[jonbeckman]

Summary

Local e2e runs now default the language model to meta/muse-ember-1.0. Ids that do not start with anthropic/ already go through openai.chat() on the Cloudflare AI Gateway /ai/v1 route, so scripts/e2e/models.ts changes only that constant. E2E_DECISION_MODEL is unchanged.

Language-model calls already forward agents.default.providerOptions into generateText: field text in @e2e-dev/decision (askText) and waitFor / extract in the e2e runner. This sets providerOptions.openai.reasoningEffort to low. Installed ai 7.0.152 documents a top-level reasoning setting, and @ai-sdk/openai 4.0.101 maps providerOptions.openai.reasoningEffort onto Chat Completions reasoning_effort (that provider option wins when both are set). The runner only forwards providerOptions, so that is the option those call sites can carry. The decision model does not receive it.

The organization Actions variable E2E_LANGUAGE_MODEL is not changed. CI keeps using that variable until it is set to meta/muse-ember-1.0.

release:none: this is an internal test-harness change. It does not change the product for users, administrators, or deployment operators, so there is no .tegami/ entry.

Call site

e2e.config.ts:

providerOptions: { openai: { reasoningEffort: "low" } },

scripts/e2e/models.ts:

const DEFAULT_LANGUAGE_MODEL = "meta/muse-ember-1.0"

Proof

Same prompt, gateway id default, model meta/muse-ember-1.0, credential CF_AI_GATEWAY_E2E_TOKEN. CLOUDFLARE_API_TOKEN was not used.

Prompt: a library has 240 books, one quarter are checked out, and one third of the remainder are reserved. Reply with only the number.

Call	reasoning_effort sent	reasoning_tokens	completion_tokens	Answer
no reasoning option	omitted	154	164	120
reasoningEffort: "low"	low	108	118	120

Both responses were HTTP 200 with no SDK warnings. Low effort is honored, and on this prompt most completion tokens are still reasoning tokens (154 of 164, then 108 of 118).

Validation

• nub run config:check
• nub run typecheck
• nub run lint
• nub run format
• nub run test: 30 passed, 2 failed (13 files, 32 tests). The runner selected openai.chat/meta/muse-ember-1.0.

The two failures are opencode executes a real prompt and codex executes a real prompt. Both timed out after 180s because the container proxy could not program iptables in this VM (RULE_APPEND failed, missing kernel module). The isolate prompt test and the rest of the core suite passed. Four recorded agent.act steps replayed.

Checklist

• I added or updated tests for behavior changes, or explained why tests are not needed.
• I updated documentation and config examples when behavior or configuration changed.
• I added a .tegami/ release entry, or explained why this pull request uses release:none.
• I ran nub run config:check, nub run typecheck, nub run lint, nub run format, and nub run test.
• I did not include credentials, private data, internal URLs, or organization-specific defaults.
• I updated THIRD_PARTY_NOTICES.md for new third-party code or separately licensed assets.

  

[github-actions]

Tegami

This repository uses Tegami to manage releases. When your changes affect published packages, add a changelog file under .tegami/ before merging.

Create a changelog → · Changelog format

Release preview

Package	Bump	Version
solzero	major	2.0.0 → 3.0.0

This PR does not add changelog files. Pending changelogs from other branches are included in the preview above.

Run nub run tegami locally to create a changelog interactively.

_{Managed by Tegami.}

[github-actions]

🟢 e2e solzero-core: 32 passed

4 agent steps · 4 replayed from cache

Change 5 files changed since c7553a4
Filter decision kept all 30 tests (e2e.config.ts changed, so the decision filter left nothing out)

All 32 tests in 13 files

	Test	Agent	Time
🟢	tests/accounts.e2e.ts · 3 passed		7.0s
🟢	create, use and revoke an isolated user API key		3.6s
🟢	Better Auth 1.7 account APIs select the local account row id		161ms
🟢	Accounts settings shows GitHub as unlinked without OAuth providers		3.3s
🟢	tests/admin.e2e.ts · 1 passed		6.7s
🟢	administrator pages show live control-plane configuration		6.7s
🟢	tests/advanced-sessions.e2e.ts · 2 passed		7.8s
🟢	incognito sessions stay outside normal lists but open directly		3.6s
🟢	session MCP definitions and subagent preferences persist		4.2s
🟢	tests/api-contract.e2e.ts · 1 passed		424ms
🟢	exported API key and session schemas decode actual Worker responses		424ms
🟢	tests/auth.e2e.ts · 4 passed		17.1s
🟢	sign-in gate protects every application route		1.2s
🟢	credential sign-in opens the requested settings page		7.7s
🟢	anonymous callers cannot access any protected API group		115ms
🟢	sign-out removes authenticated UI access		8.1s
🟢	tests/auth.setup.e2e.ts · 2 passed		8.3s
🟢	authenticate isolated admin		5.9s
🟢	authenticate isolated peer		2.4s
🟢	tests/bots.e2e.ts · 1 passed		5.3s
🟢	create a bot and manage a temporary routine through the UI		5.3s
🟢	tests/branding.e2e.ts · 3 passed		325ms
🟢	the selected deployment brand reaches the browser		245ms
🟢	health identifies the configured application without exposing configuration		22ms
🟢	OpenAPI uses the selected deployment brand		58ms
🟢	tests/session-transfer.e2e.ts · 1 passed		2.3s
🟢	admin authentication transfer redeems once and retains its deep link		2.3s
🟢	tests/sessions.e2e.ts · 5 passed	1 step	42.8s
🟢	session tools, archive and restoration persist through the real Worker	1 step	6.0s
🟢	a second account cannot read or mutate another account's session		263ms
🟢	isolate executes a real prompt and persists its output		8.5s
🟢	opencode executes a real prompt and persists its output		14.5s
🟢	codex executes a real prompt and persists its output		13.5s
🟢	tests/settings.e2e.ts · 5 passed	2 steps	27.1s
🟢	all settings categories load real data		8.7s
🟢	runtime default persists after reload		6.4s
🟢	MCP deep link and legacy settings URL open the MCP tab		2.5s
🟢	secret editor creates, finds and deletes a disposable secret	1 step	4.5s
🟢	theme preference survives reload	1 step	5.0s
🟢	tests/skills.e2e.ts · 1 passed		8.2s
🟢	global skill defaults and per-user preferences persist independently		8.2s
🟢	tests/workflows.e2e.ts · 3 passed	1 step	9.2s
🟢	workflow authoring, rename, export, disable and archive persist	1 step	6.4s
🟢	manual workflow executes JavaScript and stores a real R2 artifact		2.7s
🟢	invalid workflow definitions are rejected before storage		101ms

_{e2e 0.21.3 · 3m 16s · local-cloudflare · run artifacts}