Refactor release hardening follow-ups by steipete · Pull Request #39959 · openclaw/openclaw

steipete · 2026-03-08T14:48:50Z

Summary

fail fast when the generated host-env Swift policy is stale and sync the checked-in output
guard release checks against new bundled extension root dependency gaps
centralize provider capability quirks and table-drive the Kimi/transcript regressions
block PR merge when the local prep branch has unpushed commits
simplify models-config merge preservation so stale baseUrls are only kept when provider API still matches

Validation

pnpm check:host-env-policy:swift
pnpm release:check
pnpm test -- test/release-check.test.ts
pnpm test -- src/agents/provider-capabilities.test.ts src/agents/pi-embedded-runner-extraparams.test.ts src/agents/transcript-policy.test.ts
pnpm test -- src/agents/models-config.providers.openai-codex.test.ts src/agents/models-config.fills-missing-provider-apikey-from-env-var.test.ts src/agents/models-config.runtime-source-snapshot.test.ts
pnpm build
pnpm check

cursor · 2026-03-08T14:48:54Z

PR Summary

Medium Risk
Touches release/merge automation and agent provider request-shaping logic; regressions could block releases/merges or alter tool payload compatibility for Anthropic-style providers.

Overview
Tightens release and merge hardening by (1) regenerating the macOS HostEnvSecurityPolicy.generated.swift output and moving the Swift policy check earlier in pnpm check, (2) adding a new release-check guard that fails if bundled extension dependencies drift from the root package.json (with an explicit allowlist), and (3) blocking scripts/pr merge-verify when the local prep branch HEAD no longer matches the previously pushed/prepared SHA.

Refactors agent/provider quirks by centralizing provider capability flags (e.g., Kimi aliases) and using them to gate Anthropic tool schema/tool_choice normalization and thinking-signature preservation; related tests are converted to table-driven cases.

Adjusts models-config merge behavior so an existing baseUrl is only preserved when the provider api still matches, preventing stale URLs from surviving provider API changes (with new regression coverage).

^{Written by Cursor Bugbot for commit 4c71606. This will update automatically on new commits. Configure here.}

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

cursor · 2026-03-08T14:50:45Z

src/agents/pi-embedded-runner/extra-params.ts

+            payloadObj.tool_choice = normalizeOpenAiStringModeAnthropicToolChoice(
+              payloadObj.tool_choice,
+            );
+          }


BaseUrl-based kimi detection path becomes dead code

Medium Severity

requiresAnthropicToolPayloadCompatibility can return true via baseUrl matching (e.g., a kimi.com/coding URL with a custom provider name), but the inner guards usesOpenAiFunctionAnthropicToolSchema(model.provider) and usesOpenAiStringModeAnthropicToolChoice(model.provider) both check the provider string against the capabilities table. When the match came from baseUrl alone (provider is not "kimi-coding"), both inner checks return false, so neither tool schema nor tool_choice normalization happens. The old code applied both normalizations unconditionally once the endpoint matched. This is a regression for users who configure a kimi.com/coding endpoint with a non-standard provider key.

Additional Locations (1)

src/agents/pi-embedded-runner/extra-params.ts#L805-L818

greptile-apps · 2026-03-08T14:52:39Z

Greptile Summary

This PR is a release hardening follow-up that centralizes provider-specific quirks into a new provider-capabilities.ts module, adds a bundled-extension root-dependency mirror check to the release pipeline, blocks merges when the prep branch has unpushed commits, and simplifies models-config merge logic to drop stale baseUrls only when the provider API changes.

Key changes:

provider-capabilities.ts — new table-driven capability registry (anthropicToolSchemaMode, anthropicToolChoiceMode, preserveAnthropicThinkingSignatures) shared by extra-params.ts and transcript-policy.ts. Kimi aliases (kimi-code → kimi-coding) are handled via the existing normalizeProviderId helper.
extra-params.ts — the URL-based kimi endpoint detection path in requiresAnthropicToolPayloadCompatibility now has no functional effect: it enters the outer wrapper block but the inner usesOpenAiFunctionAnthropicToolSchema / usesOpenAiStringModeAnthropicToolChoice guards both receive undefined (when model.provider is absent) and resolve to the native default, so neither tools nor tool_choice get normalized. This is a regression compared to the old isKimiCodingAnthropicEndpoint code where a URL match unconditionally triggered both transformations.
models-config.ts — removes the hard-coded AUTHORITATIVE_IMPLICIT_BASEURL_PROVIDERS set and replaces it with shouldPreserveExistingBaseUrl which compares the api field of the existing and incoming entries. Clean, provider-agnostic approach with a new test verifying API-change-driven base URL replacement.
release-check.ts — collectBundledExtensionRootDependencyGapErrors cross-checks bundled extension dependencies against the root package.json and fails fast when gaps deviate from the allowlist.
scripts/pr — verify_prep_branch_matches_prepared_head is inserted into merge_verify to abort when the local prep branch HEAD diverges from the prepared SHA.

Confidence Score: 3/5

Mostly safe, but the URL-based kimi endpoint path in extra-params.ts is now a functional no-op, which could silently break tool calls for configurations that rely on URL detection without an explicit provider: "kimi-coding" field.
The provider-capabilities refactor, models-config simplification, and release-check additions are clean and well-tested. The one concrete regression is in createAnthropicToolPayloadCompatibilityWrapper: when requiresAnthropicToolPayloadCompatibility fires via the kimi.com/coding URL heuristic but model.provider is not set, both inner normalization guards (usesOpenAiFunctionAnthropicToolSchema and usesOpenAiStringModeAnthropicToolChoice) receive undefined and return false, so neither the tools array nor tool_choice is transformed. Before this PR, the old isKimiCodingAnthropicEndpoint would have applied both transformations in this case. If any real deployment relies on URL-only detection, their tool payloads will be silently malformed after this change.
src/agents/pi-embedded-runner/extra-params.ts — specifically the interaction between requiresAnthropicToolPayloadCompatibility (URL path) and the nested usesOpenAiFunctionAnthropicToolSchema / usesOpenAiStringModeAnthropicToolChoice guards inside createAnthropicToolPayloadCompatibilityWrapper.

_{Last reviewed commit: 4c71606}

greptile-apps · 2026-03-08T14:52:43Z

src/agents/pi-embedded-runner/extra-params.ts

+        if (
+          payload &&
+          typeof payload === "object" &&
+          requiresAnthropicToolPayloadCompatibility(model)
+        ) {
          const payloadObj = payload as Record<string, unknown>;
-          if (Array.isArray(payloadObj.tools)) {
+          if (
+            Array.isArray(payloadObj.tools) &&
+            usesOpenAiFunctionAnthropicToolSchema(
+              typeof model.provider === "string" ? model.provider : undefined,
+            )
+          ) {
            payloadObj.tools = payloadObj.tools
-              .map((tool) => normalizeKimiCodingToolDefinition(tool))
+              .map((tool) => normalizeOpenAiFunctionAnthropicToolDefinition(tool))
              .filter((tool): tool is Record<string, unknown> => !!tool);
          }
-          payloadObj.tool_choice = normalizeKimiCodingToolChoice(payloadObj.tool_choice);
+          if (
+            usesOpenAiStringModeAnthropicToolChoice(
+              typeof model.provider === "string" ? model.provider : undefined,
+            )
+          ) {
+            payloadObj.tool_choice = normalizeOpenAiStringModeAnthropicToolChoice(
+              payloadObj.tool_choice,
+            );
+          }


URL-based kimi detection is now a no-op for normalization

requiresAnthropicToolPayloadCompatibility can return true via the URL-matching path (kimi.com/coding hostname) when model.provider is not set to "kimi-coding". However, both inner normalization guards (usesOpenAiFunctionAnthropicToolSchema and usesOpenAiStringModeAnthropicToolChoice) receive undefined in that case, which resolves to the default "native" capabilities — so neither tools array nor tool_choice is ever normalized.

Before this refactor, isKimiCodingAnthropicEndpoint returning true unconditionally applied both normalizations. Now, a user who configures api: "anthropic-messages" + baseUrl: "https://api.kimi.com/coding/..." without explicitly setting provider: "kimi-coding" will silently get no transformation, breaking their tool calls.

If the URL-based fallback is intentionally kept for future extensibility, consider also driving the capability resolution from the URL match so the nested guards can fire:

// inside createAnthropicToolPayloadCompatibilityWrapper onPayload if (payload && typeof payload === "object" && requiresAnthropicToolPayloadCompatibility(model)) { const payloadObj = payload as Record<string, unknown>; // Resolve capabilities from provider; fall back to kimi-coding caps when URL-matched const effectiveProvider = typeof model.provider === "string" ? model.provider : requiresAnthropicToolPayloadCompatibility(model) // URL match ? "kimi-coding" : undefined; if (Array.isArray(payloadObj.tools) && usesOpenAiFunctionAnthropicToolSchema(effectiveProvider)) { ... } if (usesOpenAiStringModeAnthropicToolChoice(effectiveProvider)) { ... } }

Alternatively, if the URL-based path is no longer needed now that provider IDs are normalized, remove the hostname check from requiresAnthropicToolPayloadCompatibility to avoid dead code.

Prompt To Fix With AI

This is a comment left during a code review. Path: src/agents/pi-embedded-runner/extra-params.ts Line: 902-926 Comment: **URL-based kimi detection is now a no-op for normalization** `requiresAnthropicToolPayloadCompatibility` can return `true` via the URL-matching path (kimi.com/coding hostname) when `model.provider` is *not* set to `"kimi-coding"`. However, both inner normalization guards (`usesOpenAiFunctionAnthropicToolSchema` and `usesOpenAiStringModeAnthropicToolChoice`) receive `undefined` in that case, which resolves to the default `"native"` capabilities — so neither tools array nor `tool_choice` is ever normalized. Before this refactor, `isKimiCodingAnthropicEndpoint` returning `true` unconditionally applied both normalizations. Now, a user who configures `api: "anthropic-messages"` + `baseUrl: "https://api.kimi.com/coding/..."` *without* explicitly setting `provider: "kimi-coding"` will silently get no transformation, breaking their tool calls. If the URL-based fallback is intentionally kept for future extensibility, consider also driving the capability resolution from the URL match so the nested guards can fire: ```ts // inside createAnthropicToolPayloadCompatibilityWrapper onPayload if (payload && typeof payload === "object" && requiresAnthropicToolPayloadCompatibility(model)) { const payloadObj = payload as Record<string, unknown>; // Resolve capabilities from provider; fall back to kimi-coding caps when URL-matched const effectiveProvider = typeof model.provider === "string" ? model.provider : requiresAnthropicToolPayloadCompatibility(model) // URL match ? "kimi-coding" : undefined; if (Array.isArray(payloadObj.tools) && usesOpenAiFunctionAnthropicToolSchema(effectiveProvider)) { ... } if (usesOpenAiStringModeAnthropicToolChoice(effectiveProvider)) { ... } } ``` Alternatively, if the URL-based path is no longer needed now that provider IDs are normalized, remove the hostname check from `requiresAnthropicToolPayloadCompatibility` to avoid dead code. How can I resolve this? If you propose a fix, please make it concise.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4c7160604d

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-03-08T14:54:03Z

src/agents/pi-embedded-runner/extra-params.ts

+            Array.isArray(payloadObj.tools) &&
+            usesOpenAiFunctionAnthropicToolSchema(
+              typeof model.provider === "string" ? model.provider : undefined,
+            )
+          ) {


Normalize Kimi tool payloads when baseUrl detection triggers

When requiresAnthropicToolPayloadCompatibility(model) is satisfied via the baseUrl fallback (for example, a custom provider using https://api.kimi.com/coding/), this branch still gates normalization on model.provider capabilities, so the tools[]/tool_choice rewrites are skipped. That leaves Anthropic-shaped tool payloads on a Kimi coding endpoint, which can reject tool calls; before this refactor, baseUrl detection alone applied the normalization.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-03-08T14:54:03Z

src/agents/models-config.ts

+
+  const existingApi = resolveProviderApi(existing);
+  const nextApi = resolveProviderApi(nextEntry);
+  return !existingApi || !nextApi || existingApi === nextApi;


Require API match before preserving existing provider baseUrl

This predicate keeps the old baseUrl whenever either side is missing api, so merge mode preserves stale endpoints from older/manual models.json entries that omit api even after the generated provider now points to a different API/backend (notably implicit openai-codex). That prevents automatic endpoint migration and can leave requests routed to incompatible URLs.

Useful? React with 👍 / 👎.

@danodoesdesign

* main: (70 commits) Refactor release hardening follow-ups (openclaw#39959) docs: clarify bot review conversation ownership (openclaw#39942) fix: harden talk silence timeout parsing (openclaw#39607) (thanks @danodoesdesign) talk: add configurable silence timeout transcript-policy: use named Set for anthropic signature-excluded providers transcript-policy: don't preserve thinking signatures for kimi-coding (openclaw#39798) fix: land mac universal release defaults (openclaw#33891) (thanks @cgdusek) Docs: clarify notarization handoff in mac release flow Docs: mark basic mac dist example as non-notarized Docs: clarify release build arch defaults for mac packaging macOS: default release app builds to universal binaries fix(issue-39839): address tool-call extra params parsing for kimi anthropic-messages docs: use alphabetical provider ordering fix: follow up openclaw#39321 and openclaw#38445 landings docs: note /landpr merge process fix: land Brave llm-context gaps (openclaw#33383) (thanks @thirumaleshp) feat: add Brave Search LLM Context API mode for web_search fix(feishu): restore @larksuiteoapi/node-sdk in root dependencies refactor: tighten codex inline api fallback follow-up macOS: set speech recognition taskHint for Talk Mode mic capture ...

* build: fail fast on stale host-env swift policy * build: sync generated host env swift policy * build: guard bundled extension root dependency gaps * refactor: centralize provider capability quirks * test: table-drive provider regression coverage * fix: block merge when prep branch has unpushed commits * refactor: simplify models config merge preservation

…39959) Partial pick — discards gutted files (models-config, pi-embedded-runner, transcript-policy, provider-capabilities, test/release-check.test.ts). Alive changes: - build: fail fast on stale host-env swift policy (move check to front) - build: guard bundled extension root dependency gaps - fix: block merge when prep branch has unpushed commits - HostEnvSecurityPolicy.generated.swift: restore blocked override keys Rebrand: openclaw→remoteclaw in release-check.ts PackageJson type and bundled extension dependency mirror validation. Cherry-picked-from: eba9dcc

…39959) (#1256) Partial pick — discards gutted files (models-config, pi-embedded-runner, transcript-policy, provider-capabilities, test/release-check.test.ts). Alive changes: - build: fail fast on stale host-env swift policy (move check to front) - build: guard bundled extension root dependency gaps - fix: block merge when prep branch has unpushed commits - HostEnvSecurityPolicy.generated.swift: restore blocked override keys Rebrand: openclaw→remoteclaw in release-check.ts PackageJson type and bundled extension dependency mirror validation. Cherry-picked-from: eba9dcc Co-authored-by: Peter Steinberger <[email protected]>

* build: fail fast on stale host-env swift policy * build: sync generated host env swift policy * build: guard bundled extension root dependency gaps * refactor: centralize provider capability quirks * test: table-drive provider regression coverage * fix: block merge when prep branch has unpushed commits * refactor: simplify models config merge preservation

steipete added 7 commits March 8, 2026 14:38

build: fail fast on stale host-env swift policy

01957da

build: sync generated host env swift policy

b270ee1

build: guard bundled extension root dependency gaps

d979225

refactor: centralize provider capability quirks

042e496

test: table-drive provider regression coverage

f1d7313

fix: block merge when prep branch has unpushed commits

6c252ec

refactor: simplify models config merge preservation

4c71606

openclaw-barnacle bot added app: macos App: macos scripts Repository scripts agents Agent runtime and tooling size: L maintainer Maintainer-authored PR labels Mar 8, 2026

steipete merged commit eba9dcc into main Mar 8, 2026
34 checks passed

steipete deleted the refactor/release-hardening-pass branch March 8, 2026 14:50

cursor bot reviewed Mar 8, 2026

View reviewed changes

greptile-apps bot reviewed Mar 8, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Mar 8, 2026

View reviewed changes

dunci mentioned this pull request Mar 8, 2026

fix(kimi-coding): use native Anthropic tool format #39881

Closed

JoshuaLelon mentioned this pull request Mar 8, 2026

fix(swiftformat): exclude generated security policy from formatters #39969

Merged

3 tasks

github-actions bot mentioned this pull request Mar 8, 2026

📡 Upstream Digest — 2026-03-08 16:18 UTC curtismercier/openclaw-mods#213

Open

alexey-pelykh mentioned this pull request Mar 10, 2026

Cherry-pick: upstream agents core updates (15 commits) remoteclaw/remoteclaw#899

Closed

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Refactor release hardening follow-ups#39959

Refactor release hardening follow-ups#39959
steipete merged 7 commits intomainfrom
refactor/release-hardening-pass

steipete commented Mar 8, 2026

Uh oh!

cursor bot commented Mar 8, 2026 •

edited

Loading

Uh oh!

Uh oh!

cursor bot left a comment

Uh oh!

cursor bot Mar 8, 2026

Uh oh!

greptile-apps bot commented Mar 8, 2026

Uh oh!

greptile-apps bot Mar 8, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Mar 8, 2026

Uh oh!

chatgpt-codex-connector bot Mar 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

steipete commented Mar 8, 2026

Summary

Validation

Uh oh!

cursor bot commented Mar 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

Uh oh!

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor bot Mar 8, 2026

Choose a reason for hiding this comment

BaseUrl-based kimi detection path becomes dead code

Uh oh!

greptile-apps bot commented Mar 8, 2026

Greptile Summary

Confidence Score: 3/5

Uh oh!

greptile-apps bot Mar 8, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 8, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Mar 8, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

cursor bot commented Mar 8, 2026 •

edited

Loading