@bastani/atomic 0.9.12 → 0.9.13-alpha.2

package/CHANGELOG.md

@@ -2,6 +2,156 @@
 
 ## [Unreleased]
 
+## [0.9.13-alpha.2] - 2026-08-12
+
+### Breaking Changes
+
+- **Streaming `message_update` events are now delta-only across the public API.** JSON, RPC, SDK, and extension subscribers receive only `type` plus the raw `assistantMessageEvent`; the cumulative `message` and wire-only `assistantMessageEvent.partial` fields are gone. `message_start` seeds the message, ordered deltas build it without deduplication, and `message_end` supplies the authoritative final message. Consumers that read `message` or `partial` from `message_update` must accumulate deltas themselves. This also removes quadratic wire amplification: an 8000-character answer wrote 8,428,280 bytes across 501 frames and now writes 65,588, with bytes growing linearly rather than quadratically.
+- **Extension provider OAuth `login` and `refreshToken` now take abort signals.** The fork-owned contracts are `login(callbacks, signal)` and `refreshToken(credentials, signal)`, both with a required concrete `AbortSignal`. Atomic forwards cancellation through both operations, so cancelling authentication or credential refresh abandons the extension's in-flight network work. Provider implementations must accept the new parameter and forward it into their network requests.
+- **`ModelRegistry.refresh()` returns `ModelsRefreshResult` instead of `Promise<void>`.** Callers now receive pi's real result — `{ aborted, errors }`, a per-provider `ReadonlyMap<string, Error>` plus a cancellation flag — so a refresh in which some providers failed, or which was aborted, is no longer indistinguishable from a clean success. Atomic previously discarded the result behind a defensive cast and an `?? {}` fallback; both are gone and the value pi returns is passed through. Callers that awaited `refresh()` purely for its completion need no change; callers that want to report provider failures or honor cancellation can now read them.
+- **`ModelRuntime.setRuntimeApiKey()` no longer refreshes the model catalog.** It now takes auth-cancellation options and does only what its name says: record the runtime key and mark the provider as configured, aborting early if the supplied signal is already aborted. Previously it awaited a full catalog refresh internally, so setting a key paid for a network refresh whether or not the caller wanted one and could not be cancelled. Callers that need a fresh catalog after setting a key must now call `refresh({ providers: [providerId], signal })` themselves.
+- **`ctx.ui.custom()` components now report input consumption.** Their `handleInput` method must return `true` when they consume a key and `false` or `undefined` when they do not; extensions whose handlers return `void` must migrate. In fullscreen mode, unhandled viewport keys may reach the transcript. A remote component that is disposed before its reply is deliberately dropped rather than replayed.
+- **Custom extension runtime and UI implementations must adopt lifecycle fields.** Exported `ExtensionRuntime` implementations must provide `trackEventBusSubscription(unsubscribe)`, and `ExtensionUIContext.getChatRenderSettings()` must return `markdownTransformers`. `pi.events` is now bound to an extension instance: subscriptions auto-unsubscribe on reload or disposal, and calling `on()` or `emit()` through a captured stale handle throws.
+- **`ctx.ui.custom()` components now report input consumption.** A `handleInput` method must return `true` when it consumes a key and `false` or `undefined` when it does not; extensions whose handlers return `void` must migrate. In fullscreen mode, unhandled viewport keys can reach the transcript, while a disposed remote component's late reply is dropped.
+- **Interactive TTY sessions now always use the fullscreen renderer.** The `--tui-mode` flag, `tuiMode` JSON setting, `/settings` TUI-mode row, settings-manager accessors, and exported `TuiMode` alias are removed. There is no replacement for `--tui-mode regular`; fullscreen is the interactive mode. `fullscreenScrollbar` remains configurable. Non-TTY output continues through print mode, while internal `TERM=dumb` and CI guards retain the main-screen renderer as a non-user-selectable fallback.
+
+### Added
+
+- Added an **experimental** `@bastani/atomic/client` entrypoint with `RemoteSession` lifecycle management and transcript projection helpers for remote protocol sessions.
+- Added the fullscreen interactive TUI layout. Wheel and trackpad gestures reach focused workflow overlays before the transcript viewport, and events those overlays do not consume fall through to the fullscreen transcript.
+- Fullscreen sessions keep the editor, status line, extension widgets, usage meter, and footer in a sticky dock while the transcript scrolls independently; terminal resizes preserve that layout.
+- Fullscreen mode now supports a draggable transcript scrollbar configured as `auto`, `always`, or `hidden` through `/settings`; themes can set `scrollbarThumb`, which falls back to `selectedBg` for existing custom themes.
+- Fullscreen transcript navigation now supports page, opt-in half-page, marked-message, and `home`/`end` shortcuts, while `ctrl+home`, `ctrl+end`, `ctrl+pageUp`, and `ctrl+pageDown` remain available to the editor. Matching viewport bindings pass through to focused fullscreen overlays and inline custom components, so host selectors and workflow stage-chat paging stay local. Prompt history can be enabled with explicit `tui.editor.historyPrevious`/`historyNext` bindings such as `ctrl+p`/`ctrl+n`; those bindings take precedence over ordinary application action handlers while the editor is focused.
+- Windows fullscreen sessions now paste system clipboard text on a secondary-button press, with focus checks and bracketed-paste framing preserved for host and engine-owned editors.
+- Added the bundled, default-on ADHD-friendly response mode. Its hidden rules message persists across turns, resumes, branches, and compaction; `/i-have-adhd on|off`, `--no-adhd`, `.i-have-adhd-off`, and the `stop adhd mode` / `normal mode` phrases control it.
+
+- Added `atomic auth check` to verify a provider or model's effective authentication before a session starts. It reports `ready`, `not_ready`, or `invalid`, supports JSON output and a no-refresh read-only mode, and does not emit credential material unless `--credentials` explicitly requests it.
+
+- Added the explicit `--credentials` opt-in to `atomic auth check` for scripts that need the resolved credential. Checks remain status-only unless that flag is present.
+- Added immutable system-prompt contribution constants for Atomic's builtin coding tools, with alignment coverage that keeps each factory's prompt metadata in sync without duplicating prompt assembly.
+
+- Added `CredentialSynchronizationError` for credential mutations that commit successfully but fail to update local model state. Credential-save failures remain distinct from post-commit synchronization failures, and interactive auth surfaces report the distinction.
+
+- Added `terminate` to blocked extension `tool_call` results. A terminating blocked call suppresses the follow-up model request only when every finalized call in its batch also terminates.
+
+- Added the configurable and package-exported `createCodingAgentHarness()` factory, with Atomic's hashline coding tools, live prompt metadata, Atomic and `PI_*` bash session variables, injectable pi-agent-core execution environments for read (including directory trees), bash, edit, write, and find, and runtime rejection of environments that omit `renameFile()`. URL reads now use the session id for cache scope, avoid unsupported session-manager access, and do not persist host-local artifacts in factory contexts. The default `search` tool remains local until its filesystem and matching pipeline gains a complete remote operations seam; read and edit retain local path-variant probes and notebook projection, read also retains local archive, SQLite, and internal-resource selectors, write retains local generated-file, shebang, conflict, and resource helpers, and bash validates its cwd locally and uses Atomic's local temp storage for overflow output.
+- Added immutable system-prompt contribution constants for Atomic's built-in coding tools, keeping custom harnesses and tool registries aligned with the prompt metadata used by the built-in factories.
+- Added `terminate` to blocked extension `tool_call` results. A terminating blocked call suppresses the follow-up model request only when every finalized call in its batch also terminates.
+
+- Added `pi.registerMarkdownTransformer()` to the Atomic extension API for ordered, synchronous, display-only Markdown transforms in user, assistant, and assistant-thinking transcript output. Each transformer receives the prior transformer's output plus message type, streaming state, and available width; errors leave the current display text intact and do not alter stored messages or model context ([#7231](https://github.com/earendil-works/pi/pull/7231) by [@xl0](https://github.com/xl0)).
+- Added themed Unicode Mermaid diagrams in interactive transcripts, with optional rendering while responses stream ([#7624](https://github.com/earendil-works/pi/pull/7624) by [@xl0](https://github.com/xl0)).
+- Added configurable terminal-friendly Unicode LaTeX math in interactive transcripts via the `markdown.latex` setting. See [Markdown settings](docs/settings.md#markdown).
+- Added per-directory `AGENTS.override.md` context files. An override replaces that directory's `AGENTS.md` or `CLAUDE.md` while global and ancestor context still layers normally; the same canonical filename works through Atomic's legacy `.pi` and `PI_*` agent-directory aliases ([#7681](https://github.com/earendil-works/pi/pull/7681) by [@m3](https://github.com/m3)).
+- Added `AI_AGENT=atomic` attribution at the CLI, RPC, and compiled-binary entry points and across Atomic-owned child-process environments, including bash/tool commands, isolated RPC children, subagent/workflow runners, MCP and web-access subprocesses, and the intercom broker. Child environments follow upstream's overwrite policy while cloning rather than mutating caller-supplied environment objects.
+
+- Added arbitrary OpenAI-compatible `samplingParams` for custom model definitions, per-model overrides, extension providers, and per-request stream options. Parameters merge by key, preserve provider-specific names, and apply to OpenAI-compatible request bodies; malformed configuration shapes now fail during `models.json` loading. Added opt-in vLLM `compat.supportsThinkingTokenBudget` handling and `compat.supportsFinishReason` stop inference for finishless OpenAI-compatible streams.
+- Added built-in Baseten support with `BASETEN_API_KEY`, the `zai-org/GLM-5.2` default, and the catalog's provider-specific thinking levels.
+- Added Qwen Token Plan Individual with the documented subscription catalog, shared international `QWEN_TOKEN_PLAN_API_KEY`, and `qwen3.8-max` as its default model.
+
+- Added the fullscreen interactive TUI layout. The transcript scrolls above a sticky dock containing the editor, status line, usage meter, extension widgets, and footer, while wheel and trackpad input routes first to focused engine-hosted overlays such as workflow graphs ([#2269](https://github.com/bastani-inc/atomic/pull/2269), [#2274](https://github.com/bastani-inc/atomic/pull/2274)).
+- Added a draggable fullscreen transcript scrollbar with `auto`, `always`, and `hidden` settings. Custom themes may define the optional `scrollbarThumb` color; existing themes fall back to `selectedBg` ([#2275](https://github.com/bastani-inc/atomic/pull/2275)).
+- Added fullscreen transcript navigation for page, half-page, marked-message, `home`, and `end` actions, explicit `tui.editor.historyPrevious`/`historyNext` prompt-history bindings, and Windows secondary-button clipboard paste. Focused overlays and custom components receive matching viewport keys first and can return `false` to let transcript scrolling continue ([#2281](https://github.com/bastani-inc/atomic/pull/2281), [#2296](https://github.com/bastani-inc/atomic/pull/2296), [#2300](https://github.com/bastani-inc/atomic/pull/2300)).
+
+- Tool output that spills to disk is now stored per session and reaped on age, so the system temp directory can no longer fill up with orphaned files. `bash` overflow logs, background-job logs, streamed spill files, and the in-memory-session tool-result fallback all move from the bare temp directory into one owner- and session-scoped tree, `<tmpdir>/atomic-<uid>/<session-id>/`, created `0700` with `0600` files where the platform supports it. Where there is no uid, as on Windows, the domain-qualified account name plus a short digest of it replaces the uid, so neither two names that reduce to the same safe path component nor two domain principals sharing a bare account name (`CONTOSO\Alice` and `FABRIKAM\Alice`) can land on one tree. A session id containing path separators or `..` is reduced to a single safe component and cannot place files outside that tree. A session directory deleted out of band — by a system temp reaper, or by another process's sweep — is recreated on the next write instead of handing back a dead path. Sessions that persist to disk keep writing tool results to `<sessionDir>/tool-results/` exactly as before.
+- That temp path is predictable, so Atomic validates it instead of trusting it. Every component below the system temp directory is created one level at a time and checked: a symlink is refused, a directory owned by another account is refused, and one of this account's left too permissive is tightened to `0700` and re-checked. Refusal fails closed — the tool runs with no spill file and advertises no path, rather than writing command output through a link someone else planted at the predictable owner root. Only the final session directory is replaceable: a stale file or link exactly where a session directory belongs is removed (a link by itself, never its target) and recreated.
+- On Windows, that ownership check is now real rather than skipped ([#2245](https://github.com/bastani-inc/atomic/issues/2245)). Before adopting an existing temp component, Atomic reads its security descriptor through PowerShell `Get-Acl` and decides from SIDs alone: the owner must be the current user, SYSTEM, or Administrators, and every access-allowed DACL entry must grant only those principals (plus inherit-only `CREATOR OWNER`/`OWNER RIGHTS` plumbing). A foreign-owned root, a root readable or writable by other accounts (a machine-wide redirected `TMP`), and a descriptor that cannot be read or fully parsed are all refused — failing closed to "no spill file", as on POSIX. A successful verification is cached per path per process, keyed to the directory's metadata identity (creation, change, and last-write times): a directory swapped underneath the path and an in-place DACL edit — which bumps the NTFS change time — both force a fresh descriptor read, and only a directory whose metadata is unchanged skips the subprocess; the symlink and directory checks still run on every use.
+- Persisted tool-output files are capped at 64 MB each. Output that lands exactly on the cap is kept whole; output that passes it is cut off there with `[Output truncated: persisted-output cap of 67108864 bytes reached]` in place of the rest. The cut lands on a character boundary even when a multi-byte character is split across two chunks of streamed output, and bytes that are not valid UTF-8 — binary command output — are written through unchanged rather than decoded. The tool result and its `fullOutputPath` are unchanged, except that a spill file which fails to write is no longer advertised: `bash` reports no path rather than one pointing at a missing file, and the underlying stream error can no longer surface as an uncaught exception. A storage fault never changes the outcome of the command that produced the output: a background job whose process exited `0` but whose spill file could not be flushed is still reported completed with its real exit code rather than failed, since reporting a storage fault as command failure would invite a caller to re-run a command that already had its effect.
+- A deferred, unref'd sweep started shortly after session startup removes session temp trees and `tool-results` directories whose newest file is more than 30 days old. It covers the default project-nested session roots and a custom session directory selected with `--session-dir`, `ATOMIC_CODING_AGENT_SESSION_DIR`, or the `sessionDir` setting, whose tool results have no project nesting to walk. The newest entry decides the whole directory — one fresh file keeps every stale sibling beside it, so a path a tool result recorded weeks ago stays readable while the tree is in use. Each target is throttled to once a day by a `.last-cleanup` marker and guarded by a `.cleanup.lock` exclusive lock, so concurrent sessions never race and a lock left by a crashed process is broken only once stale; session-storage targets keep that pair in a control directory under the temp root, out of the directory being scanned. Session transcripts and `.jsonl` session files are outside the sweep's scope entirely, symlinked scan roots and symlinked `tool-results` directories are skipped rather than followed, and a temp tree belonging to a session the running process is using is never reaped — a `Full output saved to:` path stays valid for the life of the session that produced it.
+
+### Changed
+
+- Footer and workflow stage-chat extension statuses are now joined with a dim `•` separator instead of a bare space, so adjacent statuses read as distinct items.
+- Live tool-call argument previews no longer update while a tool call's arguments are still streaming. On the default interactive path the assistant message is now rebuilt from `message_update` deltas, and `toolcall_start`/`toolcall_delta` carry an unparsable fragment of JSON rather than a usable value, so the accumulator deliberately ignores them and installs the complete parsed call at `toolcall_end`. The tool card itself still appears as soon as the call starts — it is created by `tool_execution_start`, not by the argument stream — and the arguments it finally shows are correct and complete; what is gone is the partially-filled argument text that used to fill in progressively before the call was ready. This is a deliberate trade for correctness: the previous incremental preview was only possible because every streaming frame re-sent the entire message, which is exactly the quadratic amplification removed above.
+
+- Adopted the pi 0.84.1 coding-agent, agent, AI, client, protocol, telemetry, and TUI runtime surfaces. The TUI adoption supplies the renderer behavior described above; Atomic keeps its isolated interactive engine, provider, path, and legacy `PI_*`/`.pi` compatibility surfaces.
+
+### Fixed
+
+- Fixed fullscreen mouse routing for focused workflow graphs and stage chats. SGR/X10 wheel and click sequences now reach the focused overlay through the existing bounded custom-input reply path before the alternate-screen viewport, so graph scrolling and click-to-attach work without changing keyboard viewport fallback behavior ([#2303](https://github.com/bastani-inc/atomic/issues/2303)).
+- Fixed registration-triggered offline model catalog refreshes racing newer credential-resolved refreshes, so configured `$ENV` and `!command` credentials remain authoritative ([#2318](https://github.com/bastani-inc/atomic/issues/2318)).
+- Fixed fullscreen overlay text selection when one input read contains multiple SGR mouse reports. Reports are replayed one at a time, while Shift/Option/Ctrl-modified presses and motion stay out of application selection; modifier-bearing SGR releases still close an active selection, and legacy X10 release remains excluded because pi-tui cannot identify it for selection.
+- Fixed `atomic auth check --no-refresh` resolving command-backed API keys as configured values rather than treating `!command` as a literal credential.
+
+- Fixed `atomic auth check --json` labelling an unresolved model as its provider. Invalid results now omit `provider` until a provider has actually resolved, and `print-api-key` consistently rejects `--min-expiry` before or after a `--` terminator.
+- Fixed fullscreen viewport shortcuts stealing matching keys from a focused overlay or inline custom component. A focused handler now receives the key exactly once and can return `false` to let a no-op fall through to transcript navigation. Fullscreen shutdown also avoids restarting the replacement renderer, and the `ctrl+x` copy shortcut uses a transient `Copied!` flash instead of growing the fixed status dock.
+- Fullscreen rendering now inherits pi-tui 0.84.1's stacked flashes, corrected multi-click selection and focus reset, reduced multiplexer mouse tracking, nested stack minimum sizing, and Kitty image clipping at dock/layout boundaries, single-pass redraw/retransmission, and offscreen caching fixes. Atomic deliberately uses the dependency implementations rather than carrying forked renderer or layout copies.
+- Fixed fullscreen viewport shortcuts stealing matching keys from a focused overlay or inline custom component. A focused handler now receives the key exactly once; returning `true` keeps a consumed key local, while an unhandled result falls through to transcript navigation. The `ctrl+x` copy shortcut uses a transient `Copied!` flash instead of growing the fixed status dock.
+- Verified the inherited pi-tui 0.84.1 fullscreen fixes against the upstream commits `17090d4` (`AltScreenFlashContainer` stacked flashes), `58fc043` and `171c6b5` (`TuiAltScreen` multi-click selection and focus reset), `fc3554e` (multiplexer mouse tracking), `f24ab6e` (`VStack` nested minimum sizing), `af187ee` (layout Kitty clipping), `73414d0` (layout/terminal-image redraw reuse), and `a8ee03b` (terminal-image offscreen caching). Atomic uses the dependency implementations rather than carrying forked renderer or layout copies.
+- Fixed fullscreen viewport shortcuts stealing matching keys from a focused overlay or inline custom component. When host focus policy defers transcript navigation, the matching input stays with the focused component. Fullscreen shutdown avoids restarting the replacement renderer, and `ctrl+x` shows a transient `Copied!` flash instead of growing the fixed status dock.
+- Fixed isolated fullscreen custom components swallowing viewport navigation when their child handler did not consume the key. Input replies now correlate with each forwarded key, repaint remote components after input, and let transcript scrolling proceed on an explicit `false`, a failed reply, or a bounded timeout while preserving focus and key-release semantics.
+- Fixed isolated fullscreen tool renderers rebuilding Kitty image components on every engine render, which retransmitted visible payloads and rendered the transcript twice per frame during layout changes.
+- Verified the remaining fullscreen polish against pi-tui 0.84.1: stacked flashes, phantom-selection prevention, nested stack minimum sizing, Kitty clipping at dock boundaries, redraw reuse, and offscreen caching are supplied by the dependency. Its multi-click text-selection and tmux/Zellij/Screen mouse-tracking reductions also ship there, so Atomic deliberately does not port them. The fullscreen shutdown query-leak fix is already present in Atomic's `7a5a2674e` core layer.
+- Fixed extension event-bus subscriptions surviving session reload or disposal ([#7193](https://github.com/earendil-works/pi/issues/7193)).
+- Fixed extension event-bus subscriptions surviving session reload or disposal ([#7656](https://github.com/earendil-works/pi/pull/7656) by [@tudoroancea](https://github.com/tudoroancea)).
+- Fixed stale extension-context detection for bundled extensions. The public `isStaleExtensionContextError()` predicate and its marker now belong to the host beside the error path, so session replacement and reload guards cannot silently drift from host wording.
+- Fixed `ChatSessionHost` tail-windowing settled assistant rows while a later answer streamed; only the live assistant row is now trimmed.
+- Fixed bare exact `--model` IDs shared by providers choosing catalog order. Atomic now uses the sole authenticated matching provider or reports an ambiguity with fully qualified choices ([#7327](https://github.com/earendil-works/pi/issues/7327)).
+- Fixed `/model <name>` and `/scoped-models` waiting for remote catalog refreshes before using cached models. Exact cached matches now resolve immediately, and the scoped selector renders its cache first, refreshes in the background, and cancels that work when it closes ([#7153](https://github.com/earendil-works/pi/issues/7153), [#7443](https://github.com/earendil-works/pi/issues/7443)).
+- Fixed project `retry.provider` overrides discarding unmodified global provider retry fields by recursively merging nested settings ([#7572](https://github.com/earendil-works/pi/issues/7572)).
+- Reduced automatic terminal-theme detection latency by starting color-scheme and background probes together.
+- Softened the bash session-environment prompt while retaining Atomic and legacy `PI_*` environment support.
+
+- Verified the remaining pi-tui 0.84.1 renderer fixes used by fullscreen and extension UI: stacked flashes, phantom-selection prevention, nested-stack minimum sizing, Kitty clipping at dock boundaries, redraw reuse, offscreen caching, multi-click word/line selection, and reduced mouse tracking in tmux, Zellij, and GNU Screen. The dependency also supplies the LaTeX whitespace and matrix fixes; Atomic does not vendor a second TUI implementation.
+
+- Fixed same-state `ctx.ui.setToolsExpanded()` calls still traversing the transcript or requesting a custom-UI render. Repeated bundled-extension requests to keep tool output collapsed now leave an already-collapsed interactive or RPC UI untouched.
+- Fixed extension-facing `ModelRegistry.complete()` requests to dispatch through the active model runtime, so registered providers and resolved request authentication remain available to extension code.
+- Fixed image blocks returned by builtin and extension tools entering history at their original size. They now honor `images.autoResize` after `tool_result` handlers run, including images injected by an extension; a failed conversion leaves the original image intact.
+- Fixed custom editors installed through `ctx.ui.setEditorComponent()` that expose `setAutocompleteMaxVisible()` not inheriting the default editor's active autocomplete dropdown limit.
+
+- Fixed in-process child sessions losing their admission-issued nesting depth and delegation limit. `SubagentChildPolicy` now carries both the admitted `depth` and the effective `maxSubagentDepth`, so the subagent executor can enforce the configured and inherited limits without relying on the removed process-environment bridge ([#2220](https://github.com/bastani-inc/atomic/pull/2220), regression from [#2205](https://github.com/bastani-inc/atomic/pull/2205)).
+- Release archives no longer ship packages built for a different architecture. Every archive is produced on one Linux x64 runner, and that runner's `node_modules` was copied verbatim into all of them, so 0.9.12's `atomic-linux-arm64.tar.gz` and `atomic-darwin-arm64.tar.gz` both contained `@esbuild/linux-x64`. The build now prunes the `@embedded-postgres` leaves that cannot run on the archive being staged and then fails the build if any staged package's own `os`/`cpu`/`libc` declaration is incompatible with that archive ([#2208](https://github.com/bastani-inc/atomic/issues/2208)).
+- Reduced redundant repaints of a Ctrl+O-expanded live subagent widget, which were scrolling the chat window to the bottom and clearing terminal scrollback during a run. The widget was republished on every child session event, including assistant streaming deltas, so it repainted continuously instead of when its contents changed; each repaint of a row above the terminal fold costs a pi-tui full redraw that clears scrollback. Progress is now published only at milestones that change what the widget shows. A genuine above-fold milestone change can still require that redraw when the editor/footer region plus the live widget exceed the terminal height — that path is upstream (earendil-works/pi#4785, #7194) and is not fixed here ([#2213](https://github.com/bastani-inc/atomic/pull/2213), regression from [#2205](https://github.com/bastani-inc/atomic/pull/2205)).
+- Fixed the scroll anchor in `ScrollableComponentViewport` drifting when its content changes height. The offset is stored as a distance from the bottom and was compensated only when content grew, so a component losing rows — such as a live subagent widget dropping its current-tool row at the end of each tool call — slid the rows a scrolled-up viewer was reading. The offset is now re-derived from the anchored row, which also keeps the anchor still when rows change *above* it; a plain height-delta adjustment moved the viewer in that case. Because the chat host renders the entire transcript as a single component, the anchored row is normally *inside* one component, where per-component row counts cannot say whether rows changed above or below it; a windowed component can now report a row map (`rowSegments`) identifying each run of rows by the entry that produced it, so a compaction that deletes entries above the viewer — renumbering every positional cache key — still leaves the viewer on the entry they were reading. If the anchored entry itself was compacted away, the viewer lands on the nearest surviving entry above it. A viewer already at the bottom still sticks to the bottom. This affects the workflow stage chat view and any extension surface built on `ChatSessionHost`; it is not the path the normal interactive chat uses, so it does not by itself preserve main-chat terminal scrollback ([#2214](https://github.com/bastani-inc/atomic/pull/2214), regression source [#2205](https://github.com/bastani-inc/atomic/pull/2205)).
+- Fixed a GitHub Copilot `openai-responses` session hard-stopping with `400 invalid_request_body` once a replayed tool call entered its context. The outbound payload sanitizer treated an *absent* `function_call.id` the same as a malformed one and synthesized a replacement from `call_id`; on the Copilot path across a model-identity mismatch (a mid-session model switch, a session resumed under another model, or an aliased model id) that synthesized value was always exactly 64 characters, and the server rejected it. Because the value is a pure function of the tool call, every resend rebuilt the identical rejected identifier and nothing retries HTTP 400, so the session could not recover without hand-editing its context. An absent id is now left absent, and only an id that is actually present and malformed is repaired. The synthesized-length budget and the id validator were also tightened to agree on a single maximum of 63 characters, below the 64-character length the server was observed to reject rather than at it, so the sanitizer can no longer bless a value the server refuses ([#2204](https://github.com/bastani-inc/atomic/issues/2204)).
+- Fixed the model list widening to a provider's whole built-in catalog after an extension registered an additive provider override, so a GitHub Copilot account entitled to one model could be offered every catalog model. Availability is a per-model fact — a credential can entitle an account to a subset of a provider's catalog — but the model-list refresh and registration projections rebuilt it from the provider's *configured* status, which reads as "every model this provider has". Registering an override recomposes the provider and republishes, so the credential-filtered result was replaced by the full catalog, and it stayed replaced whenever the background refresh that the registration schedules superseded the availability pass a caller was awaiting. Republishing the model list now carries the previous availability projection across by model ID, including an empty one, and a registration exposes a provider's catalog provisionally only when that registration is what newly configures the provider.
+
+
+- Fixed a `null` request-header suppression marker being discarded instead of suppressing the provider default header. pi types request headers as `ProviderHeaders` (`Record<string, string | null>`) and documents a `null` value as suppressing the provider or API default header of the same name. Atomic narrowed that to `Record<string, string>` and enforced the narrowing at runtime in **two** places — `getApiKeyAndHeaders` on `ModelRegistry`, and `_getRequiredRequestAuth` on the agent session — each filtering every `null` entry out before the headers reached the request layer. The provider default was then sent anyway, and a header map consisting only of suppression markers arrived as `{}`, which reads as nothing configured. Both filters are gone and markers now pass through untouched. Note that this is **not** reachable from `models.json`: that schema declares headers as `Record<string, string>` and rejects a `null` value outright, so a marker has to come from a source that speaks `ProviderHeaders` — pi's provider-level catalog headers, the credential-derived request auth pi returns from `getAuth()`, or the `before_provider_headers` extension hook. The agent-session filter additionally affected every request that resolves auth through that path rather than through the registry: compaction, auto-compaction, post-tool-compaction, and branch-summary requests each discarded the marker on every call.
+- Fixed a chat surface built on `ChatSessionHost` — the workflow stage chat, and any extension pane using it — showing no assistant text at all when it was opened onto a session that was already streaming. Now that `message_update` carries only the next delta, a host mounted part-way through a turn sees neither the `message_start` that opened the assistant message nor the fragments that already went past, and nothing later restates them; the message itself reaches the transcript only when the turn ends, so a stage held mid-turn stayed blank for as long as it streamed. A host now adopts the session's in-flight assistant message when it mounts, and the deltas that arrive afterwards continue that message instead of opening a second one. Streaming consumers also assemble into their own copy of the message: an in-process subscriber is handed the provider's live partial, whose `content` array pi-ai keeps appending to, so applying deltas in place wrote each fragment on top of the accumulation the provider had already done and doubled the visible text (regression from the delta-only `message_update` change above).
+- Manual `/compact` now keeps agent events live, cancels unfinished automatic compaction safely, and then executes the requested manual compaction instead of rejecting it or returning an automatic boundary with different parameters.
+- Input queued during manual compaction now sends after completion because the session exposes its idle state before `compaction_end` listeners run.
+- A response truncated before its requested output limit now gets one compact-and-retry attempt; output-cap stops retain bounded direct continuation, and the TUI reports neutral truncation text.
+- Fixed a stale automatic post-compaction retry resuming after a manual `/compact` finished during the continuation grace period.
+- Fixed a short transcript with repeated below-cap truncations retrying compaction more than once when no compaction boundary could be made, while preserving load-bearing recovery for a later actual context overflow.
+- Fixed `ChatSessionHost` leaving queued prompts stuck when compaction was cancelled or failed; queues now drain after every non-mid-turn compaction completion.
+- Fixed malformed package-manifest resource fields that could block valid resources or crash startup ([#7187](https://github.com/earendil-works/pi/issues/7187)).
+- Fixed Git package updates left incomplete after `git clean` fails, repairing missing dependencies and retrying cleanup on the next update ([#7570](https://github.com/earendil-works/pi/issues/7570)).
+- Retried transient management requests for explicit version checks, remote catalogs, managed-tool downloads, and self updates without multiplying the overall timeout budget ([#7632](https://github.com/earendil-works/pi/issues/7632)).
+- Fixed Windows fallback `find` glob patterns with paths returning no results when `fd` uses native separators ([#6817](https://github.com/earendil-works/pi/issues/6817)).
+- Fixed `find` results at filesystem roots, including POSIX `/` and Windows drive roots, so paths keep their first segment, directory matches use one trailing slash, and POSIX filenames ending in a backslash remain literal ([#6104](https://github.com/earendil-works/pi/issues/6104), [#7569](https://github.com/earendil-works/pi/pull/7569)).
+- Fixed Git Bash, MSYS, Cygwin, and WSL drive paths resolving against the current Windows drive instead of their native drive ([#7064](https://github.com/earendil-works/pi/issues/7064), [#7547](https://github.com/earendil-works/pi/issues/7547)).
+- Fixed linked project session directories being omitted from the session picker ([#7552](https://github.com/earendil-works/pi/issues/7552)).
+- Fixed standalone x64 binaries failing to start on pre-Haswell CPUs that lack AVX2/BMI2 by using Bun's baseline target, including musl archives ([#7390](https://github.com/earendil-works/pi/issues/7390)).
+- Fixed credential-resolved model endpoints — including GitHub Copilot Business and Enterprise hosts — falling back to the catalog endpoint for Verbatim Compaction, branch summaries, MCP sampling, and direct web summaries. Atomic now retains the resolved `baseUrl` alongside request auth, so request-header `null` suppression markers still pass through unchanged ([#6768](https://github.com/earendil-works/pi/issues/6768), [#7579](https://github.com/earendil-works/pi/issues/7579)).
+- Fixed native macOS `Cmd+V` so clipboard images paste into the editor without duplicate image inserts from terminal key repeats ([#2186](https://github.com/bastani-inc/atomic/issues/2186)).
+
+### Removed
+
+- Removed the `tsx` dependency from the shipped runtime. The Intercom broker's Node path now runs on Atomic's bundled `jiti` loader, which is dependency-free pure JavaScript, instead of resolving `tsx` first. `tsx` was the only thing pulling `esbuild`, so this removes `esbuild`, its 26 platform packages, and `fsevents` from `npm-shrinkwrap.json` and from every release archive — about 11.5 MB across 72 files. The `npx --no-install tsx` config pair remains a recognized compatibility sentinel, and explicit custom broker commands are unaffected ([#2208](https://github.com/bastani-inc/atomic/issues/2208)).
+- Behavior change worth noting for extension authors: TypeScript reached through the Node broker path is now transpiled by jiti (Babel-based) rather than esbuild. Both erase types without type-checking, but they are not identical — jiti defaults JSX support and `tsconfig` path aliases to off, and enables legacy decorators. A third-party extension that resolved `tsx` from the bundled `node_modules` was relying on an undocumented dependency and will no longer find it ([#2208](https://github.com/bastani-inc/atomic/issues/2208)).
+
+- Removed the Atomic-only workflow custom-terminal mouse-scroll control from the isolated engine protocol and host terminal-mode arbiter; remote overlays retain the typed autowrap control only. Focused workflow overlays now leave `ctrl+t` to the host `app.thinking.toggle` action, while the tree selector continues to receive the same key through `app.tree.filter.noTools`. Fullscreen pi-tui owns application selection over overlays; copy still uses OSC 52, and terminals that refuse OSC 52 writes require the modifier-drag bypass (Shift/Option, as provided by the terminal).
+
+## [0.9.13-alpha.1] - 2026-08-05
+
+### Added
+
+- `CreateAgentSessionOptions` and `ExtensionContext` now carry an immutable typed subagent child capability policy for in-process tool registration ([#2188](https://github.com/bastani-inc/atomic/issues/2188)).
+- `CreateAgentSessionOptions` now supports construction-time system-prompt and inherited-context transforms, allowing in-process child sessions to apply typed policy without extension injection ([#2188](https://github.com/bastani-inc/atomic/issues/2188)).
+
+### Changed
+
+- Typed child policy now includes MCP direct-tool selection and admission-issued Intercom identity/capability, allowing in-process extensions to consume child policy without inheritable environment state ([#2188](https://github.com/bastani-inc/atomic/issues/2188)).
+- Bundled subagent launches now use in-process `AgentSession` children with typed admission, canonical identities, and live async continuation; `async: true` work is owned by the parent process and does not survive parent exit ([#2188](https://github.com/bastani-inc/atomic/issues/2188)).
+
+### Fixed
+
+- Workflow runs awaiting human input now use the blue `？` indicator in the BACKGROUND panel, `/workflow connect` picker, and `/workflow status` listing, including prompts raised by hidden nested workflow children; the indicator returns to the run's current state when the prompt resolves.
+- Fixed main-chat model fallback to classify provider failures consistently with workflows, advance rejected credentials and incompatible or unavailable models to the next candidate, and restore the user-selected model at the next turn without overriding an explicit `/model` choice. A failure that requesting the same model again cannot repair now takes that model out of the chain for the rest of the turn at every reasoning level, so a fallback entry differing only by its `:low`/`:high` suffix is skipped instead of retrying the same dead credential; transient rate-limit and transport failures keep those reasoning variants ([#2170](https://github.com/bastani-inc/atomic/issues/2170)).
+- Fixed a context overflow that compaction cannot resolve to advance the configured `fallbackModels` chain instead of ending the turn, so a larger-context candidate can answer. Compaction still runs first, and a compactable overflow spends no fallback candidate ([#2170](https://github.com/bastani-inc/atomic/issues/2170)).
+- Fixed a reasoning-level change during a model fallback stranding the session on the fallback model. Changing reasoning effort is not a model choice, so it no longer cancels the pending restore; the next turn returns to the user-selected primary and keeps the reasoning level that was chosen. Only an explicit `/model` selection or model cycle cancels the restore ([#2170](https://github.com/bastani-inc/atomic/issues/2170)).
+- Fixed gRPC `ResourceExhausted` provider errors, seen from providers such as NVIDIA NIM, bypassing the same-model auto-retry budget. They are now retried like other transient provider failures, matching upstream pi-ai, and still advance a configured fallback chain when the retries are exhausted ([#2170](https://github.com/bastani-inc/atomic/issues/2170)).
+
 ## [0.9.12] - 2026-08-04
 
 Cumulative release of the `0.9.12-alpha.1` prerelease. The summary below covers the user-visible outcome of that work; the per-change detail remains in the prerelease section below.

package/README.md

@@ -15,7 +15,7 @@
 
 Atomic is a minimal terminal coding harness. Adapt Atomic to your workflows, not the other way around, without having to fork and modify Atomic internals. Extend it with TypeScript [Extensions](#extensions), [Skills](#skills), [Prompt Templates](#prompt-templates), and [Themes](#themes). Put your extensions, skills, prompt templates, and themes in [Atomic Packages](#atomic-packages) and share them with others via npm or git.
 
-Atomic ships with powerful defaults plus first-party bundled extensions for workflows, subagents, MCP, web access, and intercom. You can still adapt or replace any workflow by writing extensions, skills, prompt templates, themes, or Atomic packages that match how you work.
+Atomic ships with powerful defaults plus first-party bundled extensions for workflows, subagents, MCP, web access, intercom, and ADHD-friendly response style. You can still adapt or replace any workflow by writing extensions, skills, prompt templates, themes, or Atomic packages that match how you work.
 
 Atomic runs in four modes: interactive, print or JSON, RPC for process integration, and an SDK for embedding in your own apps. See [openclaw/openclaw](https://github.com/openclaw/openclaw) for a real-world SDK integration.
 
@@ -93,12 +93,16 @@ For each built-in provider, Atomic maintains a list of tool-capable models, upda
 - GitHub Copilot
 - OpenRouter (browser PKCE, optional key minting)
 - Kimi Code (device authorization with token refresh)
+- xAI (Grok/X subscription)
+- Radius
 
 **API keys:**
 - Anthropic
+- Ant Ling
 - OpenAI
 - Azure OpenAI
 - DeepSeek
+- NVIDIA NIM
 - Google Gemini
 - Google Vertex
 - Amazon Bedrock
@@ -111,13 +115,22 @@ For each built-in provider, Atomic maintains a list of tool-capable models, upda
 - OpenRouter
 - Vercel AI Gateway
 - ZAI
+- ZAI Coding Plan (China)
 - OpenCode Zen
 - OpenCode Go
+- Radius
 - Hugging Face
 - Fireworks
 - Together AI
+- Baseten
 - Kimi For Coding
 - MiniMax
+- MiniMax (China)
+- Moonshot AI
+- Moonshot AI (China)
+- Qwen Token Plan
+- Qwen Token Plan (Individual)
+- Qwen Token Plan (China)
 - Xiaomi MiMo
 - Xiaomi MiMo Token Plan (China)
 - Xiaomi MiMo Token Plan (Amsterdam)
@@ -125,9 +138,9 @@ For each built-in provider, Atomic maintains a list of tool-capable models, upda
 
 See [docs/providers.md](docs/providers.md) for detailed setup instructions.
 
-Atomic's Pi 0.82.1 model/runtime surface includes Claude Opus 5 on Anthropic and Bedrock, model-capability-aware strict JSON-schema and OpenAI Lark/regex tool sampling, ETag-revalidated cached catalogs, persisted llama.cpp models, and live layered `.pi`/`.atomic` `models.json` reload whenever `/model` opens. A populated `ANTHROPIC_AUTH_TOKEN` makes Anthropic models available/selectable while sending header-only bearer auth—with no synthesized API key or `x-api-key`—across chat and summary/compaction requests. Typed RPC model listings expose constrained-sampling flags through optional `ModelInfo.compat`; capability flags are enforcement claims, so unsupported `require` constraints fail instead of being silently weakened. See [models](docs/models.md), [providers](docs/providers.md), [RPC](docs/rpc.md#get_available_models), and [extensions](docs/extensions.md#constrained-sampling).
+Atomic's Pi 0.84.1 model/runtime surface includes Claude Opus 5 on Anthropic and Bedrock, model-capability-aware strict JSON-schema and OpenAI Lark/regex tool sampling, ETag-revalidated cached catalogs, persisted llama.cpp models, arbitrary OpenAI-compatible `samplingParams`, opt-in vLLM thinking-token budgets through `compat.supportsThinkingTokenBudget`, and finishless-stream stop inference through `compat.supportsFinishReason`. A populated `ANTHROPIC_AUTH_TOKEN` makes Anthropic models available/selectable while sending header-only bearer auth—with no synthesized API key or `x-api-key`—across chat and summary/compaction requests. Typed RPC model listings expose constrained-sampling flags through optional `ModelInfo.compat`; capability flags are enforcement claims, so unsupported `require` constraints fail instead of being silently weakened. Custom `models.json` is read from the active Atomic agent directory only; it is not merged across `.pi` and `.atomic` paths. See [models](docs/models.md), [providers](docs/providers.md), [RPC](docs/rpc.md#get_available_models), and [extensions](docs/extensions.md#constrained-sampling).
 
-**Custom providers & models:** Add providers via `~/.atomic/agent/models.json` (legacy `~/.pi/agent/models.json` also works) if they speak a supported API (OpenAI, Anthropic, Google). For custom APIs or OAuth, use extensions. See [docs/models.md](docs/models.md) and [docs/custom-provider.md](docs/custom-provider.md).
+**Custom providers & models:** Add providers via the active agent directory's `models.json` (normally `~/.atomic/agent/models.json`; `ATOMIC_CODING_AGENT_DIR`/`PI_CODING_AGENT_DIR` can select another directory) if they speak a supported API (OpenAI, Anthropic, Google). For custom APIs or OAuth, use extensions. See [docs/models.md](docs/models.md) and [docs/custom-provider.md](docs/custom-provider.md).
 
 ---
 
@@ -185,7 +198,7 @@ Type `/` in the editor to trigger commands. [Extensions](#extensions) can regist
 
 ### Keyboard Shortcuts
 
-See `/hotkeys` for the full list. Customize via `~/.atomic/agent/keybindings.json` (legacy `~/.pi/agent/keybindings.json` also works). See [docs/keybindings.md](docs/keybindings.md).
+See `/hotkeys` for the full list. Customize via the active agent directory's `keybindings.json` (normally `~/.atomic/agent/keybindings.json`). See [docs/keybindings.md](docs/keybindings.md).
 
 **Commonly used:**
 
@@ -213,7 +226,7 @@ Submit messages while the agent is working:
 
 On Windows Terminal, `ALT+Enter` is fullscreen by default. Remap it in [docs/terminal-setup.md](docs/terminal-setup.md) so Atomic can receive the follow-up shortcut.
 
-Configure delivery in [settings](docs/settings.md): `steeringMode` and `followUpMode` can be `"one-at-a-time"` (default, waits for response) or `"all"` (delivers all queued at once). `transport` selects provider transport preference (`"sse"`, `"websocket"`, or `"auto"`) for providers that support multiple transports.
+Configure delivery in [settings](docs/settings.md): `steeringMode` and `followUpMode` can be `"one-at-a-time"` (default, waits for response) or `"all"` (delivers all queued at once). `transport` selects provider transport preference (`"sse"`, `"websocket"`, `"websocket-cached"`, or `"auto"`) for providers that support multiple transports.
 
 ---
 
@@ -289,14 +302,14 @@ Use `--offline` or `ATOMIC_OFFLINE=1` (`PI_OFFLINE=1` remains a legacy alias) to
 
 ## Context Files
 
-Atomic loads `AGENTS.md` (or `CLAUDE.md`) at startup from:
-- `~/.atomic/agent/AGENTS.md` (global; legacy `~/.pi/agent/AGENTS.md` also works)
+Atomic loads `AGENTS.override.md`, `AGENTS.md`, or `CLAUDE.md` at startup from:
+- `~/.atomic/agent/` (global; legacy `~/.pi/agent/` also works)
 - Parent directories (walking up from cwd)
 - Current directory
 
-Use for project instructions, conventions, common commands. All matching files are concatenated.
+An `AGENTS.override.md` file replaces that directory's `AGENTS.md` or `CLAUDE.md`; context from other directories is still concatenated.
 
-Disable context file loading with `--no-context-files` (or `-nc`).
+Use for project instructions, conventions, common commands. Disable context file loading with `--no-context-files` (or `-nc`).
 
 ### System Prompt
 
@@ -460,11 +473,11 @@ See [docs/rpc.md](docs/rpc.md) for the protocol.
 
 ## Philosophy
 
-Atomic is aggressively extensible so it does not have to dictate your workflow. The distribution bundles first-party extensions for workflows, subagents, MCP, web access, and intercom, while still letting you build or install alternative [extensions](#extensions), [skills](#skills), and [Atomic packages](#atomic-packages). This keeps Atomic adaptable without forcing every workflow into the core CLI.
+Atomic is aggressively extensible so it does not have to dictate your workflow. The distribution bundles first-party extensions for workflows, subagents, MCP, web access, intercom, and ADHD-friendly response style, while still letting you build or install alternative [extensions](#extensions), [skills](#skills), and [Atomic packages](#atomic-packages). This keeps Atomic adaptable without forcing every workflow into the core CLI.
 
 **Bundled subagents and workflows.** Atomic includes the first-party subagents and workflows packages. Use them as shipped, customize them, or replace them with your own extensions if your workflow needs a different orchestration model.
 
-**Bundled MCP, web access, and intercom.** Atomic ships first-party integrations for MCP tools, web/repository/document/video extraction, and cross-session coordination. These are packaged extensions rather than hard-coded agent behavior, so they remain configurable and replaceable.
+**Bundled MCP, web access, intercom, and ADHD-friendly output.** Atomic ships first-party integrations for MCP tools, web/repository/document/video extraction, cross-session coordination, and concise action-first responses. These are packaged extensions rather than hard-coded agent behavior, so they remain configurable and replaceable.
 
 **Security stays explicit.** Run in a container, use project trust controls, or build confirmation flows with [extensions](#extensions) inline with your environment and security requirements.
 
@@ -503,7 +516,7 @@ atomic config                        # Enable/disable package resources
 
 | Flag | Description |
 |------|-------------|
-| (default) | Interactive mode |
+| (default) | Interactive mode (fullscreen TUI) |
 | `-p`, `--print` | Print response and exit |
 | `--mode json` | Output all events as JSON lines (see [docs/json.md](docs/json.md)) |
 | `--mode rpc` | RPC mode for process integration (see [docs/rpc.md](docs/rpc.md)) |
@@ -564,14 +577,14 @@ Project trust gates `.atomic`/legacy `.pi` project resources, project package se
 | Option | Description |
 |--------|-------------|
 | `-e`, `--extension <source>` | Load extension from path, npm, or git (repeatable) |
-| `--no-extensions` | Disable extension discovery |
+| `--no-extensions`, `-ne` | Disable extension discovery |
 | `--skill <path>` | Load skill (repeatable) |
-| `--no-skills` | Disable skill discovery |
+| `--no-skills`, `-ns` | Disable skill discovery |
 | `--prompt-template <path>` | Load prompt template (repeatable) |
-| `--no-prompt-templates` | Disable prompt template discovery |
+| `--no-prompt-templates`, `-np` | Disable prompt template discovery |
 | `--theme <path>` | Load theme (repeatable) |
 | `--no-themes` | Disable theme discovery |
-| `--no-context-files`, `-nc` | Disable AGENTS.md and CLAUDE.md context file discovery |
+| `--no-context-files`, `-nc` | Disable context-file discovery and loading |
 
 Combine `--no-*` with explicit flags to load exactly what you need, ignoring settings.json (e.g., `--no-extensions -e ./my-ext.ts`).
 
@@ -581,6 +594,7 @@ Combine `--no-*` with explicit flags to load exactly what you need, ignoring set
 |--------|-------------|
 | `--system-prompt <text>` | Replace default prompt (context files and skills still appended) |
 | `--append-system-prompt <text>` | Append to system prompt |
+| `--offline` | Disable startup network operations, including update checks, package updates, and telemetry |
 | `--verbose` | Force verbose startup |
 | `-h`, `--help` | Show help |
 | `-v`, `--version` | Show version |
@@ -620,7 +634,7 @@ atomic --model sonnet:high "Solve this complex problem"
 atomic --models "claude-*,gpt-4o"
 
 # Read-only mode
-atomic --tools read,grep,find,ls -p "Review the code"
+atomic --tools read,search,find,ls -p "Review the code"
 
 # High thinking level
 atomic --thinking high "Solve this complex problem"
@@ -630,12 +644,16 @@ atomic --thinking high "Solve this complex problem"
 
 | Variable | Description |
 |----------|-------------|
+| `AI_AGENT` | Set to `atomic` by Atomic's CLI, RPC, and compiled binary entry points, and in every Atomic-owned child-process environment, so generic tooling can attribute those processes correctly. Child environments force this value even when the caller supplies another value; the caller's environment object is not mutated. |
 | `ATOMIC_CODING_AGENT_DIR` | Override config directory (default: `~/.atomic/agent`; `PI_CODING_AGENT_DIR` is a legacy alias) |
 | `ATOMIC_CODING_AGENT_SESSION_DIR` | Override session storage directory (overridden by `--session-dir`; `PI_CODING_AGENT_SESSION_DIR` is a legacy alias) |
 | `ATOMIC_PACKAGE_DIR` | Override package directory (useful for Nix/Guix where store paths tokenize poorly; `PI_PACKAGE_DIR` is a legacy alias) |
 | `ATOMIC_OFFLINE` | Disable startup network operations, including update checks, package update checks, and install/update telemetry (`PI_OFFLINE` is a legacy alias) |
 | `ATOMIC_SKIP_VERSION_CHECK` | Skip the Atomic version update check at startup. `PI_SKIP_VERSION_CHECK` is a legacy alias. |
 | `ATOMIC_TELEMETRY` | Override install/update telemetry. Use `1`/`true`/`yes` to enable or `0`/`false`/`no` to disable. This does not disable update checks (`PI_TELEMETRY` is a legacy alias). |
+| `ATOMIC_REDUCED_MOTION` | Set to `1` to skip startup choreography and use a static working identity |
+| `ATOMIC_NO_PTY` | Set to `1` to disable PTY use for bash commands (`PI_NO_PTY` is a legacy alias) |
+| `NODE_COMPILE_CACHE` | Override Node's persistent compile-cache directory; set `NODE_DISABLE_COMPILE_CACHE=1` to opt out |
 | `PI_CACHE_RETENTION` | Provider/upstream-specific prompt-cache retention knob; set to `long` where supported. This is not an `ATOMIC_*` alias and has no Atomic-prefixed equivalent. |
 | `VISUAL`, `EDITOR` | External editor for CTRL+G |
 

package/dist/builtin/i-have-adhd/CHANGELOG.md

@@ -0,0 +1,17 @@
+# Changelog
+
+All notable changes to the `@bastani/i-have-adhd` extension will be documented in this file.
+
+## [Unreleased]
+
+## [0.9.13-alpha.2] - 2026-08-12
+
+### Added
+
+- Added the upstream `i-have-adhd` skill and Atomic extension as a bundled first-party package.
+- Added `/i-have-adhd [on|off]`, the `--no-adhd` startup flag, the `.i-have-adhd-off` agent-directory flag file, and the `stop adhd mode` / `normal mode` stop phrases.
+
+### Changed
+
+- Enabled ADHD-friendly output by default for new sessions while preserving saved per-session state across restarts, branches, and compaction.
+- Renamed the footer status label from `ADHD ON` to `ADHD Mode`; the status is only shown while the mode is on, so the `ON` suffix was redundant.

package/dist/builtin/i-have-adhd/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Ayoub Ghriss
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/dist/builtin/i-have-adhd/README.md

@@ -0,0 +1,18 @@
+# i-have-adhd
+
+Atomic bundles this first-party extension from [ayghri/i-have-adhd](https://github.com/ayghri/i-have-adhd). It shapes responses for an ADHD-friendly reading flow: action first, numbered steps, visible progress, and no filler.
+
+## Atomic use
+
+ADHD mode is enabled by default for new sessions. Its rules are injected once as a hidden message and remain active across turns, branches, restarts, and compaction.
+
+- Use `/i-have-adhd on` or `/i-have-adhd off` to set the mode for the session.
+- Start a session with `--no-adhd` to disable the default.
+- Create `.i-have-adhd-off` in the active agent directory to keep the mode off by default.
+- Say `stop adhd mode` or `normal mode` to turn it off and confirm the change.
+
+The saved session setting wins over the default on later resumes. The full rules live in [`skills/i-have-adhd/SKILL.md`](./skills/i-have-adhd/SKILL.md).
+
+## License
+
+MIT. See [`LICENSE`](./LICENSE).