server/src/services/openai-agents-llm.ts
• apps/server/src/services/openai-llm.ts

💤 Files with no reviewable changes (1)

• apps/server/src/services/openai-llm.ts