fix: preserve GPT-5 conversation continuity after subtask completion

[roomote]

Description

This PR fixes an issue where GPT-5 models (specifically gpt-5-codex) were unable to read subtask results after completion. The root cause was that the parent task was unconditionally setting skipPrevResponseIdOnce = true after a subtask completed, which broke the conversation continuity chain that GPT-5 models rely on.

Problem

When using gpt-5-codex:

• Parent task launches a subtask
• Subtask completes and returns results
• Parent task resumes but cannot see the subtask results
• This happens because skipPrevResponseIdOnce causes the next API call to omit the previous_response_id, breaking continuity

Solution

Modified the completeSubtask() method to conditionally handle the skipPrevResponseIdOnce flag:

• For GPT-5 models: Do NOT skip the previous_response_id (maintains continuity)
• For other models: Continue existing behavior (skip to ensure full conversation is sent)

Changes

• Modified src/core/task/Task.ts to check if the model is GPT-5 before setting the skip flag
• Added comprehensive tests in src/core/task/__tests__/Task.spec.ts covering:
  • GPT-5 models preserving continuity
  • Non-GPT-5 models maintaining existing behavior
  • Edge case handling when model ID is undefined

Testing

• ✅ All existing tests pass
• ✅ New tests added specifically for GPT-5 subtask completion
• ✅ Type checking passes
• ✅ Linting passes

Fixes #8402

---

Important

Fixes GPT-5 conversation continuity issue after subtask completion by conditionally setting a flag in Task.ts.

• Behavior:
  • Fixes issue where GPT-5 models lost conversation continuity after subtask completion by modifying completeSubtask() in Task.ts.
  • For GPT-5 models, maintains continuity by not setting skipPrevResponseIdOnce.
  • For non-GPT-5 models, continues to set skipPrevResponseIdOnce to ensure full conversation is sent.
• Testing:
  • Added tests in Task.spec.ts to verify behavior for GPT-5 and non-GPT-5 models.
  • Tests cover edge cases, including undefined model ID.

^{This description was created by for f9a8ccd. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Self-review initialized: auditing my own code with the emotional range of a diff algorithm.

[roomote]

[P1] Unrelated git submodule/gitsubtree artifact committed. No matching .gitmodules update and not referenced by the code changes. Please drop this file from the PR.

[roomote]

[P1] Unrelated submodule/gitreference introduced. This appears accidental and unrelated to GPT‑5 continuity fix. Please remove from the PR.

[roomote]

[P1] GPT‑5 detection robustness. Model ids can be prefixed or vary in case across providers. Suggest normalizing and using includes:\n\nsuggestion\n\t\t\tconst modelId = this.api.getModel().id\n\t\t\tconst isGpt5Model = typeof modelId === \"string\" && modelId.toLowerCase().includes(\"gpt-5\")\n\n\nAlso consider centralizing this in a small helper to keep parity with other call sites.

[roomote]

[P2] Completeness. Add a test that, after completeSubtask() with GPT‑5, the subsequent attemptApiRequest() includes metadata.previousResponseId (and omits suppressPreviousResponseId). This validates the continuity chain beyond just the flag.