* feat(native-chat): answer /context in chat view for Claude
Over a terminal-backed chat, Claude's /context paints a grid only the hidden
terminal sees and records nothing in the transcript, so the chat showed
nothing. The chat host now answers it itself: the transcript decoder keeps
each assistant response's API usage and model, and the composer replies with
the same estimate the CLI statusline shows (input + cache tokens against the
model's window) instead of typing the command into the PTY. The command is
also listed in the curated Claude catalog so the slash menu offers it.
* fix(native-chat): only state a /context window the host can establish
The /context answer sized the window from the transcript's model id, which
never carries the [1m] suffix, so a 1M session read as 200k (e.g. 450k / 200k,
225%). The window now comes from the session's tracked model resolved through
the host CLI's own model listing (the resolvedModel field the list_models probe
already returned), and only a [1m] id that matches the model the transcript
last answered with is sized; anything else reports the used figure alone.
Also:
- Offer /context only in the desktop terminal-backed composer; the shared
catalog is mobile's too, and mobile cannot answer it.
- Answer a typed /context even with images attached, as the picker does, and
keep the attachments armed instead of sending them to the terminal.
- Report nothing until a response lands after a /compact or /clear sent from
the chat, instead of the pre-compaction size.
* test(native-chat): pin resolvedModel through Claude model discovery
* fix(native-chat): say /context is unavailable when the host never reports usage
A host that predates usage decoding, or a scraped view, decodes Claude's
replies without a model or usage, so the chat promised an answer after the
next response that would never come. Once the agent has answered and no
answer names its model, say usage is not available for this session.
* feat(native-chat): answer /context for OMP instead of Claude
Claude and Codex answer /context themselves in the structured chat lane, so
the terminal-backed Claude answer, its [1m] window rule, and the resolved-model
plumbing through the model listing go away.
OMP's /context only paints a panel inside its TUI. The OMP decoder now keeps
the provider, model and usage of each reply (none for aborted or errored
replies, which OMP's own gauge skips) and turns compaction rows into the
existing compaction boundary. The OMP model listing keeps each model's
context window, and the chat answers /context from the last reply's prompt
size against the window of the provider/model that served it.
* feat(native-chat): show OpenClaude's own /context report in chat
OpenClaude writes its /context grid to the transcript as a
<local-command-stdout> row, which the chat hid as harness noise. The live
message preparation now turns the stdout row that answers an opted-in command
(/context for OpenClaude) into command output with the ANSI colors stripped,
and the notice row renders command output in monospace so the grid keeps its
columns. Replies to other local commands stay hidden, and Claude is not opted
in. OpenClaude's slash menu now lists /context; Claude's does not.
* test(native-chat): pin OpenClaude /context output through live message preparation
* test(native-chat): pin the OMP context window through model discovery
* test(native-chat): pin monospace rendering of command output
* test(native-chat): restore the notice row suite the command-output test replaced
The previous commit rewrote NativeChatNoticeRow.test.tsx wholesale, deleting
main's compaction, tone, plan, provider-notice and old-reader coverage. Restore
it and pin monospace command output through the message row instead.
* fix(mobile): keep OpenClaude /context off mobile, which cannot show its report
The shared slash catalog now lists /context for OpenClaude because the desktop
chat surfaces its transcript reply. Mobile's transcript fold does not, so the
command did nothing there. Mobile now offers the catalog minus commands answered
only by a surfaced reply, which returns its OpenClaude menu and send
classification to main's behavior.
* refactor(native-chat): drop the unread estimated flag from context usage
Left from the earlier design; the answer text states the estimate itself.
* fix(native-chat): keep /model replies hidden after an OpenClaude /context
Skill surfacing ran first and rewrote non-catalog envelopes such as /model into
plain user text, so command-output pairing no longer saw them. A later /model
reply then paired with the earlier /context envelope and appeared in the chat.
Pair outputs against the raw envelopes before skill surfacing.
* fix(native-chat): word the OMP /context answer for a compacted session and show it as an aside
After a compaction the agent has already answered, so the unknown-usage
line now promises the next response instead. The host answer is one
sentence, so it draws as the muted line it replaces rather than a
monospace grid. The provider/model selector now comes from one helper
shared by the OMP listing parser and the answer, and new tests drive
real OpenClaude and OMP transcript lines through the decoders.
Co-Authored-By: Claude <noreply@anthropic.com>
* refactor(native-chat): pair command replies by row link and declare /context replies on the catalog
A `<local-command-stdout>` row now answers the command row its `parentUuid`
names, carried on decoded messages as the optional `parentId`, instead of the
newest command row at or before its timestamp. A reply with no loaded linked
row (older host, command outside the tail window) stays hidden.
How each agent's `/context` is answered is declared once, as `reply` on its
catalog row (`composer` for OMP, `transcript` for OpenClaude). The slash menu,
the composer's answer, the transcript surfacing and the mobile catalog all
derive from it, and both send paths share one intercept. Mobile no longer lists
OMP `/context`, which did nothing there.
* perf(native-chat): skip command-output surfacing for agents that declare no transcript reply
Every Claude-format row now carries its parent link, so once a session holds any
local-command reply (/model, /compact) the surfacing pass built a map of every
message on each live update and changed nothing for agents like Claude. Agents
whose catalog declares no transcript reply now return before the scan.
---------
Co-authored-by: Claude <noreply@anthropic.com>
Sixth and final wave over the modules that export symbols only tests import.
Deletes private-predicate cases whose behavior is already asserted through the
module's real entry point, then makes the symbol private again.
Also removes three distinct junk shapes the earlier detectors missed:
- a self-comparison whose expected empty row was produced by the helper under
test (`worktree-palette-search`), now a literal;
- expected values computed by a sibling helper rather than asserted
(`terminal-theme`), now read through the production `getBuiltinTheme`;
- a negative control that cannot fail — `expect('json' in jsonlMonarchLanguage)
.toBe(false)`, where `IMonarchLanguage` has no such key, so it guarded nothing
while appearing to guard "does not attach the JSON language service".
Dead production code removed where tests were its only callers:
`refreshWindowsTerminalCapabilities` (a one-line alias for
`loadWindowsTerminalCapabilities({force: true})`), `readSpoolRecords`,
`buildAgentPromptSubmitBytes`, and `getCommitMessageModelCapability`.
About 70% of everything this detector flagged across the whole vein was a false
positive, so most modules were left untouched. Bounds consumed as test input,
`*ForTests` seams, non-hook cores of `useSyncExternalStore` hooks, and
value-position registrations all look identical to a leaked internal from the
outside and are not.
* feat(source-control-ai): support OMP text generation
Read prompts on stdin, retain OMP configured model by default, and reuse JSON model discovery.
Co-authored-by: unknown <1784931579@qq.com>
* test(source-control-ai): cover OMP large input and model overrides
* fix(omp): keep configured model default out of discovered catalog
* fix(omp): hide config default from model discovery catalog
* fix(omp): separate terminal discovery from generation defaults
* test(omp): keep model probe import compatible with CLI typecheck
* test: align Source Control AI registry contracts with OMP
---------
Co-authored-by: unknown <1784931579@qq.com>
* fix(source-control-ai): stop duplicating singleton CLI flags in agent argv
Recipe CLI arguments and agent command overrides were appended on top of
Orca's generated flags, so a user-supplied --model produced a repeated
flag. yargs collapses a repeated flag into an array, crashing OpenCode
with "j.split is not a function"; clap rejects it outright for Codex.
Declare the at-most-once option groups per agent spec and fold every
user-supplied occurrence into the generated slot, with recipe args
outranking a command-override prefix.
Fixes#12305
* fix(source-control-ai): preserve command override argv order
#14397 split `shared/types.ts` into 46 per-domain modules but kept the path as
a re-export barrel so the import sites did not have to change. This removes
the barrel: every consumer now imports from the module that actually declares
the type, and `src/shared/types.ts` is deleted.
Barrels hide where a type lives, make every consumer look like it depends on
the whole domain, and let an unrelated edit invalidate a module that ~2,000
files transitively import.
2,323 import declarations across 2,321 files. Rewritten mechanically: each
specifier was resolved to an absolute path via the TypeScript AST and
recomputed, rather than string-substituted, so alias forms (`@/../../shared/
types`) and per-specifier `type` modifiers survive.
Four cases the mechanical pass had to handle, each found by a gate rather than
by reading the diff:
- Modules inside `src/shared` import the barrel as `./types`, not
`shared/types`. A pre-filter on the latter string skipped 176 of them and
left imports dangling at a deleted file, which surfaced as confusing
`Property 'x' is optional in type 'Repo' but required in Pick<Repo, ...>`
errors rather than "module not found".
- The barrel RENAMED one type on the way through
(`WorkspaceSource as WorkspaceCreateTelemetrySource`), so the original name
in the owning module has to be re-aliased at each consumer.
- Three test files put `;(globalThis as ...)` on the line after the import.
TypeScript parses that `;` as the import statement's terminator, so
replacing through `statement.getEnd()` deletes it and breaks ASI. The
rewrite now stops at the module specifier.
- A file that already imported directly from a module got a SECOND import
from it, because the barrel re-exported those same names — which trips
`import/no-duplicates` under `--deny-warnings`. A post-pass merges
declarations sharing a specifier and type-only-ness; the `import type` plus
`import` pair from one module is left alone, since that form is allowed.
Splitting one barrel import into several genuinely adds lines, which pushed
`terminal-layout-pty-ownership.ts` to 301 counted lines: its 107-character
import must wrap, and neither local type collapses onto one line (101 and 116
characters). Rather than contort a type declaration to fit a line budget,
`collectLeafIds` and `pruneLeaves` move to `terminal-pane-layout-tree.ts` —
they are pure structural operations on the layout tree and independent of PTY
ownership. `visible-worktrees.ts` similarly loses its own mini-barrel
re-export of `isDefaultBranchWorkspace`, with the four real consumers
repointed at the declaring module. No `max-lines` bypass added.
Verified: cold `tsc --noEmit` green on node, cli, and web (buildinfo deleted
first — these projects are `composite: true` and reuse stale caches); the full
`pnpm lint` green, not just bare oxlint — the narrower local check is what let
the duplicate imports reach CI; max-lines ratchet OK at 344.
* fix(commit-message): use Kimi --prompt instead of Claude --print
kimi-code rejects --print (suggesting --prompt). Deliver the generation
prompt as the --prompt argv value so branch auto-rename and commit
message generation work when Kimi is the selected agent.
Fixes#11669
* test(commit-message): cover Kimi argument defaults
* feat(native-chat): add model and effort pickers for grok
Grok had no session-option catalog, so the native chat composer showed no
pills and every launch ran the CLI's own defaults with no way to change them.
Adds a `GROK_SESSION_OPTION_CATALOG` (model via `-m`/`/model`, reasoning
effort via `--reasoning-effort`/`/effort`) and the discovery plumbing behind
it. Grok's selectable ids depend on the signed-in account and on `[model.*]`
config, so the seed carries only `grok-4.5` and a runtime `grok models` probe
supplies the rest as authoritative — a retired id must be droppable, since
launching one is a fatal exit rather than a warning.
Because `grok models` publishes `Default model:` and marks the row
`(default)`, the picker can name the model a fresh session is actually
running: `defaultModelIsCliDefault` plus an untracked record means no `-m`
was ever emitted, so the CLI is on its own default. That default scopes the
effort row but is never written to persisted settings — that field is what
authorizes `-m` on every later launch, and adopting a model the user never
picked would pin today's default forever, fatally so on an account without
it. `grok --help` publishes no default for `--reasoning-effort`, so the
effort value stays unnamed until something sets it.
Known gap: that refusal to persist is also a limit. An option set while on
the CLI default is dispatched and honored in-session, but reaches no later
launch — it persists under the default's id with `model` left unset, and
both `resolveNativeChatSessionOptionDefaults` and
`resolveAgentSessionOptionLaunch` bail without that key. Picking a model
explicitly persists normally. Closing this means teaching both to resolve
options from the default model while still refusing to emit `-m`, which is
the launch-args path and wants its own review.
Known gap: the picker infers "no `-m` was emitted" from its own in-memory
record, so a model reaching argv from outside it — the user's own
`agentDefaultArgs`, or a renderer reload that drops the record while the
flagged PTY lives on — leaves the pill claiming the CLI default while
another model runs. No wrong model is persisted.
Extracts `hasFlag` and `labelFromModelId`, and splits the model-probe spec
out of the commit-message registry so discovery no longer implies an agent
can write commit messages.
Co-authored-by: Orca <help@stably.ai>
* docs(native-chat): note the invariant keeping modelIsCliDefault agent-safe
The flag is computed without checking the catalog, so it reads as unsafe for
the four agents with no CLI default. It is safe only because `persist` bails
unless `modelId` is truthy, which for those agents implies a tracked model.
Widening that guard would silently change persistence for every agent.
Co-authored-by: Orca <help@stably.ai>
* fix: retire persisted models on mount and handle -- terminator
- When a pane mounts after model discovery has already settled, it now checks
the cache and retires persisted models that are no longer available.
- CLI flag detection now respects the `--` option terminator, treating
everything after it as positional arguments rather than flags.
* Fix: persist grok session options under probe-confirmed defaults
Options set under the CLI default were silently lost on restart.
Distinguish seed guesses from probe-confirmed defaults by renaming
`modelIsCliDefault` to `modelIsUnverifiedDefault`. Once confirmed,
adopt the default as a persisted flag so options survive restarts.
* fix(native-chat): close the retired-model fatal-launch paths from counsel review
Counsel report C1/C2 (High), C3, P1, C4:
- Untrack a session model an authoritative discovery dropped and gate every
persist path, so option writes can never re-adopt a retired id (C1).
- Resolve launch defaults through the enrichment cache: a persisted model
missing from every settled probe no longer becomes a fatal `-m` (C2).
- Serialize retirement and picks on one settings write queue that re-reads
live state at apply time (C3).
- Stabilize onSwitchToTerminal so the session-option surface is not rebuilt
every TerminalPane render (P1), and cap the enrichment host map (C4).
Co-authored-by: Orca <help@stably.ai>
* Store agent in enrichment entry and extract token utilities
Refactor enrichment to store the agent field directly instead of
parsing it from a composite key, and extract CLI flag token filtering
into a shared utility. Use a dedicated function for tracked model ID
lookup. Improves code reuse and reduces parsing overhead.
* Rename modelIsUnverifiedDefault to adoptModelAsLaunchDefault
Move the model adoption gate into the core session-options module, where probe confirmation and discovered-model status are known. This ensures adoption decisions are gate-checked before persisting to avoid fatal launch flags, and simplifies the picker surface by moving the logic to where it belongs.
* Keep model probe evidence by agent, not host
Store probed model IDs in agent-keyed cache independent of host cache, so
evidence persists across host eviction. Prevents retired models from being
treated as valid when host cache entries are evicted.
* Store agent in enrichment entries instead of separate proof-evidence map
Model probe evidence is now tied to enrichment entries rather than maintained in a separate per-agent map, eliminating the need for eviction logic that could disconnect proof from entries.
---------
Co-authored-by: Orca <help@stably.ai>
* feat(native-chat): track Claude models from the installed CLI per host (STA-3330)
The Claude seed no longer pins version labels to aliases that resolve
differently across CLI versions, and the catalog now defines listModels
backed by a one-shot list_models control request over --print stream-json.
Hosts whose CLI predates the request answer with a control error and keep
the seed. Discovery also feeds Source Control AI via the commit-message
spec, and the /model echo detector matches resolved model names.
* fix(native-chat): preserve discovered Claude capabilities
* fix(native-chat): tolerate malformed Claude model entries
* fix(native-chat): discover models in folder workspaces
* fix(native-chat): trust discovered Claude capabilities
* fix(native-chat): remove Claude model fallbacks
* fix(native-chat): keep the Claude model picker rendered
The Claude picker rendered nothing until the per-host `list_models` probe
returned, so it popped in ~1s after mount and never appeared at all when
the probe failed — an old CLI without `list_models`, no `claude` on PATH,
or an older remote runtime whose response omits `catalogOrigin`.
Restore the version-neutral family seed as the starting list; discovery
still replaces it wholesale on success, so a host with a real catalog
never shows an obsolete hardcoded row.
Separately, the tracked model could fall outside the active list: the
terminal header scrape yields family ids (`opus`) while a current CLI
lists `opus[1m]` and no plain `opus`. That blanked the picker trigger and
dropped the model's effort and fast-mode controls. Reconcile the tracked
id into the active list once, so the snapshot, the appliers, and typed
command recording all see a labelled, operable row for it.
* Add agent enablement controls
Implements the Enable/Disable Agents Dashboard behavior described in docs/enable-disable-agents-dashboard.md, including persisted agent enablement state, filtered launch surfaces, and settings UI affordances.
* Clarify agent availability controls
* Respect disabled agents across workspace and AI defaults
- Filter disabled TUI agents from mobile and quick workspace selection
- Avoid implicitly choosing disabled agents for commit/PR AI settings
- Broadcast settings changes to open windows for disabled-agent updates
* rm design doc
* fix: complete agent enablement propagation
Brennan BensonandClaude
Neilandczzczz
Neil
BingZ
JinjingandOrca
Jinwoo Hong
