Stop the AI flow when a worker or reviewer turn is incomplete

[luisorofino]

What does this PR do?

Adds an IncompleteResponseError exception and a require_complete flag to ReActProcess.start(). When require_complete=True, the method raises if the terminal stop reason is anything other than END_TURN (i.e. MAX_TOKENS or OTHER). The flag is opted in at the worker and reviewer call sites in AgenticPhase._start_task, _run_reviewer_once, and the worker-retry path in _drive_goal_loop. Subagents and the memory step are deliberately left on the default require_complete=False.

Motivation

During a demo run, a worker agent hit max_tokens mid-task — its output was cut off. Because AgenticPhase ignored the stop reason, the truncated result was handed straight to the goal reviewer. The reviewer (correctly) rejected the incomplete work, and the retry path then crashed the run.

The root issue is that a truncated turn is not a valid basis for goal validation or for advancing the flow — the model wanted to produce more output and was stopped. Rather than silently proceeding into the reviewer with half-finished work, we now fail fast with a clear, deliberate error.

Subagents keep their existing soft behavior ([SUBAGENT HIT MAX_TOKENS — RESPONSE MAY BE TRUNCATED] prefix) because a truncated subagent is a localized, recoverable degradation, not a reason to abort the whole run.

Review checklist (to be filled by reviewers)

• Feature or bugfix MUST have appropriate tests (unit, integration, e2e)
• Add qa/required if this PR needs QA validation, or qa/skip-qa if it does not. Exactly one of the two is required.
• If you need to backport this PR to another branch, you can add the backport/<branch-name> label to the PR and it will automatically open a backport PR once this one is merged

[datadog-prod-us1-4]

   

✨ Fix all issues with BitsAI

⚠️ Warnings

🚦 9 Pipeline jobs failed

PR | test / test (linux, ubuntu-22.04, ddev, ddev on Linux) / ddev on Linux     

PR | test / test (windows, windows-2022, ddev, ddev on Windows) / ddev on Windows     

PR | test / test-minimum-base-package (linux, ubuntu-22.04, ddev, ddev on Linux) / minimum-base-package-ddev on Linux     

View all 9 failed jobs.

🧪 2 Tests failed in 1 job

PR | run  

test_from_names_multiple_native from test_registry.py  

assert ('web_search', 'web_fetch') == ['web_search', 'web_fetch']
  
  Full diff:
  - [
  + (
        'web_search',
        'web_fetch',
  - ]
  + )

test_from_names_multiple_native from test_registry.py  

assert ('web_search', 'web_fetch') == ['web_search', 'web_fetch']
  
  Full diff:
  - [
  + (
        'web_search',
        'web_fetch',
  - ]
  + )

ℹ️ Info

No other issues found (see more)

❄️ No new flaky tests detected

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 88.50%

Useful? React with 👍 / 👎

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: b71179f | Docs | Datadog PR Page | Give us feedback!}

[luisorofino]

@codex

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b71179f44e

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Preserve token accounting for incomplete required turns

When require_complete=True and the terminal response is MAX_TOKENS/OTHER, this raises before returning a ReActResult or attaching the accumulated total_input/total_output. In the orchestrated phase path, _start_task only increments phase totals after process.start(...) returns, so a truncated worker/reviewer turn produces a failed checkpoint that omits the tokens spent on that API call; during goal retries it can also drop already accumulated reviewer/compaction tokens. Please carry the token totals on the exception or record them before raising so failure checkpoints remain accurate.

Useful? React with 👍 / 👎.

[dd-octo-sts]

Validation Report

All 21 validations passed.

Show details

Validation	Description	Status
agent-reqs	Verify check versions match the Agent requirements file	✅
ci	Validate CI configuration and code coverage settings	✅
codeowners	Validate every integration has a CODEOWNERS entry	✅
config	Validate default configuration files against spec.yaml	✅
dep	Verify dependency pins are consistent and Agent-compatible	✅
http	Validate integrations use the HTTP wrapper correctly	✅
imports	Validate check imports do not use deprecated modules	✅
integration-style	Validate check code style conventions	✅
jmx-metrics	Validate JMX metrics definition files and config	✅
labeler	Validate PR labeler config matches integration directories	✅
legacy-signature	Validate no integration uses the legacy Agent check signature	✅
license-headers	Validate Python files have proper license headers	✅
licenses	Validate third-party license attribution list	✅
metadata	Validate metadata.csv metric definitions	✅
models	Validate configuration data models match spec.yaml	✅
openmetrics	Validate OpenMetrics integrations disable the metric limit	✅
package	Validate Python package metadata and naming	✅
qa-label	Validate the pull request declares whether it needs QA for the next Agent release	✅
readmes	Validate README files have required sections	✅
saved-views	Validate saved view JSON file structure and fields	✅
version	Validate version consistency between package and changelog	✅

View full run