[WRONG BRANCH] feat(terminal-guard): opt-in no-tool-call continuation guard for openai-chat providers

[TooSpace]

What

Extend the no-tool-call terminal continuation guard from the anthropic adapter to
openai-chat routed models, gated behind a new per-provider opt-in flag
terminalContinuationGuard (default off).

Fixes #1651.

Why

The guard added in #394 issues one bounded internal re-ask when a model announces work
(an edit/plan) but ends the turn without emitting a tool call. It is currently wired to
the anthropic adapter only. Self-hosted OpenAI-compatible gateways (GLM / Kimi-family and
similar) routed through openai-chat hit the exact same premature-completion pattern and
stop mid-work, because they never reach analyzeTerminalTurn.

Design / why opt-in

analyzeTerminalTurn's suspicious-no-tool-stop heuristic (the ACTIONABLE_REQUEST_RE /
PLAN_OR_COMPLETION_RE / WAITING_FOR_USER_RE regexes) was tuned on Anthropic turns.
Its false-continue rate on other model families and non-English output is not yet
characterized. The openai-chat adapter is shared by many registry providers, so turning
the guard on globally there could inject unexpected continuations for existing users
(xAI, z.ai, etc.).

To keep this strictly do-no-harm, the guard is enabled only when a provider sets
terminalContinuationGuard: true. Anthropic behavior is unchanged; every other provider
is unchanged unless it explicitly opts in.

Changes

• src/types.ts: add documented optional terminalContinuationGuard?: boolean to OcxProviderConfig (passthrough config bool, same pattern as parallelToolCalls / promptCacheKey).
• src/server/responses/core.ts: terminalGuardEnabled now also true for openai-chat when route.provider.terminalContinuationGuard === true (still excludes combo attempts and routed compaction).
• src/server/responses/terminal-guard.ts: guardTerminalEventStream runs analyzeTerminalTurn for openai-chat as well as anthropic; all other adapters still short-circuit to pass.
• tests/terminal-guard.test.ts: add coverage that an openai-chat stream gets exactly one continuation, and that an unrelated adapter (openai-responses) is never guarded.

Not included

No change to analyzeTerminalTurn's heuristics themselves. If maintainers later gain
confidence in the cross-family false-continue rate, flipping openai-chat to default-on
would be a separate follow-up.

Testing

• bun x tsc --noEmit clean.
• bun test for terminal-guard, terminal-guard-server, anthropic-tail-guard,
  anthropic-image-guard, bridge-terminal-singleness, responses-terminal-repair,
  openai-chat-hardening, openai-chat-eof, openai-chat-dangling-toolcalls,
  openai-chat-parallel-stream, parallel-tool-calls-optin, cl01-openai-chat-review-regressions:
  185 pass, 0 fail (includes the 2 new cases).

Review readiness checklist

This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:

• All CI tests are green on my local testing.

• I pushed my PR to the latest dev commit.

• I resolved all correct Codex and CodeRabbit findings.

• My PR is ready for review.

Summary by CodeRabbit

• New Features

  • Added an optional terminal continuation guard for supported chat providers.
  • The guard can limit repeated no-tool-call continuations and ensure clean completion events.
  • Anthropic-based responses now receive terminal guard analysis automatically.
  • OpenAI chat responses support the guard when explicitly enabled.

• Bug Fixes

  • Improved handling of assistant response boundaries and final completion signals.
  • Other response modes remain unaffected unless specifically supported or enabled.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[github-actions]

⏳ DRAFT

• wrong target branch (main); retarget to dev.

What to do

• Retarget this PR to dev — all contributions go to dev.
• Tick all four boxes in the PR description once you're done (currently 0/4).

Review readiness checklist

• ⬜ All CI tests are green on my local testing.
• ⬜ I pushed my PR to the latest dev commit.
• ⬜ I resolved all correct Codex and CodeRabbit findings.
• ⬜ My PR is ready for review.

0/4 boxes ticked.

Its title has been prefixed with [WRONG BRANCH].
This pull request was already a draft. Its draft status will be preserved after every issue above is resolved.
@TooSpace Tick the boxes once your local CI is green, your branch is on the latest dev commit, and every correct Codex and CodeRabbit finding is resolved.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: ddb1e566-e01b-4686-a35b-502f245d29df

📥 Commits

Reviewing files that changed from the base of the PR and between 1a937d8 and 64c1121.

📒 Files selected for processing (4)

• src/server/responses/core.ts
• src/server/responses/terminal-guard.ts
• src/types.ts
• tests/terminal-guard.test.ts

---

📝 Walkthrough

Walkthrough

The terminal continuation guard now supports opted-in openai-chat routes. Anthropic behavior remains enabled, while combo attempts, routed compaction, and openai-responses remain excluded.

Changes

Terminal guard extension

Layer / File(s)	Summary
Guard configuration and eligibility src/types.ts, src/server/responses/core.ts, src/server/responses/terminal-guard.ts	Adds the optional terminalContinuationGuard provider flag. Enables terminal analysis and bounded continuation for configured openai-chat routes while retaining existing exclusions.
Guard behavior validation tests/terminal-guard.test.ts	Verifies one openai-chat continuation, assistant-boundary emission, and one final done event. Confirms openai-responses remains unguarded.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 64c11

This is a localized, opt-in behavior change with no actionable merge-blocking risk remaining; it is merge-ready after normal checks and review.

Sequence Diagram(s)

sequenceDiagram
  participant Provider
  participant CoreResponses
  participant TerminalGuard
  participant OpenAIChat
  Provider->>CoreResponses: Enable terminalContinuationGuard
  CoreResponses->>OpenAIChat: Process routed turn
  OpenAIChat-->>CoreResponses: Suspicious no-tool-call termination
  CoreResponses->>TerminalGuard: Analyze terminal turn
  TerminalGuard-->>CoreResponses: Request bounded continuation
  CoreResponses-->>Provider: Emit assistant boundary and final done event

Loading

Possibly related PRs

• lidge-jun/opencodex#394: Introduces terminal-guard behavior extended by this change.
• lidge-jun/opencodex#419: Adds related terminal-guard logic in the same response-handling files.
• lidge-jun/opencodex#1569: Introduces the openai-chat adapter path targeted by this guard extension.

Suggested labels: review-ready

Suggested reviewers: ingwannu, lidge-jun, wibias

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the opt-in continuation guard for openai-chat providers, which is the primary change.
Linked Issues check	✅ Passed	The PR addresses issue #1651 by adding provider opt-in support and applying the existing bounded guard to openai-chat streams.
Out of Scope Changes check	✅ Passed	The changes are limited to guard configuration, adapter eligibility, terminal analysis, and focused tests related to issue #1651.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[TooSpace]

Reopening against dev per enforce-pr-target (this repo takes all contributions on dev, not main). Superseded by a dev-targeted PR.