eve 0.6.0-beta.8 → 0.7.0

package/CHANGELOG.md

@@ -1,5 +1,283 @@
 # eve
 
+## 0.7.0
+
+### Minor Changes
+
+- aa5cfdc: Connection and tool `auth` definitions accept an optional `displayName`, and authorization challenges carry it to channels — sign-in buttons now show e.g. "Sign in with Salesforce" instead of a title-cased file name. The value is presentation-only (scope, token cache keys, and callback URLs stay keyed by the path-derived name), a definition-level `displayName` wins over one the strategy stamps on the challenge, and channels keep title-casing the name when neither is set.
+- 20f0fc8: The dev TUI's `/channels` is now an action list: pick an unregistered channel to run its add flow, return to the repainted list (added channels show locked), and leave with the Done row or Esc. Web Chat (the Next.js app) is now addable in projects that already have the scaffolded `agent/channels/eve.ts` session channel — the row locks only when the project actually depends on `next`, and the REPL channel shows as its own locked row.
+- fe98d0b: Compaction is now owned by the framework instead of individual tools. The per-tool `onCompact` hook (and the `CompactionInput` / `CompactionHookResult` types) is removed: the framework automatically preserves its own tool state across compaction — it resets read-before-write tracking and re-injects the active todo list. `defineReadFileTool` no longer accepts an `onCompact` option, and the agent-info response drops the now-meaningless `hasCompactionHook` field.
+- 20f0fc8: The dev TUI now discovers its slash commands: typing `/` opens a suggestion list above the prompt with each command's argument hint and description — `↑`/`↓` move the highlight, `Tab` completes, `Enter` runs, `Esc` dismisses — and a new `/help` command prints the full table. Unknown `/text` is still sent to the agent as a normal message.
+- 407c6e2: The dev TUI startup header is now a single `▲ eve <agent name>` brand line plus one rotating tip ("Use /channels to add more ways to reach your agent.", "Use /deploy to see your agent go live.", "Type /help to see every command."). The config rows (Model, Instructions, Tools, Skills, Subagents, Server) and the key-hints line are gone. The model moved to the status line, the empty input row shows a dim `Type to chat · / for commands` placeholder, discovery error/warning counts still render, and `eve info` keeps the full configuration detail.
+- edf9f1a: Redesign the `eve dev` terminal UI and add `Client.info()`.
+
+  The TUI now streams its transcript into the terminal's native scrollback instead of a bordered alt-screen, so scrolling, copy/paste, and the transcript all survive after you exit. The visual language follows the Vercel / Next.js CLI: a monochrome base with the `▲` brand mark, colored gutter glyphs per speaker, the `dots` spinner, and status icons (`✓` done, `⨯` error). Tool calls collapse to a one-line summary (`✓ get_weather  city="SF" → 73°F`) that expands under `--tools full`, subagent work is indented beneath a `◆` header, and a startup header prints the connected agent's model, instructions, tools, skills, and subagents. Subagents now default to `auto-collapsed`.
+
+  The prompt input gains shell-style editing: `↑`/`↓` cycle through your previously sent messages, `←`/`→`/Home/End/`Ctrl+A`/`Ctrl+E` move the caret, and `Ctrl+U`/`Ctrl+K`/`Ctrl+W` kill the line/word. Escape sequences that arrive split across reads (common with PTYs) are reassembled, and pasted text is sanitized of stray control bytes.
+
+  Failures read more cleanly: a terminal `session.failed` (or a dead-socket transport error) no longer leaves you stuck — the TUI starts a fresh session before the next prompt and notes it inline, so you can keep going. A failure cascade (`step.failed` → `turn.failed` → `session.failed`) renders one error block instead of three, and code bugs escaping user code show their stack trace dim beneath the headline (capped to a dozen lines; recognized provider/config failures keep their curated one-line summary). Error blocks no longer double-print a `Code: Code:` prefix and render docs links in cyan, and denied tool approvals settle to a `→ denied` line instead of a frozen `?`.
+
+  Captured server `stdout`/`stderr` renders as quiet, indented log runs: consecutive lines from the same source coalesce behind one `stdout ·` / `stderr ·` label with a hanging indent, runs get a blank line of breathing room on both sides, and nothing is ever truncated — a transcript can't be clicked open, so the full output is always shown.
+
+  `Client.info()` fetches the agent inspection payload (`GET /eve/v1/info`) — model id, instructions source, tools, skills, subagents, schedules, and sandbox — for both local and remote targets, returning the new `AgentInfoResult` type.
+
+- 407c6e2: The dev TUI footer now ends with a persistent status line: the agent's model, the session's token flow (`⁕ ↑ 394.4K ↓ 4.3K`, input up, output down, `⁕ ↑ 0 ↓ 0` at startup), the linked Vercel project and team, and a `deploy pending` marker that appears when `/channels` adds a channel and clears when `/deploy` ships it. The Vercel segment stays hidden until the directory is linked. Token usage moved here from the working row, and `--context-size` appends a context-fill percentage.
+- 5a6ac17: Add a required `evals/evals.config.ts` (authored with `defineEvalConfig`) that declares run-wide eval defaults: a mandatory scorer `model`, plus optional run-level `reporters`, `maxConcurrency`, and `timeoutMs`. Model-backed scorers now fall back to the config `model`, so `model` is optional on `defineEval` and a shared reporter (e.g. one `Braintrust()`) no longer needs to be repeated in every eval. CLI flags and per-eval values still take precedence over the config defaults.
+- 1d65cd6: Reworked the eval authoring API around a single imperative `test(t)` function. An eval now drives the agent and asserts on what it produced in one place — `async test(t) { await t.send(...); t.completed(); t.calledTool(...); t.check(t.reply, includes(...)); t.judge.autoevals.closedQA(...).atLeast(0.6); }` — replacing the separate `run`/`input`, `checks`, `scores`, `expected`, and `thresholds` fields. Assertions carry their own severity: run-level checks and `eve/evals/expect` value assertions (`includes`/`equals`/`matches`/`similarity`) are hard gates by default, while `t.judge.autoevals.*` and `.soft(...)` assertions are tracked and only fail under `--strict`. LLM-as-judge moved under `t.judge.autoevals.*`, and the judge model is now configured via `judge: { model }` (optional) on `defineEvalConfig`/`defineEval` instead of `model`. The `eve/evals/checks` and `eve/evals/scores` entry points are removed in favor of `t`'s built-in vocabulary and `eve/evals/expect`.
+- 1cf3593: Evals gain hard assertions and CI-grade pass/fail. Suites and cases accept a `checks` array (built-ins on `eve/evals/checks`: `Checks.completed`, `Checks.didNotFail`, `Checks.toolCalled` with input/output matchers, `Checks.toolOrder`, `Checks.subagentCalled`, and more); any failed check fails the case and the `eve eval` exit code, while scores stay soft unless `--strict`. Derived facts are now typed records — `toolCalls`/`subagentCalls` carry inputs, outputs, and error state (breaking: previously `string[]`), plus new `inputRequests` and `parked` fields, and `Run.didNotFail` no longer passes runs parked on HITL input. The CLI takes positional suite ids (replacing `--suite`; `--all` is removed) and adds `--tag`, `--case`, `--strict`, and `--list`. Suite `model` is now optional and only required when a model-backed scorer needs it.
+- 9bb6371: Adds the evals interaction API: scripted cases, `task.run`, `EveEvalSession` helpers for HITL responses and file attachments, multi-session event capture, and client-level HITL request results with retry handling for stream registration races.
+- 6fbf1e8: Evals can now verify target requirements and drive non-session surfaces. `eve eval` discovers live target capabilities from `/eve/v1/info`, supports `--mock-models`, `--no-skips`, and `--junit`, exposes `ctx.target.fetch/dispatchSchedule/attachSession`, and adds `requires` to eval and case definitions with a visible `skipped` verdict for unmet requirements.
+- 6fbf1e8: Eval runner polish ahead of the e2e migration: `eve eval --verbose` streams `ctx.log` lines to stdout (and they now land in case artifacts), checks receive the same requirement-scoped target handle as the case run, JUnit failure bodies are trimmed to verdict/checks/scores, and reporter callbacks run off the case-execution hot path.
+- 24c4b85: `eve channels add` now runs the channel picker, scaffold, and deploy boxes through the shared runner. The picker enforces existing-registration conflicts before writes, and the Slackbot question runs before side effects.
+- a648895: Make `agent.ts` optional. Agents without an `agent.ts` now default to
+  `anthropic/claude-sonnet-4.6`; when `agent.ts` is present, `model` remains required.
+- e55bb46: `eve init` now accepts an existing project directory as its target (e.g. `eve init .`): the project must have a `package.json` and use npm, pnpm, or yarn (bun is not supported yet), the `agent/` files must not already exist, and the missing `eve`/`ai`/`zod` dependencies are added without touching anything else the project owns. pnpm projects additionally receive the pnpm workspace policy. In both modes the final handoff now runs the `eve dev` binary through the project's package manager instead of the project's `dev` script, which in an existing app could start unrelated processes.
+
+  The bootstrap install disables npm's and pnpm's minimum-release-age cooldown for that single run — the scaffold pins versions younger than typical windows, so the gate would fail every fresh bootstrap. Later installs follow the project's own configuration.
+
+- 0a8d93b: Add `eve init <name>` for non-interactive agent creation, with optional Web Chat through `--channel-web-nextjs`. The command installs dependencies, initializes Git, and starts the development server, but does not provision or deploy a Vercel project.
+- 20f0fc8: New `eve link` and `eve deploy` commands, also available inside the `eve dev` TUI as `/vercel`, `/channels`, and `/deploy` when the server runs locally. `eve link` picks the Vercel team and project and pulls AI Gateway credentials into `.env.local`; `eve deploy` links when needed and remains successful when its production URL cannot be verified; AI Gateway authentication failures in the TUI now suggest `/vercel`, and the dev server picks up pulled credentials without a restart.
+- 2bc514e: Share the unified select renderer, resolved-prompt rail output, BYOK provider scaffolding, and non-interactive Vercel project linking with scaffold setup flows.
+- e55bb46: `eve init <name>` now scaffolds the new project with the package manager that launched it: `npx` produces an npm-managed project and `yarn dlx` a yarn one, while `pnpm dlx` and direct binary runs keep today's pnpm scaffold. Non-pnpm scaffolds no longer receive `pnpm-workspace.yaml`, `--channel-web-nextjs` requires a pnpm launch, and a bun launch is rejected until bun support lands.
+- ca290a0: Remove the legacy `eve dev --repl` line-based REPL. The terminal UI is the only interactive dev UI; `--no-ui` starts the server headless. The deprecated `--no-repl` alias is no longer accepted — pass `--no-ui` instead.
+- a33fa66: The dev TUI now shows dev-server rebuilds as one status row that updates in place: `tui/setup-panel.ts changed · rebuilding…`, then `· rebuilt`. Only the latest rebuild shows, paths shrink to their last two components, and the transcript no longer stacks a full log line per rebuild.
+- f733fe5: The deterministic mock model adapter (`EVE_MOCK_AUTHORED_MODELS=1`) now synthesizes tool inputs for string properties anchored to quoted spans in the message (e.g. `with label "x"`), honors exact-reply fixture directives (`reply with the exact string … and nothing else`, `include the exact token … verbatim`) found in the current turn's user messages and per-turn client context, discovers static skill advertisements embedded in instruction text (not just standalone announcements), derives skill activation from `load_skill` calls in history so skills are never re-loaded in a loop, and no longer matches `load_skill` by explicit name. This widens what agent smoke evals can assert deterministically without a real model.
+- 9d806d0: Add a `/model` command to the `eve dev` TUI. Run it bare to pick from a list (the running model flagged, a catalog-validated shortlist, and a freeform entry), or `/model <provider/model-id>` to set one directly. The pick rewrites `agent.ts`, and the dev server's HMR watcher applies it on the next prompt.
+- 432deaa: Vercel Connect steps in Slack channel setup run under hard deadlines, so an abandoned browser OAuth can no longer stall `connect create slack` forever.
+- 9061941: The dev TUI's `/model` now opens a configure menu uniting the model and provider setup, and the separate `/vercel` command is removed: "Change model" runs the searchable AI Gateway catalog picker, and "Configure provider" (bold yellow with "Required to enable the agent" until a Vercel link or gateway credential is detected) runs the provider questions `/vercel` used to ask. Each action returns to the menu, which shows each row's current value on its own line — e.g. "AI Gateway (Linked to my-project)" — and keeps the latest outcome visible beneath the options ("✓ Model changed to …"); Esc leaves. Error messages and the startup attention line now point at `/model`, and `/model <provider/model-id>` still applies a model directly.
+- e55bb46: Run setup and generated-project configuration through package-manager strategies for npm, pnpm, Yarn, and Bun. `eve init --channel-web-nextjs` now works with every supported manager instead of requiring a pnpm launch.
+- 22dda94: Rename the eval authoring helper from `defineEvalSuite` to `defineEval` and update the public eval types, reporter hooks, CLI wording, and JSON result shape to use eval terminology consistently.
+- 5a6ac17: `defineEval` is now always a single case, with identity fully derived from the file path — `cases`, `load`, `task`, per-case `id`, and `maxConcurrency` are removed. Declare `input` or `run` (plus `expected`, `checks`, `scores`, `parseOutput`, …) at the top level, organize related evals with directory nesting (`evals/runtime/multi-turn.eval.ts` → `runtime/multi-turn`), and default-export an array of `defineEval(...)` values for dataset fan-out (ids get a zero-padded index suffix, e.g. `weather/0000`). The runner now executes eval files concurrently (default 8, `--max-concurrency`), positional `eve eval` ids match by directory prefix, `--case` is removed, reporters use a run-level lifecycle (`onRunStart`/`onEvalComplete`/`onRunComplete`), check/scorer args expose `evaluation` instead of `case`, and artifacts land under one `.eve/evals/<timestamp>/` directory per run.
+- 455866d: Slack authorization prompts are now visible only to the triggering user. The default `authorization.required` handler sends the sign-in link and device code as an ephemeral message, and falls back to a public link-free status only when it has no user to target. Breaking: an `events["authorization.required"]` override now receives a private-delivery context (`postEphemeral`, `postDirectMessage`, `state`) instead of the full event context, so it can change the message but not the audience. Adds `SlackThread.postDirectMessage`, which DMs a user via `conversations.open` and requires the `im:write` scope.
+- a33fa66: The dev TUI now starts with stdout and stderr logs hidden. Use `/loglevel <all|stderr|none>` to change what the transcript shows. Logs stay buffered either way, so `/loglevel all` brings back earlier output in its original order.
+- 7319957: Add `createStaticSourceChange`, a central registry that lets upstream consumers (CLI, web setup UI) apply targeted edits to authored agent source. Its first operation, `updateModelName`, rewrites the `model` literal in `agent.ts` in place using oxc, preserving surrounding formatting and quote style. It bails with a source location when the value isn't a string literal, like an env reference or an inlined SDK model.
+- 432deaa: Setup multi-select prompts now render checkboxes and end with a bold Submit row. Enter toggles the highlighted option, and only enter on Submit confirms, so a stray enter can no longer skip the checklist. Single-select prompts mark the highlighted row with a `›` arrow instead of a check.
+- 20f0fc8: The dev TUI's bare `/model` now opens the same searchable AI Gateway catalog picker that `eve init` uses — full catalog with the featured shortlist, pre-selected on the model the runtime is serving — instead of a short hand-curated list. `/model <provider/model-id>` still applies directly.
+- 20f0fc8: The dev TUI's `/vercel` now asks which model provider you want before any linking: pick AI Gateway and connect via a project or by pasting your own `AI_GATEWAY_API_KEY` (saved to `.env.local`), or pick another provider to get wiring instructions instead.
+
+### Patch Changes
+
+- 20f0fc8: `eve channels add` and the dev TUI `/channels` command now run `pnpm install` after recording channels, so a running `eve dev` can load newly scaffolded channel modules (for example `@vercel/connect` for Slack) immediately instead of failing to resolve them until the next deploy.
+- a8363e6: Fix `authorization.required` not being emitted when a tool combines `needsApproval` with interactive auth. Approval-resume auth signals are now routed through the authorization park path instead of being replayed to the model as a plain tool result.
+- a8363e6: Authorization-pending tool results no longer expose OAuth URLs, user codes, or hook URLs to the model. Channels still receive full `authorization.required` events.
+- 1acc94e: Apply `build.externalDependencies` to authored module bundling and hosted output tracing, including imports from workspace packages and subagents.
+- 431e5f6: Upgrade Eve's Nitro runtime dependency to `3.0.260610-beta`, picking up the latest Nitro 3 beta fixes and bundler updates.
+- adc6118: Bump the AI SDK to the latest canary releases (`ai@7.0.0-canary.171`, `@ai-sdk/anthropic@4.0.0-canary.65`, `@ai-sdk/google@4.0.0-canary.80`, `@ai-sdk/mcp@2.0.0-canary.62`, `@ai-sdk/openai@4.0.0-canary.72`, `@ai-sdk/otel@1.0.0-canary.117`, `@ai-sdk/provider-utils@5.0.0-canary.47`, `@ai-sdk/react@4.0.0-canary.174`) and the Workflow packages to the latest betas (`@workflow/core@5.0.0-beta.14`, `@workflow/world@5.0.0-beta.8`, `@workflow/world-local@5.0.0-beta.15`).
+- 790865c: Bump the AI SDK to the latest canary releases: `ai@7.0.0-canary.169`, `@ai-sdk/anthropic@4.0.0-canary.64`, `@ai-sdk/mcp@2.0.0-canary.60`, `@ai-sdk/openai@4.0.0-canary.71`, `@ai-sdk/otel@1.0.0-canary.115`, and `@ai-sdk/react@4.0.0-canary.172`.
+- a0ca3bb: Scaffolded projects now pin `@vercel/connect@0.2.2`. The previously pinned 0.1.1 predates the `@vercel/connect/eve` entrypoint, so generated channels and connections failed to build on deploy with `Package subpath './eve' is not defined by "exports"`.
+- 3af511f: Bump `@workflow/core` to `5.0.0-beta.13`.
+- 02d5272: Discover Markdown-authored agent sources case-insensitively, including instructions, schedules, and packaged skills, while keeping module and directory names case-sensitive.
+- c99668f: Correct the public bash tool helper export from `defineBeveTool` to `defineBashTool`.
+- a0ca3bb: Fix interactive post-setup deploys failing with "The `projectSettings` object is required for new projects". The setup deploy now passes `--yes` to `vercel deploy --prod` in interactive runs too, so the Vercel API auto-detects framework settings for projects that eve provisioned through the projects API.
+- 407c6e2: The dev TUI's bottom question panel (setup flows, select/text/acknowledge questions) opens with a stronger `▔` full-width rule and carries a consistent one-space left margin. Gutter glyphs sit at column 1 and text aligns at column 3 instead of touching the terminal edge.
+- 20f0fc8: `eve dev` no longer clears the terminal scrollback when the terminal UI starts, and skips the redundant `server listening at` line in TUI mode — the header already shows the URL. Headless and REPL modes still print it.
+- 8afd7bc: Emit `$eve.cache_write_tokens` as a per-turn observability tag, alongside the existing `$eve.input_tokens`, `$eve.output_tokens`, and `$eve.cache_read_tokens`. The value is read from the AI SDK's `usage.inputTokenDetails.cacheWriteTokens` and accumulated across tool-loop steps within a turn.
+- 790865c: Empty-response recovery now also catches the AI SDK's `NoOutputGeneratedError` rejection (new in `ai@7.0.0-canary.169`): a model stream that closes without producing output is reissued once with the recovery nudge, as before, instead of failing the turn.
+- e1cd134: `eve eval --url` now authenticates to remote targets with the same Vercel OIDC headers as the dev client, including the trusted OIDC IDP token that bypasses Deployment Protection — so evals can run against protected deployments without configuring a `VERCEL_AUTOMATION_BYPASS_SECRET`.
+- 8aeba87: Expand the deterministic authored-model mock so smoke fixtures can exercise ask-question, bash command, loaded-skill, and subagent-token flows without live model variance.
+- 20f0fc8: Make `eve init` install the current Eve release through the npm `latest` dist-tag, including when scaffolding Web Chat.
+- 2aec0cf: Raise the derived `$eve.title` cap (`EVE_SESSION_TITLE_MAX_CHARS`) from 80 to 125 code points so Agent Runs titles keep more of the original prompt.
+- 9a35ea7: Fix `eve init` to launch pnpm-managed projects through `pnpm exec` and generate Web Chat configuration that matches the installed Eve Next.js API.
+- 5923ebc: fix(eve): prevent provider-executed tool calls from being emitted as client actions
+- ca290a0: Ctrl+C during a streaming `eve dev` response now aborts the underlying HTTP stream and releases its iterator instead of leaving the connection dangling.
+- 74cd2e5: Store local just-bash sandbox templates and sessions as normal on-disk directories instead of JSON filesystem snapshots, and prune stale `eve dev` runtime snapshots and local sandbox templates in the background.
+- 8507473: Update Workflow SDK beta packages and isolate Eve's workflow queue traffic behind the SDK's `eve` queue namespace so co-deployed apps can use their own Workflow output without topic collisions.
+- beddb37: Prepare Eve for open source: remove the internal `eve/dev-tui-test` subpath export and remove internal doc references from runtime error messages.
+- 3f71a5c: Add `repository`, `homepage`, and `bugs` metadata to the published package, and ship the Apache-2.0 NOTICE file in npm tarballs.
+- e1cd134: Workflow runs only route to `deploymentId: "latest"` on Vercel production. Preview and CLI deployments carry no git branch reference, so latest resolution failed with HTTP 400 ("Source deployment has no git branch") and every turn errored; they now pin workflow runs to their own immutable deployment.
+- ca290a0: `eve dev` no longer clears the terminal (including native scrollback) when a source edit refreshes the agent header — the refreshed header is committed beneath the existing transcript instead.
+- 8d087ff: Run the development server and terminal UI when `eve` is invoked without a command. Use `eve info` explicitly to inspect the resolved application.
+- abaca02: Recover from empty model responses instead of silently ending the turn. When a model call completes with no content (finishReason 'other', e.g. an AI Gateway 200 with an empty stream), the harness reissues the call once within the same step. If the reissue also comes back empty, conversations end with a recoverable, channel-visible failure and task runs end with a failed task result rather than nothing.
+- 20f0fc8: Replace removed `create-eve` and `eve setup` guidance with the current `eve init` and `eve link` commands.
+- 5b69978: Fix `eve dev` source snapshots so monorepo-relative config paths and local workspace package links keep resolving from immutable dev-runtime snapshots.
+- 20f0fc8: Restore the `eve link` and `eve deploy` command registrations after the `eve init` migration.
+- 0a8d93b: Scaffolded projects now carry a `packageExtensions` entry in `pnpm-workspace.yaml` that restores eve's missing `oxc-parser` dependency, so fresh agents boot against the published `eve` 0.6.0-beta.13/14 builds that import it without declaring it. The entry lives in `pnpm-workspace.yaml` because pnpm 11 reads settings only from there, not from the `package.json` `pnpm` field. Remove it from the template once the scaffolded `eve` range only resolves to versions that declare `oxc-parser` themselves.
+- f010437: Allow Next.js development integrations up to three minutes to start Eve by default, and support overriding the wait with the `withEve` `devServerTimeoutMs` option.
+- 380693d: Token eviction on a rejected bearer now cascades to the authorization strategy's own cache, not just Eve's per-step cache. `AuthorizationDefinition` gains an optional `evict(opts)` hook, and the shared `evictScopedToken` path (used by both authored tools and MCP connections) invokes it after dropping the per-step entry. This lets `@vercel/connect`-backed connections purge their in-process token cache so a revoked-but-unexpired grant is genuinely re-fetched instead of re-read from a lower cache layer.
+- ca1ca18: Per-tool authorization now evicts the rejected bearer from its per-step cache before re-running the consent flow, so a tool that re-reports `Required` (e.g. after a downstream `401`/`requireAuth()`) re-resolves a fresh token instead of re-reading the rejected one. This mirrors the MCP client's behavior. Documented the recommended pattern: map a provider `401` inside a tool's `execute` to `ctx.requireAuth()` so Eve re-challenges.
+- 5e87d70: Per-tool interactive authorization never leaks a raw `Required` error into
+  the model. When a tool's `auth` strategy is interactive but no callback URL
+  can be minted for the run, the tool now fails with a classified
+  `ConnectionAuthorizationFailedError` (`reason: "authorization_callback_unavailable"`,
+  non-retryable) instead of rethrowing the raw authorization error — which the
+  model would otherwise surface as a sign-in URL and loop on. Non-interactive
+  strategies still rethrow their original error, since they have no consent
+  flow to park on.
+- 20f0fc8: Keep superseded Vercel project-name warnings inside the `/vercel` panel when linking ultimately succeeds.
+- 20f0fc8: Scaffolding channels, connections, and new projects from an unstamped dev build (for example under `pnpm dev`'s watch emit) now resolves dependency versions from the live workspace catalog instead of failing with "unstamped version token". Published builds are unchanged: they are stamped at build time, and an unstamped token outside a dev tree still fails loudly.
+
+## 0.6.0-beta.20
+
+### Minor Changes
+
+- 407c6e2: The dev TUI startup header is now a single `▲ eve <agent name>` brand line plus one rotating tip ("Use /channels to add more ways to reach your agent.", "Use /deploy to see your agent go live.", "Type /help to see every command."). The config rows (Model, Instructions, Tools, Skills, Subagents, Server) and the key-hints line are gone. The model moved to the status line, the empty input row shows a dim `Type to chat · / for commands` placeholder, discovery error/warning counts still render, and `eve info` keeps the full configuration detail.
+- 407c6e2: The dev TUI footer now ends with a persistent status line: the agent's model, the session's token flow (`⁕ ↑ 394.4K ↓ 4.3K`, input up, output down, `⁕ ↑ 0 ↓ 0` at startup), the linked Vercel project and team, and a `deploy pending` marker that appears when `/channels` adds a channel and clears when `/deploy` ships it. The Vercel segment stays hidden until the directory is linked. Token usage moved here from the working row, and `--context-size` appends a context-fill percentage.
+- 1d65cd6: Reworked the eval authoring API around a single imperative `test(t)` function. An eval now drives the agent and asserts on what it produced in one place — `async test(t) { await t.send(...); t.completed(); t.calledTool(...); t.check(t.reply, includes(...)); t.judge.autoevals.closedQA(...).atLeast(0.6); }` — replacing the separate `run`/`input`, `checks`, `scores`, `expected`, and `thresholds` fields. Assertions carry their own severity: run-level checks and `eve/evals/expect` value assertions (`includes`/`equals`/`matches`/`similarity`) are hard gates by default, while `t.judge.autoevals.*` and `.soft(...)` assertions are tracked and only fail under `--strict`. LLM-as-judge moved under `t.judge.autoevals.*`, and the judge model is now configured via `judge: { model }` (optional) on `defineEvalConfig`/`defineEval` instead of `model`. The `eve/evals/checks` and `eve/evals/scores` entry points are removed in favor of `t`'s built-in vocabulary and `eve/evals/expect`.
+- a33fa66: The dev TUI now shows dev-server rebuilds as one status row that updates in place: `tui/setup-panel.ts changed · rebuilding…`, then `· rebuilt`. Only the latest rebuild shows, paths shrink to their last two components, and the transcript no longer stacks a full log line per rebuild.
+- a33fa66: The dev TUI now starts with stdout and stderr logs hidden. Use `/loglevel <all|stderr|none>` to change what the transcript shows. Logs stay buffered either way, so `/loglevel all` brings back earlier output in its original order.
+
+### Patch Changes
+
+- 407c6e2: The dev TUI's bottom question panel (setup flows, select/text/acknowledge questions) opens with a stronger `▔` full-width rule and carries a consistent one-space left margin. Gutter glyphs sit at column 1 and text aligns at column 3 instead of touching the terminal edge.
+- e1cd134: `eve eval --url` now authenticates to remote targets with the same Vercel OIDC headers as the dev client, including the trusted OIDC IDP token that bypasses Deployment Protection — so evals can run against protected deployments without configuring a `VERCEL_AUTOMATION_BYPASS_SECRET`.
+- 9a35ea7: Fix `eve init` to launch pnpm-managed projects through `pnpm exec` and generate Web Chat configuration that matches the installed Eve Next.js API.
+- e1cd134: Workflow runs only route to `deploymentId: "latest"` on Vercel production. Preview and CLI deployments carry no git branch reference, so latest resolution failed with HTTP 400 ("Source deployment has no git branch") and every turn errored; they now pin workflow runs to their own immutable deployment.
+
+## 0.6.0-beta.19
+
+### Minor Changes
+
+- 9061941: The dev TUI's `/model` now opens a configure menu uniting the model and provider setup, and the separate `/vercel` command is removed: "Change model" runs the searchable AI Gateway catalog picker, and "Configure provider" (bold yellow with "Required to enable the agent" until a Vercel link or gateway credential is detected) runs the provider questions `/vercel` used to ask. Each action returns to the menu, which shows each row's current value on its own line — e.g. "AI Gateway (Linked to my-project)" — and keeps the latest outcome visible beneath the options ("✓ Model changed to …"); Esc leaves. Error messages and the startup attention line now point at `/model`, and `/model <provider/model-id>` still applies a model directly.
+
+## 0.6.0-beta.18
+
+### Minor Changes
+
+- aa5cfdc: Connection and tool `auth` definitions accept an optional `displayName`, and authorization challenges carry it to channels — sign-in buttons now show e.g. "Sign in with Salesforce" instead of a title-cased file name. The value is presentation-only (scope, token cache keys, and callback URLs stay keyed by the path-derived name), a definition-level `displayName` wins over one the strategy stamps on the challenge, and channels keep title-casing the name when neither is set.
+- e55bb46: `eve init` now accepts an existing project directory as its target (e.g. `eve init .`): the project must have a `package.json` and use npm, pnpm, or yarn (bun is not supported yet), the `agent/` files must not already exist, and the missing `eve`/`ai`/`zod` dependencies are added without touching anything else the project owns. pnpm projects additionally receive the pnpm workspace policy. In both modes the final handoff now runs the `eve dev` binary through the project's package manager instead of the project's `dev` script, which in an existing app could start unrelated processes.
+
+  The bootstrap install disables npm's and pnpm's minimum-release-age cooldown for that single run — the scaffold pins versions younger than typical windows, so the gate would fail every fresh bootstrap. Later installs follow the project's own configuration.
+
+- e55bb46: `eve init <name>` now scaffolds the new project with the package manager that launched it: `npx` produces an npm-managed project and `yarn dlx` a yarn one, while `pnpm dlx` and direct binary runs keep today's pnpm scaffold. Non-pnpm scaffolds no longer receive `pnpm-workspace.yaml`, `--channel-web-nextjs` requires a pnpm launch, and a bun launch is rejected until bun support lands.
+- ca290a0: Remove the legacy `eve dev --repl` line-based REPL. The terminal UI is the only interactive dev UI; `--no-ui` starts the server headless. The deprecated `--no-repl` alias is no longer accepted — pass `--no-ui` instead.
+- e55bb46: Run setup and generated-project configuration through package-manager strategies for npm, pnpm, Yarn, and Bun. `eve init --channel-web-nextjs` now works with every supported manager instead of requiring a pnpm launch.
+
+### Patch Changes
+
+- 1acc94e: Apply `build.externalDependencies` to authored module bundling and hosted output tracing, including imports from workspace packages and subagents.
+- 8afd7bc: Emit `$eve.cache_write_tokens` as a per-turn observability tag, alongside the existing `$eve.input_tokens`, `$eve.output_tokens`, and `$eve.cache_read_tokens`. The value is read from the AI SDK's `usage.inputTokenDetails.cacheWriteTokens` and accumulated across tool-loop steps within a turn.
+- ca290a0: Ctrl+C during a streaming `eve dev` response now aborts the underlying HTTP stream and releases its iterator instead of leaving the connection dangling.
+- ca290a0: `eve dev` no longer clears the terminal (including native scrollback) when a source edit refreshes the agent header — the refreshed header is committed beneath the existing transcript instead.
+
+## 0.6.0-beta.17
+
+### Minor Changes
+
+- 20f0fc8: The dev TUI's `/channels` is now an action list: pick an unregistered channel to run its add flow, return to the repainted list (added channels show locked), and leave with the Done row or Esc. Web Chat (the Next.js app) is now addable in projects that already have the scaffolded `agent/channels/eve.ts` session channel — the row locks only when the project actually depends on `next`, and the REPL channel shows as its own locked row.
+- 20f0fc8: The dev TUI now discovers its slash commands: typing `/` opens a suggestion list above the prompt with each command's argument hint and description — `↑`/`↓` move the highlight, `Tab` completes, `Enter` runs, `Esc` dismisses — and a new `/help` command prints the full table. Unknown `/text` is still sent to the agent as a normal message.
+- 20f0fc8: New `eve link` and `eve deploy` commands, also available inside the `eve dev` TUI as `/vercel`, `/channels`, and `/deploy` when the server runs locally. `eve link` picks the Vercel team and project and pulls AI Gateway credentials into `.env.local`; `eve deploy` links when needed and remains successful when its production URL cannot be verified; AI Gateway authentication failures in the TUI now suggest `/vercel`, and the dev server picks up pulled credentials without a restart.
+- 20f0fc8: The dev TUI's bare `/model` now opens the same searchable AI Gateway catalog picker that `eve init` uses — full catalog with the featured shortlist, pre-selected on the model the runtime is serving — instead of a short hand-curated list. `/model <provider/model-id>` still applies directly.
+- 20f0fc8: The dev TUI's `/vercel` now asks which model provider you want before any linking: pick AI Gateway and connect via a project or by pasting your own `AI_GATEWAY_API_KEY` (saved to `.env.local`), or pick another provider to get wiring instructions instead.
+
+### Patch Changes
+
+- 20f0fc8: `eve channels add` and the dev TUI `/channels` command now run `pnpm install` after recording channels, so a running `eve dev` can load newly scaffolded channel modules (for example `@vercel/connect` for Slack) immediately instead of failing to resolve them until the next deploy.
+- 20f0fc8: `eve dev` no longer clears the terminal scrollback when the terminal UI starts, and skips the redundant `server listening at` line in TUI mode — the header already shows the URL. Headless and REPL modes still print it.
+- 20f0fc8: Make `eve init` install the newest Eve prerelease through the npm `beta` dist-tag, including when scaffolding Web Chat.
+- 20f0fc8: Replace removed `create-eve` and `eve setup` guidance with the current `eve init` and `eve link` commands.
+- 20f0fc8: Restore the `eve link` and `eve deploy` command registrations after the `eve init` migration.
+- 20f0fc8: Keep superseded Vercel project-name warnings inside the `/vercel` panel when linking ultimately succeeds.
+- 20f0fc8: Scaffolding channels, connections, and new projects from an unstamped dev build (for example under `pnpm dev`'s watch emit) now resolves dependency versions from the live workspace catalog instead of failing with "unstamped version token". Published builds are unchanged: they are stamped at build time, and an unstamped token outside a dev tree still fails loudly.
+
+## 0.6.0-beta.16
+
+### Minor Changes
+
+- 5a6ac17: Add a required `evals/evals.config.ts` (authored with `defineEvalConfig`) that declares run-wide eval defaults: a mandatory scorer `model`, plus optional run-level `reporters`, `maxConcurrency`, and `timeoutMs`. Model-backed scorers now fall back to the config `model`, so `model` is optional on `defineEval` and a shared reporter (e.g. one `Braintrust()`) no longer needs to be repeated in every eval. CLI flags and per-eval values still take precedence over the config defaults.
+- 5a6ac17: `defineEval` is now always a single case, with identity fully derived from the file path — `cases`, `load`, `task`, per-case `id`, and `maxConcurrency` are removed. Declare `input` or `run` (plus `expected`, `checks`, `scores`, `parseOutput`, …) at the top level, organize related evals with directory nesting (`evals/runtime/multi-turn.eval.ts` → `runtime/multi-turn`), and default-export an array of `defineEval(...)` values for dataset fan-out (ids get a zero-padded index suffix, e.g. `weather/0000`). The runner now executes eval files concurrently (default 8, `--max-concurrency`), positional `eve eval` ids match by directory prefix, `--case` is removed, reporters use a run-level lifecycle (`onRunStart`/`onEvalComplete`/`onRunComplete`), check/scorer args expose `evaluation` instead of `case`, and artifacts land under one `.eve/evals/<timestamp>/` directory per run.
+
+### Patch Changes
+
+- a8363e6: Fix `authorization.required` not being emitted when a tool combines `needsApproval` with interactive auth. Approval-resume auth signals are now routed through the authorization park path instead of being replayed to the model as a plain tool result.
+- a8363e6: Authorization-pending tool results no longer expose OAuth URLs, user codes, or hook URLs to the model. Channels still receive full `authorization.required` events.
+
+## 0.6.0-beta.15
+
+### Minor Changes
+
+- 0a8d93b: Add `eve init <name>` for non-interactive agent creation, with optional Web Chat through `--web`. The command installs dependencies, initializes Git, and starts the development server, but does not provision or deploy a Vercel project.
+- f733fe5: The deterministic mock model adapter (`EVE_MOCK_AUTHORED_MODELS=1`) now synthesizes tool inputs for string properties anchored to quoted spans in the message (e.g. `with label "x"`), honors exact-reply fixture directives (`reply with the exact string … and nothing else`, `include the exact token … verbatim`) found in the current turn's user messages and per-turn client context, discovers static skill advertisements embedded in instruction text (not just standalone announcements), derives skill activation from `load_skill` calls in history so skills are never re-loaded in a loop, and no longer matches `load_skill` by explicit name. This widens what agent smoke evals can assert deterministically without a real model.
+
+### Patch Changes
+
+- 0a8d93b: Scaffolded projects now carry a `packageExtensions` entry in `pnpm-workspace.yaml` that restores eve's missing `oxc-parser` dependency, so fresh agents boot against the published `eve` 0.6.0-beta.13/14 builds that import it without declaring it. The entry lives in `pnpm-workspace.yaml` because pnpm 11 reads settings only from there, not from the `package.json` `pnpm` field. Remove it from the template once the scaffolded `eve` range only resolves to versions that declare `oxc-parser` themselves.
+
+## 0.6.0-beta.14
+
+### Minor Changes
+
+- 9d806d0: Add a `/model` command to the `eve dev` TUI. Run it bare to pick from a list (the running model flagged, a catalog-validated shortlist, and a freeform entry), or `/model <provider/model-id>` to set one directly. The pick rewrites `agent.ts`, and the dev server's HMR watcher applies it on the next prompt.
+
+### Patch Changes
+
+- c99668f: Correct the public bash tool helper export from `defineBeveTool` to `defineBashTool`.
+
+## 0.6.0-beta.13
+
+### Minor Changes
+
+- fe98d0b: Compaction is now owned by the framework instead of individual tools. The per-tool `onCompact` hook (and the `CompactionInput` / `CompactionHookResult` types) is removed: the framework automatically preserves its own tool state across compaction — it resets read-before-write tracking and re-injects the active todo list. `defineReadFileTool` no longer accepts an `onCompact` option, and the agent-info response drops the now-meaningless `hasCompactionHook` field.
+- 6fbf1e8: Evals can now verify target requirements and drive non-session surfaces. `eve eval` discovers live target capabilities from `/eve/v1/info`, supports `--mock-models`, `--no-skips`, and `--junit`, exposes `ctx.target.fetch/dispatchSchedule/attachSession`, and adds `requires` to eval and case definitions with a visible `skipped` verdict for unmet requirements.
+- 6fbf1e8: Eval runner polish ahead of the e2e migration: `eve eval --verbose` streams `ctx.log` lines to stdout (and they now land in case artifacts), checks receive the same requirement-scoped target handle as the case run, JUnit failure bodies are trimmed to verdict/checks/scores, and reporter callbacks run off the case-execution hot path.
+- 4408988: Onboarding now asks where the agent runs last, after the agent itself is described: name, model, channels, connections, then deployment (Vercel, or locally for now). Picking Slack or a Connect-backed connection resolves the deployment question to Vercel automatically with a note instead of disabling those rows up front, and scaffolding into a directory that is already linked to a Vercel project (with a logged-in CLI) skips the deployment, team, and project questions entirely — the AI Gateway authenticates through your Vercel login. The "own provider key" path now derives the scaffolded `byok` block from the model you picked (e.g. `openai/…` → `OPENAI_API_KEY`) instead of pinning the Anthropic default, and on headless `--skip-vercel` runs `--model` is now honored and validated against the AI Gateway catalog instead of being silently replaced by that default.
+- 7319957: Add `createStaticSourceChange`, a central registry that lets upstream consumers (CLI, web setup UI) apply targeted edits to authored agent source. Its first operation, `updateModelName`, rewrites the `model` literal in `agent.ts` in place using oxc, preserving surrounding formatting and quote style. It bails with a source location when the value isn't a string literal, like an env reference or an inlined SDK model.
+
+### Patch Changes
+
+- 431e5f6: Upgrade Eve's Nitro runtime dependency to `3.0.260610-beta`, picking up the latest Nitro 3 beta fixes and bundler updates.
+- adc6118: Bump the AI SDK to the latest canary releases (`ai@7.0.0-canary.171`, `@ai-sdk/anthropic@4.0.0-canary.65`, `@ai-sdk/google@4.0.0-canary.80`, `@ai-sdk/mcp@2.0.0-canary.62`, `@ai-sdk/openai@4.0.0-canary.72`, `@ai-sdk/otel@1.0.0-canary.117`, `@ai-sdk/provider-utils@5.0.0-canary.47`, `@ai-sdk/react@4.0.0-canary.174`) and the Workflow packages to the latest betas (`@workflow/core@5.0.0-beta.14`, `@workflow/world@5.0.0-beta.8`, `@workflow/world-local@5.0.0-beta.15`).
+- 790865c: Bump the AI SDK to the latest canary releases: `ai@7.0.0-canary.169`, `@ai-sdk/anthropic@4.0.0-canary.64`, `@ai-sdk/mcp@2.0.0-canary.60`, `@ai-sdk/openai@4.0.0-canary.71`, `@ai-sdk/otel@1.0.0-canary.115`, and `@ai-sdk/react@4.0.0-canary.172`.
+- 790865c: Empty-response recovery now also catches the AI SDK's `NoOutputGeneratedError` rejection (new in `ai@7.0.0-canary.169`): a model stream that closes without producing output is reissued once with the recovery nudge, as before, instead of failing the turn.
+- 8507473: Update Workflow SDK beta packages and isolate Eve's workflow queue traffic behind the SDK's `eve` queue namespace so co-deployed apps can use their own Workflow output without topic collisions.
+- 4408988: The one-shot "Next steps" note now lists the exact commands in execution order — `cd`, `pnpm install`, `vercel link` (or set `AI_GATEWAY_API_KEY` in `.env.local` manually), `eve dev` — with the commands in bold.
+- f010437: Allow Next.js development integrations up to three minutes to start Eve by default, and support overriding the wait with the `withEve` `devServerTimeoutMs` option.
+
+## 0.6.0-beta.12
+
+### Minor Changes
+
+- 4c1dd92: The connections picker moved into the onboarding interview: it now asks right after channel selection, before any files are written. The selected connections are still scaffolded and provisioned after the project link, exactly as before.
+- 432deaa: The setup model picker now opens on a curated shortlist (Claude Sonnet 4.6 as the default, Claude Opus 4.8, GPT-5.5, and Gemini 3.5) and surfaces the rest of the AI Gateway catalog through scrolling or search. The search filter now accepts spaces. Vercel Connect steps in Slack channel setup run under hard deadlines, so an abandoned browser OAuth can no longer stall `connect create slack` forever.
+- 4c1dd92: Onboarding deploys to Vercel only when the Slack channel was added — its connector needs a public production URL. A web-only onboarding run links the project but skips the deploy; Web Chat runs locally through `eve dev`. Adding channels to an existing project (`eve setup` in-project, `eve channels add`) still deploys for any channel.
+- 4c1dd92: The onboarding flow now asks how much to set up after the agent name: "Complete setup" (the existing full flow, recommended) or "One-shot", which scaffolds the base template with the default model and skips provisioning, channels, deploy, and chat, ending with a next-steps note. `eve setup` in a directory that is not an Eve project now asks for the agent name and where the agent should live instead of force-scaffolding in place over the existing repo.
+- 432deaa: A failed Slackbot provision no longer aborts onboarding or `eve setup`. The run warns, skips Slack, and keeps scaffolding and deploying, and you can add Slack later with `eve channels add slack`. The `eve channels add slack` command itself still fails hard, since Slack is its whole purpose.
+- 432deaa: Setup multi-select prompts now render checkboxes and end with a bold Submit row. Enter toggles the highlighted option, and only enter on Submit confirms, so a stray enter can no longer skip the checklist. Single-select prompts mark the highlighted row with a `›` arrow instead of a check.
+
+### Patch Changes
+
+- 432deaa: Refreshed onboarding copy: the deploy question now asks where to deploy (Vercel vs. elsewhere), the team and project pickers use shorter titles, and the intro banners spell the product as 𝐞𝐯𝐞.
+
+## 0.6.0-beta.11
+
+### Minor Changes
+
+- 9bb6371: Adds the evals interaction API: scripted cases, `task.run`, `EveEvalSession` helpers for HITL responses and file attachments, multi-session event capture, and client-level HITL request results with retry handling for stream registration races.
+- 22dda94: Rename the eval authoring helper from `defineEvalSuite` to `defineEval` and update the public eval types, reporter hooks, CLI wording, and JSON result shape to use eval terminology consistently.
+
+### Patch Changes
+
+- a0ca3bb: Scaffolded projects now pin `@vercel/connect@0.2.2`. The previously pinned 0.1.1 predates the `@vercel/connect/eve` entrypoint, so generated channels and connections failed to build on deploy with `Package subpath './eve' is not defined by "exports"`.
+- a0ca3bb: Fix interactive post-setup deploys failing with "The `projectSettings` object is required for new projects". The setup deploy now passes `--yes` to `vercel deploy --prod` in interactive runs too, so the Vercel API auto-detects framework settings for projects that eve provisioned through the projects API.
+- 380693d: Token eviction on a rejected bearer now cascades to the authorization strategy's own cache, not just Eve's per-step cache. `AuthorizationDefinition` gains an optional `evict(opts)` hook, and the shared `evictScopedToken` path (used by both authored tools and MCP connections) invokes it after dropping the per-step entry. This lets `@vercel/connect`-backed connections purge their in-process token cache so a revoked-but-unexpired grant is genuinely re-fetched instead of re-read from a lower cache layer.
+
+## 0.6.0-beta.10
+
+### Minor Changes
+
+- 24c4b85: Dissolves the scaffold engine's step flows into the setup boxes. `eve channels add` now runs the same channel boxes as onboarding through the shared runner: the picker enforces the existing-registration conflict checks directly, the slackbot question moves ahead of any file write, and prompts can never interleave with side effects.
+
+  Removes the `eve/setup/scaffold/cli` and `eve/setup/scaffold/primitives` subpath exports. The terminal prompt adapters and process primitives now live in the setup island (`src/setup/cli/`, `src/setup/primitives/`), and the names `create-eve` consumes (`createPromptCommandOutput`, `detectDeployment`, `runPnpmInstall`, `spawnPnpm`, `runVercel`) are exported from `eve/setup` instead. The `eve/setup/scaffold` barrel is unchanged.
+
+- 24c4b85: Route setup questions through a single ask channel. Setup boxes now ask keyed question values through one injected `Asker` whose decorator stack (interactive base, headless base, pre-supplied answers, detected/recommended policies) decides how each question resolves; the select-model, select-chat, select-channels, and resolve-target boxes migrate onto the unified single-gather contract. The channel gains a paired `askMany` method for multi-select questions (whose `T[]` answers cannot ride `ask<T>`), with option-level `locked`/`disabled` row semantics enforced both by the interactive picker and by pre-supplied answer validation. A headless run that reaches a required question without an answer now fails with a structured `InteractionRequired` error naming the question key instead of a `HeadlessPromptError`.
+
+  The Vercel provisioning, channel-scaffold, connection, and deploy boxes now flow through that same channel: the provisioning deploy tree (deploy gate, new/link project, credential wiring, and the AI Gateway key as a sensitive text question rendered through the masked prompt), the slackbot create confirm, and the connection picker plus its custom MCP/OpenAPI sub-prompts are all keyed questions on the asker; the deploy box collapses its two constant gather faces into one. A headless run missing any required provisioning decision now refuses with `InteractionRequired` rather than `HeadlessPromptError`.
+
+  The remaining no-prompt boxes (preflight, scaffold, detect-ai-gateway, link-project, apply-ai-gateway-credential) collapse their dual gather faces into one. Boxes whose two faces differed by mode (preflight derives a model to validate only on a headless `--model`; scaffold skips the write on a headless re-run over an existing Eve project) take a `headless` factory option fixed at the composition site, mirroring the deploy box. With every box migrated, the `SetupBox` contract is now unified-gather only: the `SetupBox` type is the single unified interface and the `LegacySetupBox`/`UnifiedSetupBox` union, the `isUnifiedSetupBox` guard, and the `InteractiveOutcome` type are gone. Both runners call `box.gather` directly; cancel is signaled solely by a thrown `WizardCancelledError`.
+
+- 455866d: Slack authorization prompts are now visible only to the triggering user. The default `authorization.required` handler sends the sign-in link and device code as an ephemeral message, and falls back to a public link-free status only when it has no user to target. Breaking: an `events["authorization.required"]` override now receives a private-delivery context (`postEphemeral`, `postDirectMessage`, `state`) instead of the full event context, so it can change the message but not the audience. Adds `SlackThread.postDirectMessage`, which DMs a user via `conversations.open` and requires the `im:write` scope.
+
+## 0.6.0-beta.9
+
+### Minor Changes
+
+- 1cf3593: Evals gain hard assertions and CI-grade pass/fail. Suites and cases accept a `checks` array (built-ins on `eve/evals/checks`: `Checks.completed`, `Checks.didNotFail`, `Checks.toolCalled` with input/output matchers, `Checks.toolOrder`, `Checks.subagentCalled`, and more); any failed check fails the case and the `eve eval` exit code, while scores stay soft unless `--strict`. Derived facts are now typed records — `toolCalls`/`subagentCalls` carry inputs, outputs, and error state (breaking: previously `string[]`), plus new `inputRequests` and `parked` fields, and `Run.didNotFail` no longer passes runs parked on HITL input. The CLI takes positional suite ids (replacing `--suite`; `--all` is removed) and adds `--tag`, `--case`, `--strict`, and `--list`. Suite `model` is now optional and only required when a model-backed scorer needs it.
+- a648895: Make `agent.ts` optional. Agents without an `agent.ts` now default to
+  `anthropic/claude-sonnet-4.6`; when `agent.ts` is present, `model` remains required.
+
+### Patch Changes
+
+- 74cd2e5: Store local just-bash sandbox templates and sessions as normal on-disk directories instead of JSON filesystem snapshots, and prune stale `eve dev` runtime snapshots and local sandbox templates in the background.
+
 ## 0.6.0-beta.8
 
 ### Patch Changes

package/README.md

@@ -45,14 +45,14 @@ Every authored directory has a typed helper. Import each from the matching subpa
 | ------------------------------------------------------------------------------------------------------------------- | ------------------------------------- | ------------------------------------------------ |
 | `defineAgent(...)`                                                                                                  | `eve`                                 | `agent.ts`, `subagents/<id>/agent.ts`            |
 | `defineInstructions(...)`                                                                                           | `eve/instructions`                    | `instructions.ts` (or `instructions.md`)         |
-| `defineTool(...)`, `defineBeveTool(...)`, `defineReadFileTool(...)`, `defineWriteFileTool(...)`, `disableTool(...)` | `eve/tools`                           | `tools/<name>.ts`                                |
+| `defineTool(...)`, `defineBashTool(...)`, `defineReadFileTool(...)`, `defineWriteFileTool(...)`, `disableTool(...)` | `eve/tools`                           | `tools/<name>.ts`                                |
 | `defineSkill(...)`, `getSkill(...)`                                                                                 | `eve/skills`                          | `skills/<name>.ts` (or `skills/<name>.md`)       |
 | `defineHook(...)`                                                                                                   | `eve/hooks`                           | `hooks/<slug>.ts`                                |
 | `defineChannel(...)`, `POST`, `GET`                                                                                 | `eve/channels`                        | `channels/<name>.ts`                             |
 | `eveChannel(...)`, `slackChannel(...)`, `vercelOidc(...)`                                                           | `eve/channels/eve`, `/slack`, `/auth` | reused from `channels/<name>.ts`                 |
 | `defineSandbox(...)`                                                                                                | `eve/sandbox`                         | `sandbox.ts` (or `sandbox/sandbox.ts`)           |
 | `defineSchedule(...)`                                                                                               | `eve/schedules`                       | `schedules/<name>.ts` (or `schedules/<name>.md`) |
-| `defineEvalSuite(...)`                                                                                              | `eve/evals`                           | `evals/<name>.eval.ts`                           |
+| `defineEval(...)`, `defineEvalConfig(...)`                                                                          | `eve/evals`                           | `evals/<name>.eval.ts`, `evals/evals.config.ts`  |
 
 Runtime accessors live on the subpath that owns the concern:
 
@@ -105,15 +105,18 @@ export default defineAgent({
 ## Quick Start
 
 ```bash
-pnpm create eve@beta
-cd my-agent
-pnpm dev
+npx eve@latest init my-agent
 ```
 
-The wizard scaffolds the project, picks a model, and (for the REPL channel) installs dependencies and starts the dev server for you. To scaffold into the current empty directory, run `pnpm create eve@beta .`.
+`eve init` writes a new agent with Eve's default model. Pass `--channel-web-nextjs` to add the
+Web Chat application. It installs dependencies, initializes Git, and starts the
+development server. Targeting an existing project directory (`eve init .`) adds
+the agent files and missing dependencies instead. It does not create a Vercel
+project or deploy the agent.
 
 CLI commands:
 
+- `eve init <name>` — create a new agent
 - `eve info` — discovery results and compiled artifacts
 - `eve build` — compile `.eve/` and build the host output
 - `eve start` — serve the built `.output/` app

package/dist/docs/public/README.md

@@ -30,7 +30,7 @@ Read in this order:
 14. [TypeScript Client](./client/overview.mdx)
 15. [Subagents](./subagents.mdx)
 16. [Schedules](./schedules.mdx)
-17. [Evals](./advanced/evals.md)
+17. [Evals](./evals/overview.mdx)
 18. [Auth And Route Protection](./advanced/auth-and-route-protection.md)
 19. [Vercel Deployment](./advanced/deployment.md)
 20. [CLI, Build, And Debugging](./reference/cli.md)

package/dist/docs/public/advanced/auth-and-route-protection.md

@@ -184,7 +184,7 @@ export default defineChannel({
 
 ## Replace `placeholderAuth` before production
 
-`pnpm create eve@beta` sometimes scaffolds `agent/channels/eve.ts` with a `placeholderAuth()` guardrail:
+`eve init` scaffolds `agent/channels/eve.ts` with a `placeholderAuth()` guardrail:
 
 ```ts
 import { eveChannel } from "eve/channels/eve";
@@ -263,6 +263,8 @@ Declaring `auth` adds two accessors to the tool's `ctx`:
 
 Throw `ConnectionAuthorizationRequiredError` anywhere in `execute` (directly, via `requireAuth()`, or implicitly from `getToken()`) and you trigger the consent flow, keyed by the tool's name. Calling either accessor on a tool that does not declare `auth` throws.
 
+By default the sign-in affordance title-cases the tool's path-derived name — a tool file named `sfdc_lookup.ts` renders "Sign in with Sfdc_lookup". Set `displayName` on the `auth` definition to control what users see instead: `auth: { ...connect("sfdc"), displayName: "Salesforce" }`. It is presentation-only; the tool's name still keys the authorization scope, token cache, and callback URL, and a definition-level `displayName` wins over one the strategy stamps on the challenge.
+
 ## What to read next
 
 - [Security model](./security-model): trust boundaries and the pre-production checklist

package/dist/docs/public/advanced/default-harness.md

@@ -18,7 +18,7 @@ export default defineAgent({
 });
 ```
 
-Compaction is also a hook point for tools. When the harness compacts history, it calls each tool's `onCompact(input, ctx)` in registration order, so a tool can re-inject the facts it needs to survive the summary: appending a short message, or patching session state. See [`onCompact`](./dynamic-capabilities) for the hook contract.
+Compaction also preserves the framework's own tool state automatically. When the harness compacts history, it resets read-before-write tracking (so a write afterward re-reads the file whose read evidence was summarized away) and re-injects the active todo list, so the model keeps its task list across the summary. There is no per-tool hook to configure.
 
 ## Built-in tools
 
@@ -89,6 +89,6 @@ With it on, the model can orchestrate the agent's own subagents from model-autho
 
 ## What to read next
 
-- [Tools](../tools): define your own tools, gate them on approval, and shape their output (`toModelOutput` / `onCompact`)
+- [Tools](../tools): define your own tools, gate them on approval, and shape their output with `toModelOutput`
 - [Dynamic capabilities](./dynamic-capabilities): generate the tool set per session with `defineDynamic`
 - [Sandbox](../sandbox): the sandbox the shell and file tools run in

package/dist/docs/public/advanced/dev-tui.md

@@ -9,22 +9,20 @@ description: "Drive an Eve agent locally in an interactive terminal UI: chat, st
 eve dev
 ```
 
-On startup the TUI prints a header for the connected agent — the model, instructions prompt, and the tools, skills, and subagents it has available:
+On startup the TUI prints a brand line with your agent's name, plus a rotating tip (local sessions only):
 
 ```text
- ▲ Weather Agent
- · Model         openai/gpt-5
- · Instructions  agent/instructions.md
- · Tools         get_weather, get_forecast, geocode
- · Subagents     researcher
- · Server        http://localhost:3000
-
- Type to chat  ·  ↑ history  ·  /new reset session  ·  /exit quit  ·  Ctrl+C interrupt
+ ▲ eve weather-agent
+ Use /channels to add more ways to reach your agent.
 ```
 
-From there the conversation streams straight into your terminal's normal scrollback — your prompts, the agent's replies, reasoning, tool calls, nested subagents, connection-authorization prompts, and any captured `stdout`/`stderr` — so you keep native scrolling, copy/paste, and a transcript that persists after you exit. Each turn is rendered without boxes: a colored gutter glyph marks who's speaking, tool calls collapse to a one-line summary (`✓ get_weather  city="SF" → 73°F`), and a subagent's work is indented beneath its `◆` header. A sticky line at the bottom shows the input prompt or the live status (spinner, token usage). Press `Enter` to send; `Ctrl+C` interrupts a running turn or quits at the prompt. Two slash commands: `/new` starts a fresh session and `/exit` quits.
+If agent discovery reported problems, an error/warning count renders between the two lines. Instructions, tools, skills, and subagents are one `eve info` away, and `/help` lists every command.
 
-The prompt input behaves like a shell line editor: `↑`/`↓` cycle through the messages you've sent this session, `←`/`→`, Home/End, and `Ctrl+A`/`Ctrl+E` move the caret, and `Ctrl+U`/`Ctrl+K`/`Ctrl+W` kill the line, the rest of the line, or the previous word. If a turn fails terminally — the server session dies or the connection drops — the TUI starts a fresh session and notes it inline so you can keep going (server-side context resets with the old session). Errors render compactly, with docs links highlighted, and a code bug escaping your agent's own code shows its stack trace dim beneath the error headline. Captured server `stdout`/`stderr` renders as dim, indented log runs behind a `│` rule — consecutive lines from the same source share one label, and nothing is ever hidden.
+From there the conversation streams straight into your terminal's normal scrollback — your prompts, the agent's replies, reasoning, tool calls, nested subagents, connection-authorization prompts, and any captured `stdout`/`stderr` — so you keep native scrolling, copy/paste, and a transcript that persists after you exit. Each turn is rendered without boxes: a colored gutter glyph marks who's speaking, tool calls collapse to a one-line summary (`✓ get_weather  city="SF" → 73°F`), and a subagent's work is indented beneath its `◆` header. A sticky footer keeps you oriented. The input prompt shows a dim `Type to chat · / for commands` placeholder while empty, and beneath it a persistent status line shows the model, the session's token flow (`⁕ ↑ 394.4K ↓ 4.3K`), the linked Vercel project and team (`▲ my-agent (acme)`), and a yellow `deploy pending` marker once a channel added this session still needs `/deploy`. The Vercel segment stays hidden until the directory is linked. Press `Enter` to send; `Ctrl+C` interrupts a running turn or quits at the prompt. Slash commands: `/new` starts a fresh session and `/exit` quits.
+
+When `eve dev` runs the server locally, three more slash commands manage the project without leaving the session. Bare `/model` opens a two-row configure menu that loops until Esc. "Change model" runs the same searchable model picker setup uses (the AI Gateway catalog, pre-selected on the model the runtime is serving); a model change is written into your agent's authored source, and the command reports success only after Eve confirms the new id (`/model <provider/model-id>` applies one directly, skipping the menu). The provider row opens the provider questions: which model provider to use (picking something other than AI Gateway shows wiring instructions for your own provider and stops there, leaving any existing setup untouched) and how to connect to AI Gateway — paste your own `AI_GATEWAY_API_KEY`, saved straight to `.env.local`, or connect via a project, which walks the same Vercel team/project pickers as setup (picking again re-links) and pulls the project's environment so an AI Gateway credential lands in `.env.local`; the dev server reloads env files automatically, no restart needed. The row demands attention (a bold yellow "Configure provider" with "Required to enable the agent") until a link or gateway credential is detected, then names the connection (e.g. "AI Gateway (Linked to my-project in my-team)") after, and each action's latest outcome stays visible beneath the menu (e.g. "✓ Model changed to openai/gpt-5.5"). `/channels` shows the agent's channel list — already-registered channels render as locked rows — and adds the one you pick, including the Slack Connect provisioning, then installs the dependencies the scaffold added so the dev server can load the new channels right away; after each addition the list repaints with the new channel locked, until Done (or Esc) leaves the flow. `/deploy` ships the agent to Vercel production, linking first when the directory is unlinked. Each command echoes as an invocation line, asks through a bordered panel that takes the input area's place — one question at a time, clearly separate from the chat transcript — and finishes with a one-line `⎿` result; loading states stay on the ephemeral status line instead of piling into the transcript. These commands are not available when connected to a remote server with `--url`, and when a turn fails because AI Gateway authentication is missing or stale, the error points you at `/model` directly. The TUI also checks at startup: a missing model-provider setup surfaces as an attention line — `⚠ 1 setup issue: model provider not linked · /model` — so the fix is visible before the first message fails, and each command's outcome hangs under it with the `⎿` connector.
+
+The prompt input behaves like a shell line editor: `↑`/`↓` cycle through the messages you've sent this session, `←`/`→`, Home/End, and `Ctrl+A`/`Ctrl+E` move the caret, and `Ctrl+U`/`Ctrl+K`/`Ctrl+W` kill the line, the rest of the line, or the previous word. If a turn fails terminally — the server session dies or the connection drops — the TUI starts a fresh session and notes it inline so you can keep going (server-side context resets with the old session). Errors render compactly, with docs links highlighted, and a code bug escaping your agent's own code shows its stack trace dim beneath the error headline. Captured server `stdout`/`stderr` renders as dim, indented log runs behind a `│` rule — consecutive lines from the same source share one label. Dev-server rebuilds condense further, into one status row that updates in place: `tui/setup-panel.ts changed · rebuilding…`, then `· rebuilt`. Only the latest rebuild shows, and paths shrink to their last two components. The TUI hides logs by default. `/loglevel <all|stderr|none>` switches what the transcript shows, and because logs stay buffered either way, the switch is retroactive: `/loglevel all` brings back everything captured so far, in its original order. Bare `/loglevel` reports the current mode; the `--logs` flag sets the starting one.
 
 The agent will sometimes need something from you, and the TUI asks inline. Tool approvals are a `y`/`n`. Option questions let you pick with `↑`/`↓` and `Enter`, or you can type a freeform answer. If a tool needs an authorized [connection](../connections), the URL shows up right in the transcript, and the turn picks back up once you've finished the flow.
 

package/dist/docs/public/advanced/instrumentation.md

@@ -162,4 +162,4 @@ When `eve build` fails on discovery errors, the CLI prints the full diagnostics
 - [`agent.ts`](../agent-config)
 - [Hooks](./hooks): observe the runtime event stream
 - [Dev TUI](./dev-tui): drive the agent locally
-- [Evals](./evals): repeatable scored checks
+- [Evals](../evals/overview): repeatable scored checks

package/dist/docs/public/advanced/meta.json

@@ -14,7 +14,6 @@
     "deployment",
     "instrumentation",
     "dev-tui",
-    "evals",
     "execution-model-and-durability",
     "sessions-runs-and-streaming"
   ]

package/dist/docs/public/advanced/security-model.md

@@ -42,7 +42,7 @@ A [channel](../channels/overview) is your agent's front door, which makes authen
   claims. A body field is attacker-controlled; treating it as identity is
   cross-user impersonation.
 
-The `support-fixture` dashboard channel is a concrete custom-channel fixture for these rules: it authenticates the raw body with an HMAC, compares signatures in constant time, and trusts the body-supplied principal only after the signature verifies.
+A custom channel that accepts dashboard-style webhooks should follow the same shape: authenticate the raw body with an HMAC, compare signatures in constant time, and trust any body-supplied principal only after the signature verifies.
 
 ## Authored markdown is data
 

package/dist/docs/public/advanced/session-context.md

@@ -60,7 +60,7 @@ Important behavior:
 
 ```ts
 const sandbox = await ctx.getSandbox();
-const result = await sandbox.run({ command: "pnpm test" });
+const result = await sandbox.run({ command: "npm test" });
 ```
 
 Important behavior:

package/dist/docs/public/agent-config.md

@@ -3,11 +3,11 @@ title: "agent.ts"
 description: "The agent's runtime config: defineAgent, the model, and compaction."
 ---
 
-Every agent has an `agent.ts` that calls `defineAgent` (from `eve`) to set its runtime config.
+An agent's `agent.ts` calls `defineAgent` (from `eve`) to set its runtime config.
 
 ## A good default
 
-Most agents need exactly this, a model:
+A typical config selects a model:
 
 ```ts title="agent/agent.ts"
 import { defineAgent } from "eve";
@@ -17,6 +17,9 @@ export default defineAgent({
 });
 ```
 
+The root `agent.ts` can be omitted when no runtime config is needed. In that case, Eve defaults
+to `anthropic/claude-sonnet-4.6`. When `agent.ts` is present, `model` is required.
+
 `model` can be a gateway model id string (which routes through the [Vercel AI Gateway](https://vercel.com/docs/ai-gateway)) or a provider-authored `LanguageModel`, when you want to call the provider directly, bypassing the gateway and configuring the model in code:
 
 ```ts title="agent/agent.ts"
@@ -61,7 +64,7 @@ A structured return type for task-mode runs: a subagent, schedule, or remote job
 
 ### `build`
 
-Hosted-build packaging, such as `externalDependencies` to keep listed packages external and trace them into the hosted output.
+Build packaging controls. `externalDependencies` keeps listed packages external while Eve compiles authored modules such as tools and channels, and traces those packages into the hosted output.
 
 ## Where adjacent settings live
 

package/dist/docs/public/channels/eve.mdx

@@ -36,7 +36,7 @@ The `auth` option decides who can call these routes. The built-in helpers are me
 
 Neither admits browser users or external clients in production. For a public app, wire the channel to your own auth (Clerk, Auth.js, or your own OIDC/JWT verification).
 
-`pnpm create eve@beta` scaffolds an `agent/channels/eve.ts` with a production placeholder so you replace it before going live. The generated channel allows Vercel OIDC and localhost, and includes `placeholderAuth()`, which returns a setup-focused 401 in production until you swap it for real auth. Delete the file and Eve falls back to `[localDev(), vercelOidc()]`, which still does not admit browser users in production.
+`eve init` scaffolds an `agent/channels/eve.ts` with a production placeholder so you replace it before going live. The generated channel allows Vercel OIDC and localhost, and includes `placeholderAuth()`, which returns a setup-focused 401 in production until you swap it for real auth. Delete the file and Eve falls back to `[localDev(), vercelOidc()]`, which still does not admit browser users in production.
 
 For the full auth model and helper list, see [Auth & route protection](../advanced/auth-and-route-protection).