Summary

• Problem: Telegram currently allocates a reasoning preview lane whenever answer preview streaming is enabled, even when the session is not explicitly set to reasoning:stream.
• Why it matters: that wider callback surface makes it possible for hidden <think> or backend reasoning deltas to surface as Telegram preview messages on normal sessions.
• What changed: gate Telegram reasoning preview creation strictly to sessions whose persisted reasoning level is stream, and add a regression test that verifies no reasoning preview callback is exposed otherwise.
• What did NOT change (scope boundary): this PR does not remove intentional Telegram reasoning previews for reasoning:stream sessions, and it does not change shared model reasoning semantics.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes #
• Related #
• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause: extensions/telegram/src/bot-message-dispatch.ts enabled the reasoning draft lane with canStreamAnswerDraft || streamReasoningDraft, so normal answer preview streaming created a reasoning-preview callback even when the session reasoning level was off.
• Missing detection / guardrail: no regression test asserted that Telegram must not expose onReasoningStream outside explicit reasoning:stream sessions.
• Contributing context (if known): some backends and model outputs can produce reasoning-style deltas or <think> traces, so an overly broad Telegram preview lane is risky.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file: extensions/telegram/src/bot-message-dispatch.test.ts
• Scenario the test should lock in: normal Telegram sessions do not expose a reasoning preview callback and only create the answer preview lane unless the session reasoning level is stream.
• Why this is the smallest reliable guardrail: the bug is localized to Telegram preview-lane setup.
• Existing test that already covers this (if any): existing reasoning preview tests cover explicit reasoning:stream behavior, but not the off/default case.
• If no new test is added, why not: N/A

User-visible / Behavior Changes

• Telegram no longer surfaces hidden reasoning traces as preview bubbles on normal streaming sessions.
• Explicit reasoning:stream Telegram sessions keep their separate reasoning preview lane.

Diagram (if applicable)

Before:
[normal Telegram preview streaming] -> [answer lane + reasoning lane allocated] -> [possible accidental reasoning preview]

After:
[normal Telegram preview streaming] -> [answer lane only]
[reasoning:stream session] -> [answer lane + reasoning lane]

Security Impact (required)

• New permissions/capabilities? (No)
• Secrets/tokens handling changed? (No)
• New/changed network calls? (No)
• Command/tool execution surface changed? (No)
• Data access scope changed? (No)
• If any Yes, explain risk + mitigation:

Repro + Verification

Environment

• OS: macOS
• Runtime/container: local Codex worktree
• Model/provider: OpenAI Codex / GPT-5.4 suspicion path reviewed in code
• Integration/channel (if any): Telegram
• Relevant config (redacted): friend sample config reviewed separately; issue here is session-state + lane wiring, not the static config itself

Steps

1. Run a normal Telegram session without reasoning:stream.
2. Stream a reply containing hidden reasoning fragments or backend reasoning deltas.
3. Observe whether Telegram creates a separate reasoning preview bubble.

Expected

• Normal sessions should not expose a Telegram reasoning preview lane.

Actual

• Before this patch, Telegram allocated the reasoning lane whenever normal answer preview streaming was enabled.

Evidence

Attach at least one:

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

• Verified scenarios: traced the Telegram reasoning-lane setup, shared onReasoningStream routing, and added a regression test for the default/off session case.
• Edge cases checked: explicit reasoning:stream sessions still allocate the reasoning preview lane; default/off sessions now only allocate the answer preview lane.
• What you did not verify: the targeted pnpm test:serial run for the new Telegram regression test reached RUN but did not yield a clean exit signal in this workspace.

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

Compatibility / Migration

• Backward compatible? (Yes)
• Config/env changes? (No)
• Migration needed? (No)
• If yes, exact upgrade steps:

Risks and Mitigations

• Risk: if some Telegram flow was implicitly relying on a reasoning lane during non-stream sessions, this narrows that behavior.
  • Mitigation: explicit reasoning:stream sessions remain unchanged, which is the intended feature boundary.

AI / Testing Notes

• AI-assisted: yes
• Testing: lightly tested locally; targeted pnpm test:serial reached RUN but did not return a clean exit signal in this workspace.