fix(composer): polish automatic mentions

[tellaho]

Category: improvement
User Impact: Automatic mentions are easier to turn off and now behave consistently across conversations, settings, drafts, and repeated agent mentions.

Problem: People found the new automatic mention behavior hard to control: turning it off in Settings did not reliably affect the composer, removing a mention could require also disabling the feature, and root/thread composers could inherit or restore surprising state. Other reported rough edges included only one of several mentioned agents becoming automatic, synthetic mentions leaking into drafts, restored mentions corrupting adjacent text, controls remaining visible in archived channels, and unclear picker feedback. See the original feedback thread.

Solution: Polish the existing feature around the problems people encountered, keeping automatic mentions controllable and scoped to the active conversation.

Reported issue	UX fix
Turning automatic mentions off in Settings did not reliably update the composer.	The global setting and composer control stay synchronized, and disabling the feature does not clear typed text.
Removing an automatic mention could require both deleting the mention and turning off the feature.	Removing or unchecking an agent excludes that agent for the current conversation, while explicitly re-adding the agent can restore automatic mention behavior.
Root and thread composers could share or restore surprising selections.	Each root or thread composer keeps its own automatic audience and restores it when the user returns. A request to enable automatic mentions only in agent threads was considered; this PR keeps them available at the channel root but prevents state from leaking between the two.
Mentioning multiple agents could leave only one saved as automatic.	Multi-agent selections remain represented in the automatic audience and restored mention chips.
Automatic mention prefixes could be saved as if the user typed them.	Synthetic prefixes stay out of persisted drafts while authored text is preserved.
Restored mentions could lose their separator and corrupt continued typing.	Restored multi-word mentions retain their trailing space and place the caret after it.
Archived channels showed automatic-mention state beside a disabled composer.	Disabled composers hide automatic-mention controls while preserving the draft and restoring state when re-enabled.
Confirmation and picker behavior made the feature feel difficult to inspect or adjust.	Confirmations dismiss with removed agents, remain open while hovered, and expose the setting before it changes; pin icons, contrast, scope copy, animation, and keyboard toggling are also clarified.
Agent suggestions and membership state could shift during directory refreshes.	Suggestions and membership labels stay stable during refreshes, while send-time authorization still revalidates access.

Changes

File changes

desktop/src/features/agents/lib/agentAutocompleteEligibility.test.mjs
Adds coverage for the channel-roster eligibility rules used by agent mention autocomplete.

desktop/src/features/agents/lib/agentAutocompleteEligibility.ts
Aligns agent autocomplete eligibility with channel membership so available agents and their labels stay trustworthy.

desktop/src/features/channels/ui/MembersSidebar.tsx
Uses the shared member-pubkey logic when presenting and acting on channel members.

desktop/src/features/messages/lib/autoPinMentionedAgentsPreference.test.mjs
Covers preference changes that must remain stable while composer controls are toggled.

desktop/src/features/messages/lib/autoPinMentionedAgentsPreference.ts
Keeps the automatic-mention preference as durable user intent rather than transient composer state.

desktop/src/features/messages/lib/mentionMemberPubkeys.ts
Centralizes which member identities count as mentionable in the current channel.

desktop/src/features/messages/lib/persistentAgentAudience.test.mjs
Expands lifecycle coverage for persistent agent audiences, explicit exclusions, and restored mentions.

desktop/src/features/messages/lib/persistentAgentAudience.ts
Models automatic, explicit, and excluded agent audiences separately so user choices survive updates without leaking across composers.

desktop/src/features/messages/lib/stripImplicitAgentMentions.test.mjs
Verifies implicit automatic mentions are removed without damaging surrounding separators or authored content.

desktop/src/features/messages/lib/stripImplicitAgentMentions.ts
Strips presentation-only automatic mentions before draft persistence while preserving whitespace and authored text.

desktop/src/features/messages/lib/useMentions.ts
Routes mention insertion and removal through the composer-local audience lifecycle.

desktop/src/features/messages/lib/useRichTextEditor.ts
Preserves mention-chip structure and caret placement when automatic mentions are restored.

desktop/src/features/messages/ui/ComposerAddressControls.test.mjs
Updates control-state expectations for disabled automatic mentions and restored pin affordances.

desktop/src/features/messages/ui/ComposerAddressControls.tsx
Makes automatic-mention state, disabled presentation, and pin controls visually explicit.

desktop/src/features/messages/ui/MentionAutocomplete.test.mjs
Adds coverage for roster labels, pin state, and picker behavior after mention selection.

desktop/src/features/messages/ui/MentionAutocomplete.tsx
Keeps the shortcut picker open for repeated selection and restores visible automatic-mention pin indicators.

desktop/src/features/messages/ui/MessageComposer.tsx
Scopes automatic mention state to each root or thread composer and coordinates restoration, draft persistence, and sending.

desktop/src/features/messages/ui/MessageComposerToolbar.tsx
Passes the effective automatic-mention state into the toolbar presentation.

desktop/src/features/messages/ui/composerAgentKeyboard.test.mjs
Updates keyboard interaction coverage for toggling agents in place.

desktop/src/features/messages/ui/useAddressedAgentMentionRestore.ts
Restores automatic mention chips after lifecycle changes without moving or duplicating authored content.

desktop/src/features/messages/ui/useAgentAddressLockPicker.test.mjs
Substantially expands coverage for toggles, exclusions, synchronization, and picker dismissal rules.

desktop/src/features/messages/ui/useAgentAddressLockPicker.ts
Keeps the picker usable across repeated choices and preserves explicit per-agent intent while settings change.

desktop/src/features/messages/ui/useAlwaysAddressShortcut.ts
Makes the keyboard shortcut toggle the highlighted automatic audience choice without replacing unrelated selections.

desktop/src/features/messages/ui/useAutoPinMentionedAgents.ts
Owns composer-local automatic-mention lifecycle behavior, including restoration, exclusions, deletion, and disabled-state handling.

desktop/src/features/messages/ui/useComposerMentionPicker.test.mjs
Adds focused picker lifecycle coverage for selection, hover, and dismissal behavior.

desktop/src/features/messages/ui/useComposerMentionPicker.ts
Prevents premature picker dismissal while the user is interacting with its controls.

desktop/src/features/messages/ui/useDraftPersistSnapshot.ts
Persists only user-authored draft content rather than implicit automatic mention decorations.

desktop/src/shared/lib/keyboard-shortcuts.ts
Updates the automatic-mention shortcut description to match its toggle behavior.

desktop/src/testing/e2eBridge.ts
Extends the desktop test bridge with the state needed to exercise roster and automatic-mention transitions.

desktop/tests/e2e/mentions.spec.ts
Covers roster-based labels, managed-agent invitation, revocation, and recovery behavior in the complete mention flow.

desktop/tests/e2e/persistent-agent-audience.spec.ts
Adds end-to-end coverage for root/thread isolation, preference synchronization, manual exclusions, draft hygiene, restored chips, separators, hover behavior, and disabled presentation.

Reproduction Steps

1. Open a channel with at least two available agents and enable automatic mentions from the composer mention control.
2. Select multiple agents, remove or uncheck one, and confirm subsequent composer updates keep that agent excluded while the others remain automatic.
3. Open a thread, choose a different automatic audience there, and switch between the thread and root composer; confirm each composer retains only its own choices.
4. Disable automatic mentions and confirm the draft text remains unchanged while automatic chips and controls show the disabled state; re-enable the setting and confirm eligible automatic chips return.
5. Delete an automatic mention chip, then explicitly add the agent again; confirm it immediately returns as an automatic mention without disturbing spaces or the caret, including for a multi-word name.
6. Reload with a saved draft and confirm implicit automatic mentions were not persisted as authored draft text.
7. Use the automatic-mention keyboard shortcut and picker repeatedly; confirm the picker remains open for additional choices and the highlighted agent toggles in place.

Validation

Validated at 34d208b47d64a9816f88e10a46bcfd479e917d75 after rebasing onto origin/main (69096c9a8):

• Desktop unit tests: 5,731 passed, 0 failed.
• Desktop TypeScript typecheck: passed.
• Desktop E2E build: passed; emitted only existing chunk and dynamic-import warnings.
• pnpm check: exited successfully; 4 warnings and 5 informational findings are in unrelated files introduced by current main.

Screenshots/Demos

The behavioral changes are covered by the focused desktop E2E scenarios above. Screenshots can be attached from the screenshot-producing automatic-mention E2E after the PR is created.

[github-actions]

🔐 Codex Security Review

Status: review required for the current range.

The current range is 2c99ee7af5a20d239e95ec5368407887449795c9...d312ca3fb2fe13b277efed091a8a703ffa3170d4.
A new review must complete for this exact range. When manual authorization
is required, a Block organization member must comment exactly
@buzz-security-review d312ca3fb2fe13b277efed091a8a703ffa3170d4 to authorize a new review.
Any previous review applies only to its recorded range.

[jedwards27]

Reviewing exact head 34d208b47d64a9816f88e10a46bcfd479e917d75 against base 69096c9a8db726eebd83422cdc5e6c72ed9d4f90, I found three author-actionable blockers.

1. [P2] Persisting a draft can delete an authored duplicate leading mention. stripImplicitAgentMentions repeatedly removes every matching leading display-name token (desktop/src/features/messages/lib/stripImplicitAgentMentions.ts:15-22), but the persistence call sites provide only the locked agents' names, not the provenance or boundary of the synthesized prefix (desktop/src/features/messages/MessageComposer.tsx:131-133,203-207,461-465; desktop/src/features/messages/hooks/useDraftPersistSnapshot.ts:129-135,206-216). With one automatic @Morgarita prefix and authored content @Morgarita authored duplicate, the combined editor value @Morgarita @Morgarita authored duplicate persists as authored duplicate; the user's authored mention is silently lost. Existing unit coverage handles wholly synthetic prefixes and non-leading mentions, but not this identical authored-leading case (desktop/src/features/messages/lib/stripImplicitAgentMentions.test.mjs:6-39).

   Author action: strip exactly the editor-inserted automatic prefix using explicit provenance/an exact captured prefix boundary, and add a regression covering an automatic prefix followed by an authored identical mention through channel/thread switch and restore.

2. [P2] Restoring an automatic mention concatenates it with the authored draft text. The PR-authored E2E implicit automatic mentions stay out of persisted drafts reproducibly restores @Morgaritadraft text instead of @Morgarita draft text after adding the automatic mention, typing draft text, opening a thread, and returning. The mention remains highlighted, so this is a user-visible draft corruption at the reinsertion/persistence boundary rather than an assertion-only failure.

   Author action: preserve exactly one separator when restoring an implicit addressed mention before authored text, retain valid mention identity/highlighting and caret behavior, and make the existing persistent-agent-audience.spec.ts case pass.

3. [P2] The required Desktop smoke gate is red on the changed mention-identity journey. In run 33110679208, job 98696932534, desktop/tests/e2e/mentions.spec.ts:325 failed all three attempts: selecting managed agent carl sends local, but readOutgoingMentionPubkeys is [] rather than [aaaaaaaa…] at lines 430-432. The aggregate Desktop gate is consequently red. This exercises exact-pubkey selection/signing behavior changed by this PR.

   Author action: fix the lost managed-agent recipient—or, if the fixture/oracle is wrong, correct it with a demonstrated causal explanation—and make Desktop Smoke E2E (2) plus aggregate Desktop green at the reviewed head.

Validation at this immutable head: full unit suite 5,731/5,731 passed; pnpm typecheck, pnpm check, pnpm build, and git diff --check passed; targeted persistent-agent-audience.spec.ts was 25 passed / 1 failed on blocker 2. DCO, Desktop Core, macOS build, release candidate, and relay-backed integration gates are green. Native GUI/manual observation was not performed; that is a reviewer-owned confidence gap, not requested author work.

[tellaho]

AI-generated update: the three blocking paths are addressed at 4a93533e2.

• Draft persistence now removes only the exact captured automatic-mention prefix, preserving an authored identical leading mention; the per-draft provenance map is bounded.
• Empty-composer insertion now arms mention settlement so the separator survives the first keystroke, while the async caret write is guarded by editor focus so settings/popovers can take focus normally.
• The exact-pubkey smoke oracle now types through the real automatic-prefix journey and inspects the actual outgoing content; no send-path behavior was changed.

Validation on the cleared tree: changed unit lanes 62/62, targeted draft pair 20/20 across ten consecutive paired runs, settings synchronization 4/4, modified desktop E2E selection 102/102, typecheck clean, and git diff --check clean. An independent pre-push review also removed the focus guard and reproduced the settings failure 4/4, then restored it and passed 4/4. The branch was force-pushed after that clearance; DCO and the current security-review workflow are green. There are no inline review threads to resolve.

[tellaho]

@buzz-security-review 2ce0137

[tellaho]

@jedwards27 AI-generated current-head update: the branch is now rebased onto b49675894, with all three requested fixes preserved at 2ce0137f6 and every required check green, including Desktop Core, all four Desktop Smoke shards, Desktop aggregate, both integration shards and aggregate, macOS build, release candidate, DCO, and security authorization.

The rebase integrated main's exact typed-mention-on-Space behavior. That behavior gives the authored duplicate its own valid mention identity after restoration, so the E2E now expects two highlighted mentions while continuing to assert the exact persisted authored duplicate content. Full desktop units pass 5,762/5,762, and the focused settings plus draft-restoration journeys pass 3/3. There are no unresolved inline threads. Please re-review the current head when ready.

[jedwards27]

Verdict: REQUEST CHANGES

Reviewed: b49675894b39f87215e4dfc8c1ad4a3c28c6097e..2ce0137f6f6988ea0621c201ab696f825e27ed42
Risk: high — this changes automatic mention identity, draft persistence, restored editor state, and exact message recipients.

Blocking finding

[P2] Preserve the inserted automatic-prefix provenance across profile/name refreshes

MessageComposer recomputes and overwrites the draft's supposedly captured prefix from the current lockedAgents display names on every render (desktop/src/features/messages/ui/MessageComposer.tsx:473-480). Those names are themselves refreshed from live profile data (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:103-125). Restore then checks the current text only for the new name and prepends it when absent (useAgentAddressLockPicker.ts:372-424), while persistence strips only the newly recomputed prefix (useDraftPersistSnapshot.ts:129-135; stripImplicitAgentMentions.ts:6-14).

That breaks the PR's persistence invariant under a normal async identity transition:

1. The automatic recipient is inserted as @OldName and the user types a draft.
2. The same pubkey resolves to NewName after a profile/relay refresh.
3. Restore considers @NewName missing and prepends it.
4. Persistence strips @NewName but leaves the synthetic @OldName as authored content.
5. A later restore can show both recipient-looking prefixes.

The new stable-name duplicate test (desktop/tests/e2e/persistent-agent-audience.spec.ts:1103-1138) does not exercise this transition, and the nearest hook test covers the name cache becoming null rather than one non-null name changing to another (useAgentAddressLockPicker.test.mjs:792-824).

Author action: retain the exact automatic prefix (or identity-bound inserted range) at insertion/restore time instead of overwriting its provenance from later profile renders. Add a regression that changes one locked pubkey from a non-null old display name to a different non-null new name while authored content exists, crosses a persistence/remount boundary, and asserts: no duplicate visible prefix, no synthetic prefix in stored draft content, and the exact pubkey remains the recipient.

What is fixed

The prior stable-name failures are causally addressed: exact-prefix stripping preserves an identical authored mention; separator/caret restoration now has focused coverage; and the outgoing-recipient regression exercises typed content and asserts the exact managed/relay pubkey. The formerly failing smoke shard is green.

Validation at exact head

• PASS: focused prefix/address-lock tests, 19/19.
• PASS: pnpm typecheck && pnpm check && pnpm build.
• PASS: fresh pnpm build:e2e plus four focused former-blocker Playwright journeys, 4/4.
• PASS: git diff --check b49675894b39f87215e4dfc8c1ad4a3c28c6097e..2ce0137f6f6988ea0621c201ab696f825e27ed42.
• PASS: exact-head CI run 33179780715: Desktop Core, all four Desktop Smoke E2E shards, Desktop E2E Integration, macOS build, release candidate, aggregate Desktop, and DCO.
• INCOMPLETE: separate local full-suite attempts exceeded reviewer harness limits under concurrent load; exact-head package CI is green. This is a reviewer/tooling confidence gap, not additional author action.
• NOT RUN: native Tauri/AX observation. Browser-level focus/caret/keyboard assertions exist, but no native semantic receipt was obtained; reviewer/tooling owns this optional confidence gap.

Verification owner: author for the causal profile-refresh/persistence regression; reviewer reruns affected Desktop gates and that journey.
Residual risk: native WebKit/AX behavior remains unwitnessed locally, but it is not the reason for this verdict; the source-level persistence defect above is.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

Verdict: REQUEST CHANGES

Reviewed exact head 2ce0137f6f6988ea0621c201ab696f825e27ed42 against base b49675894b39f87215e4dfc8c1ad4a3c28c6097e.

[P2] Preserve the identity of the automatic prefix when an agent's display name changes

MessageComposer overwrites the draft-key entry in implicitAgentMentionPrefixByDraftRef on every render from the current lockedAgents display names (desktop/src/features/messages/ui/MessageComposer.tsx:473-480). Those names are recomputed from live profile/directory data (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:103-125). Restoration likewise searches only for the current display name and prepends it when absent (useAgentAddressLockPicker.ts:372-424), while draft cleanup strips only the latest recomputed string (useDraftPersistSnapshot.ts:129-135,201-217).

A normal async identity refresh therefore breaks the promised draft invariant:

1. Automatic addressing inserts @OldName and the user types draft text.
2. The same pubkey resolves to NewName after profile/directory refresh.
3. Restoration sees no @NewName, prepends it, and the render overwrites the recorded prefix with @NewName .
4. Persistence strips @NewName but leaves the synthetic @OldName in the saved authored draft.
5. A later restore can display both recipient-looking prefixes even though only one pubkey is addressed.

The same mutable-prefix model also cannot distinguish a reintroduced synthetic prefix from an authored identical leading mention after automatic addressing is toggled off and on. The stable-name duplicate E2E at desktop/tests/e2e/persistent-agent-audience.spec.ts:1103-1138 does not cover either lifecycle transition.

Author action: record the exact synthetic insertion provenance at the editor transaction/insertion boundary rather than deriving and overwriting it from the current lock list. Add regressions for (a) one locked pubkey changing from one non-null display name to another while authored text exists and (b) off/on re-enable followed by an authored identical leading mention. Across persistence/remount, assert no duplicate or leaked synthetic prefix, preservation of authored text, and the same exact recipient pubkey.

Exact-head Desktop Core, smoke E2E, integration E2E, build, release-candidate, and DCO checks are green. This is a source-level persistence/lifecycle defect not exercised by those gates.

[morgmart]

🤖 Fixed the profile-refresh draft edge case at cf8bcefa1:

• The automatic-prefix record is now written only when Buzz actually inserts that prefix, rather than being recomputed from later agent-profile renders.
• Restoration recognizes the existing saved mention by its stable agent identity, so a display-name change does not add a second automatic prefix.
• Added a regression covering an automatic @OldName mention whose same pubkey later resolves as @NewName; restoration keeps the existing draft mention and does not insert another.

Validation: full desktop unit suite 5,763/5,763, pnpm typecheck, Biome on changed files, and the repo pre-push lanes all pass.

@jedwards27 Please re-review the new head when ready.
@buzz-security-review cf8bcefa159f2dbcf85df76771c2daac865ad30b

[morgmart]

@buzz-security-review cf8bcef

[morgmart]

🤖 Added Carl’s additional off/on lifecycle regression at a530f29dbbef34bd774430aa14b7eef29c07439f. It turns automatic mentioning on, removes it, re-enables it, then adds an authored identical leading mention and crosses a draft restore boundary. The test proves only the newly automatic prefix is removed from storage; the authored duplicate remains.\n\nValidation: fresh E2E build and the focused smoke journey pass (1/1).\n\n@wesbillman @jedwards27 Please re-review the current head when ready.

[morgmart]

@buzz-security-review a530f29

[jedwards27]

Verdict: REQUEST CHANGES

Reviewed exact head a530f29dbbef34bd774430aa14b7eef29c07439f against base b49675894b39f87215e4dfc8c1ad4a3c28c6097e.

The profile-rename blocker from the previous head is source-addressed: automatic-prefix provenance is now captured at insertion, and restoration checks existing mention identity by pubkey. Two deterministic draft-integrity failures remain in the same provenance lifecycle.

[P2] Clear automatic-prefix provenance when its synthetic mention is removed

onImplicitPrefixInserted stores one prefix per draft (desktop/src/features/messages/ui/MessageComposer.tsx:467-473), but removing the automatic lock deletes the visible prefix without clearing or updating that stored provenance (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:173-197). Draft persistence later strips whatever still matches the stale record (desktop/src/features/messages/ui/useDraftPersistSnapshot.ts:129-135,201-217).

A real mock-bridge Playwright journey reproduced this at the reviewed head:

1. Enable automatic @Morgarita.
2. Remove the lock, leaving the composer empty.
3. Manually author @Morgarita manual after removal while the agent remains unpinned.
4. Navigate away to persist the draft.

Expected stored content: @Morgarita manual after removal. Actual: manual after removal. The authored mention is silently deleted.

Author action: clear/update the draft's synthetic-prefix provenance whenever that exact synthetic prefix is removed, including lock removal and typed deletion/synchronization paths, without clearing provenance for a currently restored automatic prefix. Add a remove → identical manually authored mention → persist/remount regression.

[P2] Accumulate provenance for multiple automatic agents instead of replacing it

Each insertion reports only its newly inserted text (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:247-255,430-439), and MessageComposer replaces the draft's stored prefix with that value (MessageComposer.tsx:467-473). Persistence strips exactly that latest value (useDraftPersistSnapshot.ts:129-135; desktop/src/features/messages/lib/stripImplicitAgentMentions.ts:6-14).

A strengthened multi-agent browser journey failed 3/3 at this head: after automatically adding Morgarita, then Vogue, and authoring draft text, the editor correctly showed both automatic mentions, but navigation persisted @Morgarita draft text instead of draft text. Adding Vogue overwrote the @Morgarita provenance, turning synthetic UI into durable authored content.

Author action: track the complete current synthetic leading prefix—or equivalent identity/range provenance—across add/remove/re-add/rename. Add a multi-agent persistence/remount regression proving that stored content excludes every synthetic mention while visible restoration and exact recipient pubkeys preserve both agents.

Exact-head evidence

• PASS: focused address-lock/prefix tests, 20/20.
• PASS: fresh E2E build and typecheck.
• PASS: six focused browser journeys covering disabled/archive restore, keyboard toggle, single-agent draft hygiene, authored duplicate, off/on duplicate, and multi-word separator/caret, 6/6.
• PASS: exact-head CI, including Desktop Core, all smoke and integration shards, aggregate Desktop, macOS build, release candidate, DCO, and authorization.
• PASS: remote PR head matched local clean HEAD after testing.
• NOT RUN: native Tauri/WebKit accessibility/focus observation. Browser keyboard/caret/accessibility assertions passed; native semantics remain a reviewer-owned confidence gap, not author action.

Verification owner: author supplies the causal regressions and fix; reviewer reruns both failing journeys, affected full Desktop package gates, and exact-head CI.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

REQUEST CHANGES at exact head a530f29dbbef34bd774430aa14b7eef29c07439f (base b49675894b39f87215e4dfc8c1ad4a3c28c6097e).

Blocking findings

P2 — Preserve all generated automatic mentions when persisting a draft

Each automatic-agent insertion calls onImplicitPrefixInserted with only the newly inserted fragment (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:245-255,430-439). MessageComposer replaces the per-draft provenance entry with that fragment (desktop/src/features/messages/ui/MessageComposer.tsx:467-473), and draft persistence strips exactly that one string (desktop/src/features/messages/ui/useDraftPersistSnapshot.ts:129-135,201-217).

With automatic agent A followed by B, the composer becomes @B @A draft, but persistence records only @B and saves synthetic @A draft as authored content. Returning to the composer can then restore both automatic agents around the leaked prefix, producing a duplicate/stale visible mention and potentially stale mention-reference identity.

Required change: maintain the complete current generated leading prefix, or equivalent identity/range provenance, across add/remove/re-add/rename. Add a multi-agent persistence/remount regression proving stored draft content excludes every synthetic mention while the restored composer and exact recipient pubkeys retain both agents.

P2 — Clear or update provenance when an automatic mention is removed

onImplicitPrefixInserted stores provenance per draft (MessageComposer.tsx:467-473), but removeAddressedAgent removes the visible automatic prefix without clearing or updating that entry (useAgentAddressLockPicker.ts:173-197). If the user then manually authors the same leading mention and leaves the composer, persistence still sees the stale generated prefix and removes the user’s text.

Reproduction: enable automatic @A; remove A so the composer is empty; manually type @A authored; navigate away. The saved draft becomes authored, silently deleting the authored mention.

Required change: update synthetic-prefix provenance whenever the corresponding generated text is removed, including lock removal and typed deletion/synchronization paths. Add remove → identical manually authored mention → persist/remount coverage.

Coverage

The previous profile-rename blocker is addressed: provenance is captured at insertion and restoration recognizes stable pubkey identity. The team also traced root/thread isolation, exact-recipient delivery, picker/cursor behavior, disabled/archived presentation, membership refresh, and send-time revalidation; no additional blocker was verified. Exact-head required CI is green. Review was read-only GitHub source/diff inspection; no PR code was executed.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

REQUEST CHANGES at exact head 593f8d17587d4fd85b3c4384c0ba222e779312ca (base 2f66ee24a0bf5efa958378069a6b155e99a1d068).

P2 — Clear generated-prefix ownership when the user deletes an automatic mention

The editor synchronization path notices when a previously visible locked agent mention disappears and excludes that pubkey, but it does not clear the generated fragment owned by that pubkey (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:149-175). The separate toolbar-removal path does call onImplicitPrefixRemoved when it removes the prefix (useAgentAddressLockPicker.ts:178-213), so these two equivalent removal transitions leave different provenance state.

That stale fragment remains in useImplicitAgentMentionProvenance (desktop/src/features/messages/ui/useImplicitAgentMentionProvenance.ts:13-20,41-50) and draft cleanup later treats any matching leading text as generated (desktop/src/features/messages/ui/useDraftPersistSnapshot.ts:129-135,201-217). A user's authored text can therefore be silently changed:

1. Automatic addressing inserts @Morgarita .
2. Select all and press Backspace; the mention and lock disappear.
3. Manually type @Morgarita manual after typed deletion.
4. Navigate away so the draft persists.

The stale record still owns @Morgarita , so the stored draft becomes manual after typed deletion. The existing typed-deletion journey re-adds and sends immediately (desktop/tests/e2e/persistent-agent-audience.spec.ts:1024-1061), while the persistence regression uses toolbar removal (:1141-1172); neither crosses typed deletion into persistence. The hook tests likewise assert audience exclusion without observing provenance removal (desktop/src/features/messages/ui/useAgentAddressLockPicker.test.mjs:521-618).

Required change: clear the generated fragment by stable pubkey when text synchronization observes that automatic mention removed, while retaining fragments still present. Add a typed-delete → identical manual reauthor → persist/remount regression asserting the authored mention survives, the automatic lock stays absent, and recipients remain accurate.

The latest head fixes the earlier multi-agent accumulation and toolbar-removal defects. Stable-pubkey rename handling, off/on restoration, and current mention/caret behavior showed no additional blocker in read-only exact-head review. Exact-head Desktop Core, all smoke and integration shards, aggregate Desktop, macOS build, release candidate, and DCO are green, but they do not exercise this lifecycle.

[morgmart]

🤖 Fixed the typed-deletion provenance gap at d312ca3. When editor synchronization sees an automatic mention disappear, it now clears that pubkey’s generated fragment before excluding the automatic lock—the same ownership cleanup used by the toolbar remove path.\n\nAdded the reproduced journey: automatic mention → Select All/Backspace → manually type the identical mention → navigate away. It now persists the full authored mention and leaves the automatic lock absent.\n\nValidation: typecheck, fresh E2E build, and all six focused draft-lifecycle journeys passed.\n\n@wesbillman @jedwards27 Please re-review the current head when ready.

[morgmart]

@buzz-security-review d312ca3

[jedwards27]

:bot: Jude’s code review agent — APPROVE at exact head d312ca3fb2fe13b277efed091a8a703ffa3170d4 against base 2f66ee24a0bf5efa958378069a6b155e99a1d068.

The prior typed-deletion blocker is resolved. When editor synchronization observes a formerly tracked automatic mention disappear, the implementation now clears generated-prefix provenance for that stable pubkey before recording the exclusion (desktop/src/features/messages/ui/useAgentAddressLockPicker.ts:149-177). A later identical manually authored mention is therefore preserved rather than mistaken for generated text. The regression at desktop/tests/e2e/persistent-agent-audience.spec.ts:1141-1180 covers automatic mention → typed deletion → absent lock → identical manual mention → persisted authored text.

Systems and product review found no unresolved author-actionable defect. Reviewed contracts remain coherent across conversation-keyed audience ownership (persistentAgentAudience.ts:47-59), identity-keyed generated-fragment provenance (useImplicitAgentMentionProvenance.ts:11-53), and recipient revalidation/tag intersection before send (useMentionSendFlow.ts:430-461,540-557). Keyboard removal, focus-visible behavior, root/thread isolation, exclusions/re-additions, multi-agent persistence, and authored text preservation were also exercised.

Exact-head evidence:

• Full Desktop units: 5,788/5,788.
• pnpm --dir desktop typecheck, check, and production build: PASS.
• Fresh E2E build and focused browser journeys for typed deletion, explicit removal, authored duplicates, re-addition, multi-agent persistence, root/thread isolation, caret/chip restoration, persistence/remount, and exact-recipient send: PASS.
• Causal mutation removing the new provenance callback kept the build green but made the targeted regression fail with authored-prefix loss; restoring exact-head code passed with a clean tree.
• git diff --check: PASS; remote PR head matched the reviewed SHA.

Confidence gaps:

• Native Tauri/WebKit accessibility observation was not run. Author action: none. Verification owner: reviewer/tooling if native semantic evidence is desired.
• At submission freshness, Desktop Core remained in progress; all completed exact-head checks had no failures, and smoke shards 1–4, both Desktop integration shards, macOS build, release candidate, DCO, and authorization were green. Author action: none unless a required gate fails. Verification owner: named CI gate.

Superseded by the typed-deletion provenance fix at d312ca3, reviewed and approved on the current head.