fix(native-chat): plain wording for chat errors and status rows

[brennanb2025]

	Files	Added	Deleted	Net
Test	20	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​120	$\color{#cf222e}{\Huge{\mathbf{−}}}$​84	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​36
Prod	31	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​285	$\color{#cf222e}{\Huge{\mathbf{−}}}$​216	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​69

ELI5

When something goes wrong in the native chat, the message that explains it often used Orca's internal words: "2 journal lines could not be read", "[Orca: item truncated — 3145728 bytes exceeds the history page budget]", "The Orca running this chat has no record of it", a subagent marked "unverifiable · no contact", "Skills are unavailable for this host". Nobody reading the chat knows what a journal, a host, a page budget or "unverifiable" is, so these messages don't say what happened or what to do. Some were also wrong: "Update Orca" was offered for refusals that updating can't fix. Two errors above the composer were hard-coded English, and the two rows the host writes into a chat could never be translated.

This PR rewrites those messages in plain words that say what happened and, where there is one, what to do. It removes one line that only described Orca retrying in the background. It makes every one of these messages show in the reader's language.

What Changed

Before → after, as the user reads it:

Where it shows	Before	After
A chat action the computer running the chat doesn't recognise at all (that computer has an older Orca)	The Orca running this chat doesn't support this. Update Orca, then try again.	This needs a newer Orca on the computer running this chat. Update Orca there, then try again.
A chat action refused for a named reason that updating doesn't fix (the agent or location can't run there, chats are turned off there, or this app lacks a capability)	Same sentence as above, including "Update Orca"	This isn't available in this chat. (When the action was opening the chat's history: "This chat's history couldn't be loaded.")
A chat action refused because the previous agent might not have exited	Orca hasn't confirmed that this chat's previous agent stopped.	The previous agent in this chat may still be running.
A chat the computer running it has no record of	The Orca running this chat has no record of it.	Orca can't find this chat.
A send refused because an earlier message isn't settled	A message you sent earlier hasn't been confirmed yet.	A message you sent earlier isn't confirmed yet.
A write Orca couldn't save to the chat's history	Orca couldn't record it in this chat's history.	Orca couldn't save this to the chat's history.
Row shown while queued messages wait for the previous agent to exit	Orca couldn't confirm Claude's previous process ended. Messages wait to be sent until Orca confirms it has ended.	Claude from before may still be running. Your messages will send once it stops.
Row added after Orca repairs a damaged chat history	2 journal lines could not be read	Part of this chat's history couldn't be loaded. (now translated)
Row replacing one history item too big to send	[Orca: item truncated — N bytes exceeds the history page budget]	This part of the chat was too large to show. (now translated)
Skills menu (/ picker) errors	Skills are unavailable for this host (SSH chats only) / Could not load skills from this host	Skills aren't available in SSH chats / Couldn't load skills. The screen-reader text now says the same sentence as the screen.
A subagent or background task Orca has lost sight of	unverifiable · no contact / 2 unverifiable / 2 agents unverifiable · no contact	no recent update / 2 with no recent update / 2 agents with no recent update
A chat whose history failed to load while Orca keeps retrying	Could not load conversation + "The transcript could not be read. Orca keeps trying to load it."	Could not load conversation (the retry line is gone; Orca still retries)
Above the composer, when Orca won't resend a message it couldn't confirm	Message delivery is unconfirmed and Orca will not send it again (English only)	Orca couldn't confirm your message reached the agent. Check the chat, then send it again if needed. (translated)
Above the composer, when the message couldn't be saved locally	Message could not be saved to the outbox (English only)	Couldn't save your message. Try again. (translated)

How it works:

• Refusals. The sentences stay in the shared notice tables in src/shared/, which desktop translates and the phone shows in English. A refusal that names a reason now gets the new sentence notAvailable instead of the "update Orca" sentence (agent-session-refusal-notice.ts, the reason table). The "update Orca" sentence is kept only for a refusal with no reason. Every current host names a reason, so a refusal without one means the host didn't recognise the request (method_not_found), which proves it is older. For opening a chat's history, "This isn't available" would name nothing the reader asked for, so that case says the history couldn't be loaded.
• The two rows the host writes. The repair row and the too-large row now carry a presentation value (history-repaired, history-item-too-large), the same field the "Context compacted" row already uses. The desktop words them with translate() (NativeChatNoticeRow.tsx). Their English text stays as written for older desktop builds and the phone. Both lost the row count or byte size they used to print, which meant nothing to the reader. Nothing parses these texts: the repair row is identified by its fixed item id. The copy and the presentation values live in the new src/shared/agent-session-host-status-rows.ts.
• Composer errors. The send-outcome function in src/shared/structured-agent-session-send-disposition.ts used to return a hard-coded English sentence. It now returns sentence keys from the same table every other chat notice uses, and the desktop translates them with the existing agentSessionWriteNoticeText. The four places that report a failed local save use one shared pair of sentences instead of four copies of an English literal.
• The retrying load error. The line under the title only described Orca's own background retry, so it is removed rather than reworded. The pane shows its title, or the specific headline when there is one, and never falls back to the terminal pane's "Toggle back to the terminal" line.
• Lost-contact subagents and tasks. These now use the sidebar's existing words for this state, "no recent update", everywhere. After a count they read "2 with no recent update". The separate "no contact" reason only repeated the state, so it is gone. The internal state vocabulary (live / unverifiable / exited) is untouched.
• The waiting row's promise is true. Every 5 minutes the idle sweep retries the stop it couldn't prove (structured-agent-session-idle-sweep.ts). A new message, an option change or a Stop also retries it. Whichever retry proves the old agent exited sends the waiting messages (releaseLease → wakeDelivery in structured-agent-session-host-lifetime.ts). structured-agent-session-claude-unproven-stop-send.test.ts covers the sweep-driven send. The row is written only while a message is queued.
• Left to sibling PRs: the agent-starting refusal, the launch-failure toast, the loading pane, resume and restart toasts, the approval card, and the failed-send rows.

Why

These messages show up exactly when something has gone wrong, which is when the reader most needs to know what happened and what to do. Internal words help with neither. The new copy follows the common pattern for chat apps: short, says what happened ("Couldn't load skills", "This part of the chat was too large to show"), and gives a step only where one exists and would work ("Check the chat, then send it again if needed").

Rejected options:

• Remove the repair and too-large rows. Each is the only sign that part of the conversation isn't shown. Removing them would hide that.
• Keep the counts and byte sizes. They give the reader nothing to act on.
• Keep English text in the host rows and composer errors. A sentence written on the host, or returned as a finished string from shared code, can't follow the reader's language. A typed value that each screen words is the same contract the compaction row and the refusal notices already use.
• Keep the retry line and only reword it. It reports background work the reader can't act on, and the common pattern shows only that loading failed.

Wire compatibility: no new field, kind or opcode. presentation already exists as an open string that older readers ignore. A newer host talking to an older desktop build shows the two rows' English text, in normal text colour instead of muted grey. The phone always shows the English text. An older host keeps sending its old English rows, which a newer client shows as written. The too-large row is built on every history read, so it changes as soon as the host updates. The repair row is rewritten each time the chat is opened while the damage remains.

Differences from the common pattern:

• Intended: a subagent or task Orca has lost sight of has its own state, "no recent update". The common pattern has no word for this: its group summary counts any unknown status as stopped. Orca keeps the state separate because losing contact with an agent, for example over SSH, must never be reported as the agent exiting.
• Intended: the "update Orca" step depends on whether the refusal names a reason. The common pattern proves an older remote with a capability or version check before saying "update". Here the proof is that the host didn't recognise the request. One imperfect case remains: a host released between 2026-08-28 and 2026-09-28 named no reasons yet. If its real cause was a location it can't run (for example WSL), it gets the "update" sentence; after the update it gives the accurate "isn't available" sentence.
• Intended: the host rows are a typed value with an English fallback, rather than text the client builds alone. This is so older clients and the phone, which have no wording for a new value, still show a sentence.
• Intended: the "couldn't confirm your message reached the agent" error exists because Orca refuses to send a message twice when it can't tell whether the first send arrived. The common pattern has no such guard, so it has no such message. This PR only rewords it.
• Temporary: an item too large for a history page is replaced whole by one row. The common pattern truncates inside the item and keeps the start visible. Follow-up: truncate in place.
• Temporary: the waiting reason is a row in the transcript rather than a line on the queued message, and it stays in the transcript after the messages send. Follow-up: move it onto the queued message once the queued-messages capability ships.

Linked Issue

N/A (internal native-chat UI cleanup)

Visual Proof

Captured on Windows in two hidden, isolated test builds, with a stand-in agent in place of Claude. Before = base 306b4578aaa. After = this PR's final head 5ffbbc1c44d. These replace the earlier set, which was taken before the review fixes, on a Mac, at 13e47527fcb.

• Damaged chat history. One saved history line was corrupted on purpose, then Orca was restarted. Before: "1 journal line could not be read". After: "Part of this chat's history couldn't be loaded." The red "Claude stopped while this response was in progress" row appears in both builds. It comes from the damaged line itself, not from this PR.
  
  
• The same two rows with Orca set to Japanese (new since the earlier set). Before, both rows stay in English because the host writes them as finished text. After, both are translated: "このチャットの履歴の一部を読み込めませんでした。" and "チャットのこの部分は大きすぎるため表示できません。".
  
  
  
  
• A subagent Orca lost sight of across a restart (new since the earlier set). Before: "Ran 1 subagent unverifiable". After: "Ran 1 subagent no recent update".
  
  
• An answer too large to show, in English (a 2.6 MB reply). Kept from the earlier set, because the English text did not change after it. The final head was re-checked on Windows by reading the on-screen text: before "[Orca: item truncated — 2600434 bytes exceeds the history page budget]", after "This part of the chat was too large to show." A fresh English screenshot timed out.
  
  

The reworded rows keep the same look: 14 px muted grey in both builds, measured from the rendered page. The after build showed no new pop-up, toast, banner or status line.

Not captured:

• Skills-menu errors. One needs an SSH chat. The other needs skill loading itself to fail, and the test build offered no way to force that.
• The retrying load-error pane. Same reason: the test build offered no way to force the failure.
• The composer's "Couldn't save your message". Filling the app's storage did not make a send fail.
• The refusal notices, the waiting row and "couldn't confirm your message reached the agent". The stand-in agent can't produce these states on demand.

Each of these is covered by a rendered component test.

Testing

• I manually tested these changes locally

• Automated tests added/updated, or explained why not below

• Updated every test that asserted the old copy, and added tests for the new behaviour:

  • a refusal with a named reason says "isn't available" for every action except opening history;
  • a refused history read on a host that can't run the chat says the history couldn't load;
  • the two host rows are worded through translation;
  • the skills-menu screen-reader text in SSH chats.

• The targeted vitest files pass on the final head, including the mobile suites that use the shared refusal wording.

• Localization catalog, extraction and coverage checks pass.

• oxlint, the changed-code quality gate and React Doctor report no findings.

• Web, node and CLI typechecks pass.

• A trial merge in the planned order (main, then fix(native-chat): show "Sending…" on a message until the host confirms it #24606, then fix(native-chat): queued messages carry on in order after any turn, and nothing sends by itself after a restart #24586, then this PR) is clean at every step.

AI Disclosure

Review

Agent skill upstream boundary

• Not applicable, or this change follows docs/reference/agent-skill-sharing-upstream-boundary.md and copies or mechanically translates no upstream skill-installer source, tests, fixtures, registry entries, path tables, comments, or documentation.

Notes

• SSH: the "newer Orca" sentence names "the computer running this chat", which is the remote machine for an SSH chat and this computer for a local one.
• Phone: it shows the shared English sentences as written.
• Merge with fix(native-chat): show "Sending…" on a message until the host confirms it #24606: this PR adds the sendOutcomeLost sentence. Its follow-up for the failed-send rows should reuse that sentence rather than add its own.

Checklist

• This PR is small and focused
• I explained what changed and why (ELI5, the user-facing before/after, the mechanism, and why over the alternatives)
• Before/after screenshots or videos attached for UI changes, or N/A with reason
• Self-reviewed for correctness, security, and performance
• Cross-platform, SSH/remote, and path/shortcut impact considered (or N/A)
• pnpm lint, pnpm typecheck, pnpm test, and pnpm build pass (or CI will cover; local preferred)

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: ea39c5f3-5975-4cc9-ba30-4149fef68a6d

📥 Commits

Reviewing files that changed from the base of the PR and between a5613b9 and 5ffbbc1.

📒 Files selected for processing (3)

• src/renderer/src/components/native-chat/NativeChatStructuredSession.read-error.test.tsx
• src/shared/agent-session-refusal-notice.test.ts
• src/shared/agent-session-refusal-notice.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 7 remain after this review.

---

📝 Walkthrough

Walkthrough

The changes standardize journal-repair and oversized-history status rows, update structured-session send and save notices, and remove retry-status text from chat error states. Unverifiable background tasks now use “no recent update” labels and counted forms. Skill errors share text between visible status and screen-reader announcements. English and five other locales update related chat, history, skills, and task-status copy.

Priority: ➖ Normal

Merge Risk: 🔵 Low · up to 5ffbb

Some older mobile clients may be refused structured-chat features without being told that updating the client could help. The impact is limited to recovery guidance, so the change is mergeable with that issue in mind.

Architecture Summary

Architecture risk: 🔵 Low · up to 5ffbb

The change affects 2 systems.

Changed systems: src, mobile

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — src (service) was modified; 49 changed files map to changed impact.
• observed — mobile (service) was modified; 1 changed file maps to changed impact.

Before / after behavior

• observed — Modified behavior in mobile/src/session/mobile-structured-operation-id-retirement.test.ts: The expected failure message now specifies that the user should update Orca on the computer running the chat before trying again; the previous expectation used a generic unsupported-version message.
• observed — Modified behavior in src/main/native-chat/agent-session-journal/journal-store-open.ts: The repair disclosure is now created without supplying input.malformedRows(); the existing malformed-row and read-only checks still govern whether it is appended.
• observed — Modified behavior in src/main/native-chat/agent-session-journal/journal-store-schema.test.ts: Adds the journalRepairDisclosure import.
• observed — Modified behavior in src/main/native-chat/agent-session-journal/journal-store-schema.test.ts: Defines REPAIR_TEXT from the disclosure body text for use in test expectations.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.81% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 31 functions across 44 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly identifies the native-chat wording changes and status rows, which are the main focus of the pull request.
Description check	✅ Passed	The description is detailed and covers the change, rationale, testing, visual proof, compatibility, and checklist. It marks the linked issue as not applicable for an internal cleanup and leaves the AI…

• Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Commit to this branch
• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[pullfrog]

✅ No new issues found.

Reviewed changes — read the full 42-file diff end to end; verified the type refactor, the copied/removed literals, and ran the changed test files (228 tests passing).

• Plain-language copy — internal wording ("journal lines could not be read", "item truncated", "no contact", "unverifiable") rewritten across the shared notice/failure tables, the two host-written status rows, and the renderer fallbacks.
• Typed send disposition — StructuredAgentSessionSendDisposition.error is now 'redeliveryRefused' | null instead of a free string; the renderer words it in the new structured-agent-session-outbox-error-text.ts, which also centralizes the four duplicated "couldn't save" literals.
• State wording unified — unverifiable now reads "no recent update", with a new backgroundTaskStateCountWord giving "status unavailable" after a count; reasonUnverifiable and the row-count/byte-size parameters are removed.
• i18n — en.json, en-runtime-required.json, and es/fr/ja/ko/zh updated for every changed or new key.

Verification notes: the disposition error change is internal (mobile has its own delivery path and never reads disposition.error), detail.kind === 'reason' is only ever waiting | blocked | unverifiable so the new backgroundTaskStateWord fallback fires only for unverifiable, no stale old literals remain repo-wide, and the exact repair/oversized literals stay pinned in journal-corruption-repair.test.ts and agent-session-history-page.test.ts.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Advanced

Run ID: 310d6856-a74d-4ef0-bb80-57be1c8b15f7

📥 Commits

Reviewing files that changed from the base of the PR and between 15e4599 and a5613b9.

📒 Files selected for processing (40)

• src/main/native-chat/agent-session-journal/journal-corruption-repair.test.ts
• src/main/native-chat/agent-session-journal/journal-repair-disclosure.ts
• src/main/native-chat/agent-session-wire/agent-session-history-page-bounds.ts
• src/main/native-chat/agent-session-wire/agent-session-history-page-grouping-parity.test.ts
• src/main/native-chat/agent-session-wire/agent-session-history-page.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-claude-unproven-stop-send.test.ts
• src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.test.tsx
• src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.tsx
• src/renderer/src/components/native-chat/NativeChatEmptyState.test.tsx
• src/renderer/src/components/native-chat/NativeChatEmptyState.tsx
• src/renderer/src/components/native-chat/NativeChatNoticeRow.test.tsx
• src/renderer/src/components/native-chat/NativeChatNoticeRow.tsx
• src/renderer/src/components/native-chat/NativeChatStructuredSession.read-error.test.tsx
• src/renderer/src/components/native-chat/NativeChatStructuredSession.tsx
• src/renderer/src/components/native-chat/NativeChatSubagentRun.test.tsx
• src/renderer/src/components/native-chat/NativeChatSubagentRun.tsx
• src/renderer/src/components/native-chat/agent-session-failure-words-text.test.ts
• src/renderer/src/components/native-chat/agent-session-write-notice-text.ts
• src/renderer/src/components/native-chat/background-task-header-content.ts
• src/renderer/src/components/native-chat/background-task-roster.test.ts
• src/renderer/src/components/native-chat/background-task-roster.ts
• src/renderer/src/components/native-chat/structured-agent-session-outbox-dispatch.ts
• src/renderer/src/components/native-chat/structured-agent-session-outbox-retry.ts
• src/renderer/src/components/native-chat/structured-agent-session-read-failure-notice.ts
• src/renderer/src/components/native-chat/use-structured-agent-session-outbox.test.tsx
• src/renderer/src/components/native-chat/use-structured-agent-session-outbox.ts
• src/renderer/src/i18n/en-runtime-required.json
• src/renderer/src/i18n/locales/en.json
• src/renderer/src/i18n/locales/es.json
• src/renderer/src/i18n/locales/fr.json
• src/renderer/src/i18n/locales/ja.json
• src/renderer/src/i18n/locales/ko.json
• src/renderer/src/i18n/locales/zh.json
• src/shared/agent-session-failure-copy.ts
• src/shared/agent-session-host-status-rows.ts
• src/shared/agent-session-refusal-notice.test.ts
• src/shared/agent-session-refusal-notice.ts
• src/shared/agent-session-write-notice-copy.ts
• src/shared/native-chat-empty-state.ts
• src/shared/structured-agent-session-send-disposition.ts

💤 Files with no reviewable changes (1)

• src/shared/native-chat-empty-state.ts

🚧 Files skipped from review as they are similar to previous changes (15)

• src/main/native-chat/agent-session-wire/structured-agent-session-claude-unproven-stop-send.test.ts
• src/shared/agent-session-failure-copy.ts
• src/renderer/src/components/native-chat/NativeChatSubagentRun.tsx
• src/renderer/src/components/native-chat/background-task-roster.test.ts
• src/renderer/src/components/native-chat/structured-agent-session-outbox-dispatch.ts
• src/renderer/src/components/native-chat/structured-agent-session-outbox-retry.ts
• src/renderer/src/components/native-chat/use-structured-agent-session-outbox.ts
• src/renderer/src/i18n/en-runtime-required.json
• src/renderer/src/i18n/locales/ja.json
• src/renderer/src/i18n/locales/zh.json
• src/renderer/src/components/native-chat/use-structured-agent-session-outbox.test.tsx
• src/renderer/src/i18n/locales/ko.json
• src/renderer/src/i18n/locales/en.json
• src/renderer/src/i18n/locales/es.json
• src/renderer/src/i18n/locales/fr.json

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 5 remain after this review.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

🔎 Supported by static analysis

🏁 Script executed:

sed -n '185,250p' src/shared/agent-session-refusal-notice.ts
rg -n 'clientCapabilityMissing|hostUnsupported|notAvailable|hostFinding' src/shared/agent-session-refusal-notice* src/main/native-chat src/renderer/src/components/native-chat | head -100

Repository: stablyai/orca

Length of output: 6392

---

🏁 Script executed:

set -eu
printf '%s\n' '--- diff ---'
git diff --unified=30 306b4578aaa5c55f27add2a75607fd5eb347eca2  a5613b9e77eb017bda87cfaafed8c996d87dcd7a -- src/shared/agent-session-refusal-notice.ts src/shared/agent-session-refusal-notice.test.ts src/renderer/src/components/native-chat/agent-session-write-notice-text.ts
printf '%s\n' '--- notice definitions and rendering ---'
sed -n '1,115p' src/shared/agent-session-refusal-notice.ts
sed -n '230,330p' src/shared/agent-session-refusal-notice.ts
sed -n '1,115p' src/renderer/src/components/native-chat/agent-session-write-notice-text.ts
printf '%s\n' '--- clientCapabilityMissing references ---'
rg -n -C 8 'clientCapabilityMissing' src
printf '%s\n' '--- focused tests ---'
sed -n '280,375p' src/shared/agent-session-refusal-notice.test.ts
rg -n -C 12 'structured_agent_session_unsupported|hostDisabled|hostUnsupported|updateOrca|notAvailable' src/shared/agent-session-refusal-notice.test.ts src/main/native-chat

Repository: stablyai/orca

Length of output: 41576

---

🏁 Script executed:

printf '%s\n' 'pending-check'

Repository: stablyai/orca

Length of output: 165

---

🏁 Script executed:

set -eu
printf '%s\n' '--- diff ---'
git diff --unified=30 306b4578aaa5c55f27add2a75607fd5eb347eca2 a5613b9e77eb017bda87cfaafed8c996d87dcd7a -- src/shared/agent-session-refusal-notice.ts src/shared/agent-session-refusal-notice.test.ts src/renderer/src/components/native-chat/agent-session-write-notice-text.ts
printf '%s\n' '--- notice definitions and rendering ---'
sed -n '1,115p' src/shared/agent-session-refusal-notice.ts
sed -n '230,330p' src/shared/agent-session-refusal-notice.ts
sed -n '1,115p' src/renderer/src/components/native-chat/agent-session-write-notice-text.ts
printf '%s\n' '--- clientCapabilityMissing references ---'
rg -n -C 8 'clientCapabilityMissing' src
printf '%s\n' '--- focused tests ---'
sed -n '280,375p' src/shared/agent-session-refusal-notice.test.ts
rg -n -C 12 'structured_agent_session_unsupported|hostDisabled|hostUnsupported|updateOrca|notAvailable' src/shared/agent-session-refusal-notice.test.ts src/main/native-chat

Repository: stablyai/orca

Length of output: 41853

---

🏁 Script executed:

set -eu
printf '%s\n' '--- exact client capability references ---'
rg -n -C 15 'clientCapabilityMissing' src
printf '%s\n' '--- notice copy ---'
rg -n -C 8 'notAvailable|updateOrcaToKeepUsing|updateOrca' src/shared/agent-session-write-notice-copy.ts src/renderer/src/components/native-chat/agent-session-write-notice-text.ts
printf '%s\n' '--- notice flow ---'
sed -n '280,390p' src/shared/agent-session-refusal-notice.ts

Repository: stablyai/orca

Length of output: 20327

---

🏁 Script executed:

set -eu
printf '%s\n' '--- cause types and reason rendering ---'
sed -n '1,95p' src/shared/agent-session-refusal-notice.ts
sed -n '225,290p' src/shared/agent-session-refusal-notice.ts
printf '%s\n' '--- capability policy ---'
sed -n '1,180p' src/main/runtime/rpc/methods/structured-agent-session-policy.ts

Repository: stablyai/orca

Length of output: 8843

---

Keep update guidance for clientCapabilityMissing.

When requireStructuredCapability rejects a remote client that lacks STRUCTURED_AGENT_SESSION_RUNTIME_CAPABILITY, the refusal uses clientCapabilityMissing. The new mapping renders only “This isn't available in this chat.” This removes the update guidance for the older-client compatibility path. Restore the existing code-level mapping.

Suggested fix

-    clientCapabilityMissing: causeWords('notAvailable', 'hostFinding'),
+    clientCapabilityMissing: codeWords('updateOrca'),

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	// Thrown, so a client meets these only as an RPC error. The code's own words ask for an update,
	// which only a method the host doesn't know proves; no reason here means an older Orca. An
	// unsupported location or agent, or no chat host, is not fixed by updating, and a client missing
	// the capability words this with its own older copy.
	structured_agent_session_unsupported: {
	clientCapabilityMissing: codeWords('updateOrca'),
	hostDisabled: codeWords('hostFinding'),
	hostUnsupported: codeWords('updateOrca')
	clientCapabilityMissing: causeWords('notAvailable', 'hostFinding'),
	hostDisabled: causeWords('notAvailable', 'hostFinding'),
	hostUnsupported: causeWords('notAvailable', 'hostFinding')
	// Thrown, so a client meets these only as an RPC error. The code's own words ask for an update,
	// which only a method the host doesn't know proves; no reason here means an older Orca. An
	// unsupported location or agent, or no chat host, is not fixed by updating, and a client missing
	// the capability words this with its own older copy.
	structured_agent_session_unsupported: {
	clientCapabilityMissing: codeWords('updateOrca'),
	hostDisabled: causeWords('notAvailable', 'hostFinding'),
	hostUnsupported: causeWords('notAvailable', 'hostFinding')

[pullfrog]

✅ No new issues found.

Reviewed changes — incremental re-review of the three commits since the prior pullfrog review (15e4599 → a5613b9), reading the delta and the full 51-file diff, then confirming with pnpm tc, the localization checks, and the changed test files (20 files, 279 tests passing).

• Client-worded host status rows — the new agent-session-host-status-rows.ts gives the history-repair and oversized rows a typed presentation; the host still writes plain English into text, and NativeChatNoticeRow translates the row for the reader. presentation already existed on the journal item, so no wire field is added.
• Composer send errors use the shared notice sentences — StructuredAgentSessionSendDisposition.error is now AgentSessionWriteNoticePart[]; the "couldn't save" and redelivery copy flow through agentSessionWriteNoticeText, and the obsolete structured-agent-session-outbox-error-text.ts is deleted.
• structured_agent_session_unsupported reason copy — when the host names why, the notice now says "This isn't available in this chat." instead of asking to update Orca; the code's own update copy still stands for the no-reason (method-not-found) path.
• Read-retry line dropped — headlineSaysUnread / NATIVE_CHAT_KEEPS_TRYING_COPY / agentSessionNoticeSaysThisChatUnread removed; a named read failure shows only its own sentence.
• State wording — after a count, unverifiable reads "N with no recent update"; agentChildRowDetailText falls back to the state word; previousExitUnverifiable reworded.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes — incremental re-review of the single commit since the prior pullfrog review (a5613b9 → 5ffbbc1), reading the delta against the full 51-file diff and running the two changed test files (2 files, 64 tests passing).

• Read refusals for a host that can't run the chat now say the history didn't load — causeParts special-cases cause === 'notAvailable': a read-history write returns ['notDoneReadHistory'] ("This chat's history couldn't be loaded.") instead of "This isn't available in this chat.", while every other write keeps the shorter sentence. The pane showed the old line because the host's hostUnsupported/hostDisabled/clientCapabilityMissing reasons all map through notAvailable.
• Reason-matrix tests split reads from other writes — the every-reason test now asserts ['notDoneReadHistory'] for structured_agent_session_unsupported reads and ['notAvailable'] otherwise, and the read-error pane gains a hostUnsupported case asserting the history sentence appears once and no "isn't available"/"newer Orca" text leaks.

The change is behaviorally narrow and the new pane test fails against the prior wording, so the fix has real coverage rather than an assertion that would pass either way.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[brennanb2025]

Review summary

The problem. When something went wrong in the native chat, the message that explained it often used Orca's internal words: "2 journal lines could not be read", "[Orca: item truncated — 3145728 bytes exceeds the history page budget]", a subagent marked "unverifiable · no contact", "Skills are unavailable for this host". A reader can't tell from these what happened or what to do. The first version of this PR reworded the English, but review found three deeper problems:

• Wrong advice. The new "This needs a newer Orca… Update Orca there" sentence showed for refusals that updating can't fix. Examples: a chat whose location the host can't run (WSL, or Windows without the process check it needs), or chats turned off on that computer. A user would update and nothing would change.
• Still English only. The two rows the host writes into a chat (history repaired, item too large) were still finished English sentences, so they could never follow the reader's language. The composer's error had also grown a second, separate way of wording errors beside the shared sentence table.
• A line that shouldn't exist. "Orca keeps trying to load this chat." only describes Orca retrying in the background. The common pattern shows just that loading failed, so the line should go, not be reworded.

What changes for the user (final head 5ffbbc1c44d):

• "Update Orca" appears only when the other computer didn't recognise the request at all, which proves it runs an older Orca. A refusal for a named reason says "This isn't available in this chat." If the history itself can't be opened, it says "This chat's history couldn't be loaded."
• The repair row and the too-large row are now translated. An older desktop build, or the phone, shows the English text instead.
• Both composer errors are translated: "Orca couldn't confirm your message reached the agent. Check the chat, then send it again if needed." and "Couldn't save your message. Try again."
• A chat whose history failed to load shows only "Could not load conversation". The "keeps trying" line is gone, and Orca still retries.
• A subagent or background task Orca has lost sight of has one name everywhere, "no recent update". After a count it reads "2 agents with no recent update".
• The skills menu's screen-reader text matches what's on screen.
• The row for messages waiting on the previous agent reads "Claude from before may still be running. Your messages will send once it stops."

Fixed during review:

• P2:
  • "Update Orca" was shown for refusals it can't fix.
  • The host rows couldn't be translated.
  • The composer had a second wording path.
  • The retry line was reworded instead of removed.
• P3:
  • One state had two names ("no recent update" and "status unavailable"). The second had no real precedent and is dropped.
  • The skills-menu screen-reader text didn't match the screen.
  • The waiting row named Orca's checking loop instead of what happens to the message.
  • A refused history read briefly said only "This isn't available in this chat." This was a side effect of two earlier review fixes, caught in round 2.
  • The PR description had several claims that turned out to be false. It now has correct ones and labels every difference from the common pattern as intended or temporary.

Deferred (all P3, none blocking):

• A composer error stays in the old language if you switch languages while it's showing. Every other composer error behaves the same way, and before this PR it was always English.
• The "messages waiting" row stays in the chat after the messages send. The common pattern shows no such row, so the right fix is to remove it, which is a behaviour change outside a wording PR.
• The skills menu in SSH chats shows a heading plus "Skills aren't available in SSH chats". The common pattern has no such line.
• An item too large to show is replaced whole rather than cut short inside, and a group of them gives several identical rows.
• The "couldn't save" sentence pair lives in shared code but only the desktop uses it.
• A host released between 2026-08-28 and 2026-09-28 doesn't say why it refuses, so it can still get the "update Orca" sentence. After updating, it gets the accurate one.

Verified:

• Review: a full adversarial round with an architecture challenge, a second round on the fixes, a narrow check of the last fix, then a readiness check. No P0, P1 or P2 is open.

• On the final head: the targeted test files, the phone's suites that share this wording, the translation checks, lint, the changed-code quality gate, and the web, node and CLI typechecks all pass. CI is green.

• A trial merge in the planned order (main, then fix(native-chat): show "Sending…" on a message until the host confirms it #24606, then fix(native-chat): queued messages carry on in order after any turn, and nothing sends by itself after a restart #24586, then this PR) is clean at every step.

• Hidden, isolated test builds on Windows, with a stand-in agent, before and after:

  • the repair row in English and Japanese;
  • the too-large row in Japanese, plus its English on-screen text;
  • the lost-contact subagent label.

  The reworded rows render at the same size and colour as before, and the new build showed no new pop-up, toast or banner. Screenshots are in the PR description.

Not verified:

• Not reproduced in a live app, so covered only by rendered component tests:

  • the skills-menu errors;
  • the retrying load-error pane;
  • the composer's "couldn't save" error;
  • the refusal notices;
  • the waiting row;
  • "couldn't confirm your message reached the agent".

  The test build offered no way to force those failures, and the stand-in agent can't produce those states on demand.

• Missing screenshot: a fresh English too-large screenshot of the final head. The capture timed out, so it is verified by on-screen text only.

• Merged result not typechecked: the trial merge with fix(native-chat): show "Sending…" on a message until the host confirms it #24606 and fix(native-chat): queued messages carry on in order after any turn, and nothing sends by itself after a restart #24586 was checked for conflicts only.

• Not tested: the real Claude CLI and an SSH chat.