fix(models): use catalog for models list --all by sd1114820 · Pull Request #39678 · openclaw/openclaw

sd1114820 · 2026-03-08T09:42:31Z

Summary

openclaw models list --all was reading the raw runtime registry (discoverModels(...).getAll()) instead of the OpenClaw model catalog.

That meant synthetic catalog fallback entries from src/agents/model-catalog.ts (for example openai-codex/gpt-5.4, openai/gpt-5.4, and openai/gpt-5.4-pro) could exist in the official catalog but still not appear in CLI output.

This change makes models list --all use loadModelCatalog({ config: cfg }) for the list source, while still reusing runtime registry discovery for availability/auth hints and local filtering.

What changed

switched the --all code path in src/commands/models/list.list-command.ts from raw registry getAll() to loadModelCatalog()
preserved availability/auth behavior via availableKeys + provider fallback heuristics
preserved --local behavior by filtering against discovered runtime models when needed
added a focused test covering synthetic catalog-only openai-codex/gpt-5.4 visibility in --all output

Why

Official upstream already defines GPT-5.4 synthetic fallbacks in src/agents/model-catalog.ts, but models list --all bypassed that layer entirely. This caused a mismatch between the official catalog and CLI output.

Validation

Validated in a container (no host dependency pollution) with:

src/commands/models/list.list-command.forward-compat.test.ts
src/commands/models.list.auth-sync.test.ts
src/commands/models/list.status.test.ts

Result: 15/15 tests passed.

greptile-apps · 2026-03-08T09:44:54Z

Greptile Summary

This PR fixes models list --all so it reads from the OpenClaw model catalog (loadModelCatalog) rather than the raw runtime registry (discoverModels(...).getAll()), ensuring synthetic catalog fallback entries (e.g. openai-codex/gpt-5.4) are visible in CLI output. The non---all path and availability/auth hinting logic are preserved.

Issues found:

Runtime crash on catalog entries with no input field (list.list-command.ts:96–99): ModelCatalogEntry.input is optional (ModelInputType[] | undefined). For catalog-only entries without a discovered runtime counterpart, the fallback object { ...entry, baseUrl: "" } is cast to Model<Api> and passed to toModelRow, which calls model.input.join("+") unconditionally. If input is undefined the call throws a TypeError at runtime. The new test does not catch this because its mock always supplies input: ["text"]. A simple input: entry.input ?? [] guard in the fallback object fixes this.
Test assertion targets wrong invocation (list.list-command.forward-compat.test.ts:151): The existing "does not mark configured codex model as missing" test uses mock.calls[0] to read printModelTable output. With the new test inserted before it, calls[0] now refers to the first test's --all call, not this test's own call. The assertions pass coincidentally, but the test is no longer verifying what it claims to verify. Changing to .at(-1) (consistent with the rest of the suite) restores the intent.

Confidence Score: 3/5

The core catalog-switch logic is sound, but a runtime crash exists for catalog entries whose input field is absent.
The main behavioral fix is correct and well-scoped. However, the fallback model object spread from ModelCatalogEntry can carry input: undefined, which will throw in toModelRow for any real-world catalog entry lacking an input value. This path is reachable for synthetic fallbacks or opt-in provider models that omit the input field, so the risk is not theoretical. The test isolation issue is minor but further reduces confidence in test coverage.
src/commands/models/list.list-command.ts — the fallback model object construction at lines 96-99 needs an input null-guard before merge.

Comments Outside Diff (1)

src/commands/models/list.list-command.forward-compat.test.ts, line 151 (link)

Test silently asserts the wrong invocation

The newly-added test runs first and records a printModelTable call at index 0. This existing test then reads calls[0], which now points to the first test's --all invocation rather than this test's own invocation. The assertions pass coincidentally, but the test is no longer verifying its own call.

Use .at(-1) (consistent with the rest of the suite) so this test always references its own invocation:

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/commands/models/list.list-command.forward-compat.test.ts
Line: 151

Comment:
**Test silently asserts the wrong invocation**

The newly-added test runs first and records a `printModelTable` call at index `0`. This existing test then reads `calls[0]`, which now points to the first test's `--all` invocation rather than this test's own invocation. The assertions pass coincidentally, but the test is no longer verifying its own call.

Use `.at(-1)` (consistent with the rest of the suite) so this test always references its own invocation:



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

_{Last reviewed commit: 9ac774f}

greptile-apps · 2026-03-08T09:44:57Z

src/commands/models/list.list-command.ts

+          model: (discoveredModel ?? {
+            ...entry,
+            baseUrl: "",
+          }) as Model<Api>,


Runtime crash when catalog entry has no input field

ModelCatalogEntry.input is typed as ModelInputType[] | undefined (optional). When discoveredModel is undefined (i.e. the entry only exists in the catalog, not in the runtime registry), the fallback object { ...entry, baseUrl: "" } may carry input: undefined. This value is then cast to Model<Api> and passed to toModelRow, which unconditionally calls model.input.join("+") on line 162 of list.registry.ts — with no null-guard. This will throw TypeError: Cannot read properties of undefined (reading 'join') for any catalog-only entry that has no input in its catalog record (e.g. synthetic fallbacks derived from a template that lacked an input field, or opt-in provider models without an input value).

The existing test doesn't catch this because the mock always provides input: ["text"].

Suggested change

model: (discoveredModel ?? {

...entry,

baseUrl: "",

}) as Model<Api>,

model: (discoveredModel ?? {

...entry,

baseUrl: "",

input: entry.input ?? [],

}) as Model<Api>,

Prompt To Fix With AI

This is a comment left during a code review. Path: src/commands/models/list.list-command.ts Line: 96-99 Comment: **Runtime crash when catalog entry has no `input` field** `ModelCatalogEntry.input` is typed as `ModelInputType[] | undefined` (optional). When `discoveredModel` is `undefined` (i.e. the entry only exists in the catalog, not in the runtime registry), the fallback object `{ ...entry, baseUrl: "" }` may carry `input: undefined`. This value is then cast to `Model<Api>` and passed to `toModelRow`, which unconditionally calls `model.input.join("+")` on line 162 of `list.registry.ts` — with no null-guard. This will throw `TypeError: Cannot read properties of undefined (reading 'join')` for any catalog-only entry that has no `input` in its catalog record (e.g. synthetic fallbacks derived from a template that lacked an `input` field, or opt-in provider models without an `input` value). The existing test doesn't catch this because the mock always provides `input: ["text"]`. ```suggestion model: (discoveredModel ?? { ...entry, baseUrl: "", input: entry.input ?? [], }) as Model<Api>, ``` 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: 6eabe656f4

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-08T11:20:47Z

src/commands/models/list.list-command.ts


  if (opts.all) {
-    const sorted = [...models].toSorted((a, b) => {
+    const catalog = await loadModelCatalog({ config: cfg });


Fallback to discovered models when catalog is empty

Using loadModelCatalog as the sole source in --all can incorrectly hide all models when the catalog loader transiently fails and returns [] (it intentionally swallows errors in src/agents/model-catalog.ts and may return an empty list). In that case loadModelRegistry has already succeeded and provided discovered models, but this path still prints No models found, which is a user-visible regression from the previous behavior where --all listed runtime-discovered models.

Useful? React with 👍 / 👎.

dorukardahan · 2026-03-08T18:36:39Z

i opened #40160 as a smaller follow-up on latest main. it keeps discovered rows when catalog lookup returns [], but still fills synthetic catalog rows like openai-codex/gpt-5.4. i think it supersedes this one: #40160

steipete · 2026-03-08T23:31:16Z

Closing as superseded by #40160.

That PR lands the same models list --all synthetic-catalog visibility fix in a smaller/current shape on top of latest main.
EOF && gh pr close 39678 --repo openclaw/openclaw

fix(models): use catalog for models list --all

9ac774f

openclaw-barnacle bot added commands Command implementations size: S labels Mar 8, 2026

greptile-apps bot reviewed Mar 8, 2026

View reviewed changes

sd1114820 added 3 commits March 8, 2026 18:10

fix(tests): address greptile follow-up on catalog fallback

b6ebfd1

style(models): format list catalog mapping

786e159

Merge branch 'main' into fix/models-list-all-use-catalog

6eabe65

chatgpt-codex-connector bot reviewed Mar 8, 2026

View reviewed changes

sd1114820 mentioned this pull request Mar 8, 2026

models list --all bypasses model catalog synthetic fallbacks (e.g. GPT-5.4) #40068

Closed

dorukardahan mentioned this pull request Mar 8, 2026

fix(models): keep --all aligned with synthetic catalog rows #40160

Merged

steipete closed this Mar 8, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(models): use catalog for models list --all#39678

fix(models): use catalog for models list --all#39678
sd1114820 wants to merge 4 commits intoopenclaw:mainfrom
sd1114820:fix/models-list-all-use-catalog

sd1114820 commented Mar 8, 2026

Uh oh!

greptile-apps bot commented Mar 8, 2026 •

edited

Loading

Comments Outside Diff (1)

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!

dorukardahan commented Mar 8, 2026

Uh oh!

steipete commented Mar 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

sd1114820 commented Mar 8, 2026

Summary

What changed

Why

Validation

Uh oh!

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

Uh oh!

Greptile Summary

Confidence Score: 3/5

Comments Outside Diff (1)

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!

dorukardahan commented Mar 8, 2026

Uh oh!

steipete commented Mar 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

greptile-apps bot commented Mar 8, 2026 •

edited

Loading