fix(github-copilot): preserve reasoning IDs for Copilot Codex models by InvalidPandaa · Pull Request #71684 · openclaw/openclaw

InvalidPandaa · 2026-04-25T17:22:55Z

Summary

Problem: gpt-5.3-codex via GitHub Copilot fails with LLM request failed: provider rejected the request schema or tool payload on every request after the first reasoning turn. The raw provider error is 400 The encrypted content for item rs_2a72bff773173087 could not be verified. Reason: Encrypted content item_id did not match the target item id.
Why it matters: Any user running gpt-5.3-codex (or future Codex models) through GitHub Copilot cannot have multi-turn conversations — every second message hard-fails.
What changed: (1) connection-bound-ids.ts — skip all type: "reasoning" items during ID rewriting, not just those with a non-null encrypted_content string. (2) models.ts — add gpt-5.3-codex as a forward-compat clone target and update the catch-all regex to mark Codex-named model IDs as reasoning: true. (3) index.ts — add gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS.
What did NOT change: No change to message or function_call ID rewriting, no change to the Anthropic stream path, no change to non-Copilot providers.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Refs [Bug]: LLM request failed: provider rejected the request schema or tool payload. #68735 (GitHub Copilot reasoning-ID subcase)
Related fix(github-copilot): preserve reasoning ids with encrypted_content #71448 (the previous partial fix)
This PR fixes a bug or regression

Root Cause (if applicable)

Root cause: gpt-5.3-codex is not in the Copilot model registry and falls through to resolveCopilotForwardCompatModel's catch-all with reasoning: false. With reasoning: false, pi-ai never includes include: ["reasoning.encrypted_content"] in the request, so Copilot returns reasoning items with encrypted_content: null. The guard added in fix(github-copilot): preserve reasoning ids with encrypted_content #71448 — typeof item.encrypted_content === "string" — evaluates to false for null, so the connection-bound item ID is still rewritten to rs_<hash>. On the next turn, Copilot looks up its server-side reasoning state by item ID, finds the original connection-bound ID but receives rs_<hash>, and returns 400.
Missing detection / guardrail: The encrypted_content string check in fix(github-copilot): preserve reasoning ids with encrypted_content #71448 was too narrow. Copilot validates reasoning item IDs server-side regardless of whether the client holds the encrypted_content blob.
Contributing context: The second error in the logs (09:38 UTC) occurred after fix(github-copilot): preserve reasoning ids with encrypted_content #71448 landed (07:57 UTC), confirming that fix was insufficient for this model.

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/github-copilot/connection-bound-ids.test.ts, extensions/github-copilot/models.test.ts
Scenario the test should lock in: reasoning items with encrypted_content: null and reasoning items with no encrypted_content field must not have their IDs rewritten; Codex-named model IDs must resolve with reasoning: true from the catch-all.
Why this is the smallest reliable guardrail: the rewriting logic and model resolution are pure functions — no network or runtime needed.
Existing test that already covers this (if any): fix(github-copilot): preserve reasoning ids with encrypted_content #71448 added "preserves reasoning IDs when encrypted_content is present" — updated here to a single test covering all three cases (string, null, absent).
If no new test is added, why not: N/A — tests are added/updated.

User-visible / Behavior Changes

Multi-turn conversations with gpt-5.3-codex (and any future *-codex model) via GitHub Copilot no longer fail after the first reasoning turn. No config or defaults change.

Diagram (if applicable)

Before (broken):
reasoning item {id: "<base64>", encrypted_content: null}
  → encrypted_content check: null is not string → guard skips
  → looksLikeConnectionBoundId → true
  → id rewritten to rs_<hash>
  → Copilot 400: item_id mismatch

After (fixed):
reasoning item {id: "<base64>", encrypted_content: null}
  → type === "reasoning" → guard fires unconditionally
  → id preserved as "<base64>"
  → Copilot validates OK

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

Repro + Verification

Environment

OS: Raspberry Pi OS (server), macOS 15 (test runner)
Runtime/container: Node 22
Model/provider: gpt-5.3-codex / github-copilot
Integration/channel: Telegram
Relevant config: agents.defaults.models: ["github-copilot/gpt-5.3-codex"]

Steps

Configure OpenClaw with GitHub Copilot and model gpt-5.3-codex
Send a message via Telegram
Send a second message (triggers reasoning replay)

Expected

Second message receives a normal assistant reply

Actual

LLM request failed: provider rejected the request schema or tool payload.
Gateway log: 400 The encrypted content for item rs_2a72bff773173087 could not be verified.

Evidence

Failing test/log before + passing after
Trace/log snippets

Gateway logs showing the error (before fix):

07:42:22 warn agent/embedded "rawErrorPreview":"400 The encrypted content for item rs_2a72bff773173087 could not be verified. Reason: Encrypted content item_id did not match the target item id."
09:38:31 warn agent/embedded (same error, after #71448 was already merged)

Live test passing after fix:

[github-copilot-live] token ok via env (fetched)
[github-copilot-live] sending Responses request
[github-copilot-live] Responses request completed
Tests  1 passed (1)

Human Verification (required)

Verified scenarios: live test against real Copilot endpoint with gpt-5.3-codex; unit tests for all three encrypted_content states (string / null / absent) and for codex-pattern model IDs in the catch-all.
Edge cases checked: gpt-5.4 template cloning still works; gpt-5.3-codex prefers itself as template source over gpt-5.2-codex when both are in registry; non-reasoning models (gpt-5.4-mini, claude-sonnet-4.6) remain unaffected.
What you did not verify: end-to-end Telegram round-trip on the fixed server (requires deployment).

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

Risks and Mitigations

Risk: skipping all reasoning ID rewriting (not just encrypted ones) might leave stale connection-bound IDs in requests for models that don't need server-side validation.
- Mitigation: Copilot already rejects stale connection-bound IDs for message/function_call items — that's why rewriting was introduced. For reasoning items, the server lookup by original ID is the correct behavior even across connection changes; the rewrite was always wrong for this item type.

greptile-apps · 2026-04-25T17:24:46Z

Greptile Summary

This PR fixes a multi-turn failure for gpt-5.3-codex via GitHub Copilot by broadening the reasoning-item ID preservation guard from "skip only when encrypted_content is a string" to "skip unconditionally for all type: \"reasoning\" items," and by adding gpt-5.3-codex to the forward-compat target list and the xhigh thinking-level allow-list, plus a catch-all codex-substring check to infer reasoning: true.

The core logic is sound and the tests cover all three encrypted_content states (string / null / absent) as well as the new catch-all regex path.

Confidence Score: 4/5

Safe to merge; the fix is narrow, well-tested, and backward compatible — only P2 style nits remain.

No logic errors or security concerns found. The two findings are both P2: one is dead code in deriveReplacementId (the "rs" prefix branch is now unreachable) and one is a documentation gap around the dual role of gpt-5.3-codex in the template list. Neither affects runtime behavior.

No files require special attention.

Comments Outside Diff (1)

extensions/github-copilot/connection-bound-ids.ts, line 21 (link)

Dead "rs" prefix branch in deriveReplacementId

Since all type === "reasoning" items are now unconditionally skipped before deriveReplacementId is ever called, the type === "reasoning" ? "rs" branch is unreachable. It won't cause a bug, but leaving it silently implies the function still handles reasoning items which could mislead a future reader.

Prompt To Fix With AI

This is a comment left during a code review.
Path: extensions/github-copilot/connection-bound-ids.ts
Line: 21

Comment:
**Dead `"rs"` prefix branch in `deriveReplacementId`**

Since all `type === "reasoning"` items are now unconditionally skipped before `deriveReplacementId` is ever called, the `type === "reasoning" ? "rs"` branch is unreachable. It won't cause a bug, but leaving it silently implies the function still handles reasoning items which could mislead a future reader.



How can I resolve this? If you propose a fix, please make it concise.

Prompt To Fix All With AI

This is a comment left during a code review.
Path: extensions/github-copilot/connection-bound-ids.ts
Line: 21

Comment:
**Dead `"rs"` prefix branch in `deriveReplacementId`**

Since all `type === "reasoning"` items are now unconditionally skipped before `deriveReplacementId` is ever called, the `type === "reasoning" ? "rs"` branch is unreachable. It won't cause a bug, but leaving it silently implies the function still handles reasoning items which could mislead a future reader.

```suggestion
  const prefix = type === "function_call" ? "fc" : "msg";
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: extensions/github-copilot/models.ts
Line: 9-10

Comment:
**`gpt-5.3-codex` as a template is a no-op when it is itself the requested model**

When `resolveCopilotForwardCompatModel` is called for `gpt-5.3-codex`, the early-return guard on lines 33-36 has already confirmed the model is *not* in the registry. So `ctx.modelRegistry.find(PROVIDER_ID, "gpt-5.3-codex")` on line 42 will always return `null` for that path, making `"gpt-5.3-codex"` the first entry in `CODEX_TEMPLATE_MODEL_IDS` a no-op for `gpt-5.3-codex` requests specifically.

The template *is* useful when `gpt-5.4` is the target and `gpt-5.3-codex` is in the registry, so this isn't wrong — but a short clarifying comment would prevent future confusion about why the list includes a model that is also a target.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "fix(github-copilot): preserve all reason..." | Re-trigger Greptile}

greptile-apps · 2026-04-25T17:24:53Z

+const CODEX_FORWARD_COMPAT_TARGET_IDS: readonly string[] = ["gpt-5.4", "gpt-5.3-codex"];
+const CODEX_TEMPLATE_MODEL_IDS = ["gpt-5.3-codex", "gpt-5.2-codex"] as const;


gpt-5.3-codex as a template is a no-op when it is itself the requested model

When resolveCopilotForwardCompatModel is called for gpt-5.3-codex, the early-return guard on lines 33-36 has already confirmed the model is not in the registry. So ctx.modelRegistry.find(PROVIDER_ID, "gpt-5.3-codex") on line 42 will always return null for that path, making "gpt-5.3-codex" the first entry in CODEX_TEMPLATE_MODEL_IDS a no-op for gpt-5.3-codex requests specifically.

The template is useful when gpt-5.4 is the target and gpt-5.3-codex is in the registry, so this isn't wrong — but a short clarifying comment would prevent future confusion about why the list includes a model that is also a target.

Prompt To Fix With AI

This is a comment left during a code review. Path: extensions/github-copilot/models.ts Line: 9-10 Comment: **`gpt-5.3-codex` as a template is a no-op when it is itself the requested model** When `resolveCopilotForwardCompatModel` is called for `gpt-5.3-codex`, the early-return guard on lines 33-36 has already confirmed the model is *not* in the registry. So `ctx.modelRegistry.find(PROVIDER_ID, "gpt-5.3-codex")` on line 42 will always return `null` for that path, making `"gpt-5.3-codex"` the first entry in `CODEX_TEMPLATE_MODEL_IDS` a no-op for `gpt-5.3-codex` requests specifically. The template *is* useful when `gpt-5.4` is the target and `gpt-5.3-codex` is in the registry, so this isn't wrong — but a short clarifying comment would prevent future confusion about why the list includes a model that is also a target. How can I resolve this? If you propose a fix, please make it concise.

InvalidPandaa · 2026-04-25T17:38:20Z

(I'm on phone rn that's why I did this small change with Claude)

InvalidPandaa · 2026-04-25T18:17:01Z

Greptile Summary

This PR fixes a multi-turn failure for gpt-5.3-codex via GitHub Copilot by broadening the reasoning-item ID preservation guard from "skip only when encrypted_content is a string" to "skip unconditionally for all type: \"reasoning\" items," and by adding gpt-5.3-codex to the forward-compat target list and the xhigh thinking-level allow-list, plus a catch-all codex-substring check to infer reasoning: true.

The core logic is sound and the tests cover all three encrypted_content states (string / null / absent) as well as the new catch-all regex path.

Confidence Score: 4/5

Safe to merge; the fix is narrow, well-tested, and backward compatible — only P2 style nits remain.

No logic errors or security concerns found. The two findings are both P2: one is dead code in deriveReplacementId (the "rs" prefix branch is now unreachable) and one is a documentation gap around the dual role of gpt-5.3-codex in the template list. Neither affects runtime behavior.

No files require special attention.
Comments Outside Diff (1)
extensions/github-copilot/connection-bound-ids.ts, line 21 (link)

Dead "rs" prefix branch in deriveReplacementId

Since all type === "reasoning" items are now unconditionally skipped before deriveReplacementId is ever called, the type === "reasoning" ? "rs" branch is unreachable. It won't cause a bug, but leaving it silently implies the function still handles reasoning items which could mislead a future reader.
Prompt To Fix With AI
This is a comment left during a code review.
Path: extensions/github-copilot/connection-bound-ids.ts
Line: 21

Comment:
**Dead `"rs"` prefix branch in `deriveReplacementId`**

Since all `type === "reasoning"` items are now unconditionally skipped before `deriveReplacementId` is ever called, the `type === "reasoning" ? "rs"` branch is unreachable. It won't cause a bug, but leaving it silently implies the function still handles reasoning items which could mislead a future reader.



How can I resolve this? If you propose a fix, please make it concise.
Prompt To Fix All With AI
This is a comment left during a code review.
Path: extensions/github-copilot/connection-bound-ids.ts
Line: 21

Comment:
**Dead `"rs"` prefix branch in `deriveReplacementId`**

Since all `type === "reasoning"` items are now unconditionally skipped before `deriveReplacementId` is ever called, the `type === "reasoning" ? "rs"` branch is unreachable. It won't cause a bug, but leaving it silently implies the function still handles reasoning items which could mislead a future reader.

```suggestion
  const prefix = type === "function_call" ? "fc" : "msg";
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: extensions/github-copilot/models.ts
Line: 9-10

Comment:
**`gpt-5.3-codex` as a template is a no-op when it is itself the requested model**

When `resolveCopilotForwardCompatModel` is called for `gpt-5.3-codex`, the early-return guard on lines 33-36 has already confirmed the model is *not* in the registry. So `ctx.modelRegistry.find(PROVIDER_ID, "gpt-5.3-codex")` on line 42 will always return `null` for that path, making `"gpt-5.3-codex"` the first entry in `CODEX_TEMPLATE_MODEL_IDS` a no-op for `gpt-5.3-codex` requests specifically.

The template *is* useful when `gpt-5.4` is the target and `gpt-5.3-codex` is in the registry, so this isn't wrong — but a short clarifying comment would prevent future confusion about why the list includes a model that is also a target.

How can I resolve this? If you propose a fix, please make it concise.
_{Reviews (1): Last reviewed commit: "fix(github-copilot): preserve all reason..." | Re-trigger Greptile}

done

steipete · 2026-04-25T19:12:50Z

Codex maintainer update: I rebased this PR onto current main and pushed a small fixup commit.

What changed:

updated the stale stream wrapper test so reasoning IDs are preserved while message IDs still rewrite
tightened Copilot Codex model detection to delimiter-aware codex names
changed the issue reference from closing [Bug]: LLM request failed: provider rejected the request schema or tool payload. #68735 to Refs #68735, since this fixes the GitHub Copilot reasoning-ID subcase rather than the whole omnibus issue

Verification:

pnpm test extensions/github-copilot -> 9 files / 56 tests passed
pnpm lint:extensions -> passed
pnpm tsgo:extensions -> passed
pnpm tsgo:extensions:test -> passed
live .profile/auth-profile Copilot probe against gpt-5.3-codex: first turn produced a real reasoning item; replay with encrypted_content: null preserved the original reasoning ID and Copilot accepted the second turn (LIVE_TWO_OK)

steipete · 2026-04-25T19:19:29Z

Codex follow-up: CI exposed an unrelated stale mock in src/plugins/bundled-runtime-deps.test.ts after the rebase. I pushed a tiny test-only fix so the mocked node:child_process module still exports execFile/spawn for code imported through src/process/exec.ts.

Additional verification:

pnpm test src/plugins/bundled-runtime-deps.test.ts -> 43 tests passed
node scripts/run-oxlint.mjs --tsconfig tsconfig.oxlint.core.json src/plugins/bundled-runtime-deps.test.ts -> passed
pnpm tsgo:test:src -> passed

steipete · 2026-04-25T19:25:36Z

Codex follow-up 2: CI also exposed a brittle plugin-sdk fast-path fixture. The test wrote a CommonJS body to a .js sidecar, which can fall through to Jiti under package-type context in CI. I changed the synthetic fixture to .cjs, keeping the nodeRequire fast-path assertion explicit and stable.

Additional verification:

pnpm test src/plugin-sdk/channel-entry-contract.test.ts -> 8 tests passed
pnpm test src/plugins/bundled-runtime-deps.test.ts -> 43 tests passed
targeted oxlint for both touched test files -> passed
pnpm tsgo:test:src -> passed

@InvalidPandaa

… support The existing guard (8fd15ed) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa.

…tself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm

…lacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm

steipete · 2026-04-25T19:52:24Z

Landed. Thanks @InvalidPandaa!

Source SHA: 035532d
Merge commit: b64bfc5
Gate: PR CI green on the rebased source SHA; pnpm check:changed completed conflict markers, type lanes, lint, import-cycle, and policy guards locally; direct reruns passed pnpm test extensions/github-copilot, pnpm test src/plugin-sdk/channel-entry-contract.test.ts, pnpm tsgo:extensions, pnpm tsgo:extensions:test, pnpm tsgo:test:src, pnpm lint:extensions, and targeted core oxlint. Live gpt-5.3-codex Copilot reasoning replay probe passed earlier in review.

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (8fd15ed) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (2654ced) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (82fbd10) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (26a7738) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (0048671) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (8b1a21f) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

@InvalidPandaa

…penclaw#71684) * fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex support The existing guard (32a83e9) only skipped rewriting reasoning item IDs when encrypted_content was a non-null string. When gpt-5.3-codex is used via GitHub Copilot, the model falls through to the forward-compat catch-all with reasoning:false, so encrypted_content is never requested and arrives as null — bypassing the guard and causing a rewrite. Copilot validates reasoning item IDs server-side regardless of whether the client includes encrypted_content, so the rewritten id triggers the 400 error. Two changes: 1. connection-bound-ids.ts: skip ALL reasoning items unconditionally. Reasoning items always reference server-side state bound to their original ID; rewriting any of them breaks Copilot's lookup. 2. models.ts + index.ts: extend the forward-compat cloning logic to cover gpt-5.3-codex (adds it to the template-target set and to CODEX_TEMPLATE_MODEL_IDS so it can also serve as a template source for gpt-5.4). Adds gpt-5.3-codex to COPILOT_XHIGH_MODEL_IDS for the thinking profile. Thanks @InvalidPandaa. * docs(github-copilot): clarify gpt-5.3-codex is a no-op template for itself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): remove dead reasoning prefix branch in deriveReplacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm * fix(github-copilot): align reasoning id replay tests * test(plugin-sdk): use cjs sidecar for require fast path --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>

openclaw-barnacle Bot added the size: S label Apr 25, 2026

InvalidPandaa changed the title ~~fix(github-copilot): preserve all reasoning IDs and add gpt-5.3-codex…~~ fix(github-copilot): preserve reasoning IDs for all Copilot models, add gpt-5.3-codex support Apr 25, 2026

greptile-apps Bot reviewed Apr 25, 2026

View reviewed changes

steipete force-pushed the fix/github-copilot-preserve-reasoning-ids branch from dd9fbfa to 6d5eb01 Compare April 25, 2026 19:12

steipete changed the title ~~fix(github-copilot): preserve reasoning IDs for all Copilot models, add gpt-5.3-codex support~~ fix(github-copilot): preserve reasoning IDs for Copilot Codex models Apr 25, 2026

InvalidPandaa and others added 5 commits April 25, 2026 20:36

docs(github-copilot): clarify gpt-5.3-codex is a no-op template for i…

41645b5

…tself https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm

fix(github-copilot): remove dead reasoning prefix branch in deriveRep…

b8db36b

…lacementId https://claude.ai/code/session_01EAFmq4WyKkiUkVAqRXp4Bm

fix(github-copilot): align reasoning id replay tests

0bc4768

test(plugin-sdk): use cjs sidecar for require fast path

035532d

steipete force-pushed the fix/github-copilot-preserve-reasoning-ids branch from 2793c8c to 035532d Compare April 25, 2026 19:42

steipete merged commit b64bfc5 into openclaw:main Apr 25, 2026
64 checks passed

This was referenced Apr 25, 2026

[Bug]: Error: Encrypted content item_id did not match the target item id when using GitHub Copilot with GPT-5-mini in OpenClaw #71727

Closed

Support MCP hooks in the Codex harness #71707

Merged

github-actions Bot mentioned this pull request Apr 25, 2026

📡 Upstream Digest — 2026-04-25 20:30 UTC curtismercier/openclaw-mods#682

Open

clawsweeper Bot mentioned this pull request Apr 26, 2026

[Bug]: LLM request failed: provider rejected the request schema or tool payload. #68735

Closed

clawsweeper Bot mentioned this pull request May 2, 2026

fix(copilot): drop connection-bound replay ids #72623

Closed

ScientificProgrammer mentioned this pull request May 10, 2026

OpenAI Responses 400 'message' missing required 'reasoning' item on cross-provider failover from Gemini #80452

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(github-copilot): preserve reasoning IDs for Copilot Codex models#71684

fix(github-copilot): preserve reasoning IDs for Copilot Codex models#71684
steipete merged 5 commits into
openclaw:mainfrom
InvalidPandaa:fix/github-copilot-preserve-reasoning-ids

InvalidPandaa commented Apr 25, 2026 •

edited by steipete

Loading

Uh oh!

greptile-apps Bot commented Apr 25, 2026 •

edited

Loading

Comments Outside Diff (1)

Uh oh!

greptile-apps Bot Apr 25, 2026

Uh oh!

InvalidPandaa commented Apr 25, 2026

Uh oh!

InvalidPandaa commented Apr 25, 2026

Greptile Summary

Confidence Score: 4/5

Comments Outside Diff (1)

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		const CODEX_FORWARD_COMPAT_TARGET_IDS: readonly string[] = ["gpt-5.4", "gpt-5.3-codex"];
		const CODEX_TEMPLATE_MODEL_IDS = ["gpt-5.3-codex", "gpt-5.2-codex"] as const;

Uh oh!

Conversation

InvalidPandaa commented Apr 25, 2026 • edited by steipete Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Root Cause (if applicable)

Regression Test Plan (if applicable)

User-visible / Behavior Changes

Diagram (if applicable)

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Risks and Mitigations

Uh oh!

greptile-apps Bot commented Apr 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 4/5

Comments Outside Diff (1)

Uh oh!

greptile-apps Bot Apr 25, 2026

Choose a reason for hiding this comment

Uh oh!

InvalidPandaa commented Apr 25, 2026

Uh oh!

InvalidPandaa commented Apr 25, 2026

Greptile Summary

Confidence Score: 4/5

Comments Outside Diff (1)

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

InvalidPandaa commented Apr 25, 2026 •

edited by steipete

Loading

greptile-apps Bot commented Apr 25, 2026 •

edited

Loading