Renames the openai-codex provider's display label across the CLI
(hermes model picker, provider labels), the dashboard OAuth accounts
catalog, and the Desktop onboarding + settings provider pickers.
Slug, aliases, and auth flows are unchanged.
Hovering a sidebar session row (one-line or card) now shows the complete
title in a styled tooltip — but only when the title actually overflows
its label, so fully visible titles never grow a redundant tip.
New OverflowTip primitive in ui/tooltip.tsx: a controlled Tip that
measures scrollWidth vs clientWidth on pointerenter and arms a 600ms
deliberate-hover delay only when the content is truncated. Community
request via @fhreire on X.
`hermes update` was pruning root-level Node dependencies (agent-browser)
because npm ci always wipes and reifies node_modules according to its
active filter -- no root-first/workspace-first ordering or flag
combination (--workspaces=false, --include-workspace-root, etc.) can
reliably keep a root-only package.json dependency from being pruned by
a subsequent workspace-scoped npm ci. Confirmed empirically and via
npm/cli source (isArboristCmd hardcodes includeWorkspaceRoot=false for
ci/install), so no amount of install-order juggling fixes this for good.
Instead of chasing install order, remove the root-only dependencies
that made the npm step fragile in the first place:
- agent-browser is no longer a root package.json dependency. It
resolves lazily via `npx agent-browser` (tools/browser_tool.py
already had this as a fallback; it's now the primary path).
warm_agent_browser_npx_cache() is called fire-and-forget from both
`hermes update` and `hermes doctor --fix` to keep npx's cache warm,
preserving the "available before any session starts" property
agent-browser had as an eager dependency without re-entangling it
with the npm workspace graph.
- @streamdown/math moves to apps/desktop/package.json, where it's
actually imported (markdown-text.tsx, katex-memo.ts) -- it was
never used anywhere else and was subject to the same pruning risk.
- _update_node_dependencies() collapses to a single
`npm ci --workspace ui-tui --workspace web` call now that root has
no dependencies of its own to protect, and keeps its original spot
ahead of `_build_web_ui()` at both call sites in update_cmd.py --
with no root-only dependencies left to protect, there's no reason
for the Node refresh and the web build to run in any particular
order relative to each other.
- hermes_cli/tools_config.py's post-setup Chromium-install path and
hermes_cli/doctor.py's agent-browser check both now resolve through
the same PATH -> Homebrew/Hermes-managed-node -> npx cascade
(_find_agent_browser / _resolve_npx_bin) instead of hand-rolling
their own node_modules/.bin lookups, so they can't diverge from what
browser tools actually invoke at runtime.
- tests-js/package-json-lazy-deps.test.ts gets a lockfile-level check
mirroring the existing camofox one, so a future regression that
reintroduces agent-browser into package-lock.json fails this test
directly instead of relying on manual review to catch it.
Fixes#43564.
Adds a 'Scheduled jobs' row to the sidebar's top nav (below Artifacts,
watch codicon, wired to the existing nav.cron keybind action) so the
cron overlay opens from the same rail as New session / Capabilities /
Messaging / Artifacts.
Inside the cron overlay, the list rail now also shows the Automation
Blueprint recipes below the jobs (same search box filters both).
Clicking a recipe opens the create dialog pre-seeded to that
blueprint's typed-slot form via a new optional blueprintKey on the
create EditorState. Catalog fetch reuses the ['cron-blueprints'] query
key, so no extra request.
i18n: sidebar.nav.cron added to en/zh/zh-hant/ja (ar already had it).
ensureGatewayProfile narrows the Sessions sidebar to the activated
profile as a side effect, so every cross-profile open from a plugin
surface silently locked the user into that profile's session list. A
plugin-driven open is a navigation, not a scope choice: openSession now
restores the unified all-profiles view after a cross-profile activation
(keepAllProfilesScope, default true; pass false for the old narrowing).
Same-profile opens write no scope at all.
Final-diff simplify/review pass findings on #83785:
- Consent gate (confirm_truncate -> 4029) now checked BEFORE target
resolution, restoring the pre-PR precedence: an unconfirmed submit
carrying truncation params refuses without paying the durable-transcript
read or heal-stamping live history dicts, and an unconfirmed out-of-range
ordinal returns 4029 (not 4018). Malformed params still refuse first
with 4004. Regression test added (spy DB asserts zero reads pre-consent;
mutation-checked against the previous commit).
- _coerce_truncate_ordinal generalized to _coerce_truncate_int(param_name):
the row_id branch was inlining the exact bool-guard + int() -> 4004
pattern the helper had just extracted.
- Deleted the dead user_indices re-read after _resolve_truncate_row_id
(heal mutates dicts in place; the filter output is identical) and the
duplicate range check that had deadened the pre-existing guard.
- Desktop: exported isVisibleUserMessage from use-prompt-actions/utils and
used it in visibleUserOrdinal / visibleUserIndexAtOrdinal /
rebindSurvivorRowIds — one predicate for the ordinal parity all three
depend on instead of three verbatim copies.
- Docs: programmatic-integration.md documents survivor_user_row_ids.
check:lint failed on the two sort-named-imports errors (survivorRowIdsFrom
before type SurvivorUserRowIds) introduced by the rebind commit; also adds
the blank line eslint wanted in the new test helpers.
Review follow-up (StanleyStetson + egilewski on #83785/#83202): a successful
rewind's replace_messages(archive_dropped=True) re-inserts the surviving
prefix as NEW SQLite rows. Gateway memory picks up the fresh _row_id stamps
via lastrowid, but the Desktop's surviving bubbles kept their pre-rewind
ChatMessage.rowId — so a second rewind/edit/regenerate of an older surviving
turn sent a stale truncate_before_row_id and was (correctly) refused with
4018 until a transcript reload. Fail-closed stays untouched, per both
reviews; the fix is rebinding, not ordinal fallback.
Server: prompt.submit now returns survivor_user_row_ids (fresh post-rewrite
ids of surviving visible user turns, in visible-user-ordinal order) on both
the inline and compute-host paths whenever a durable truncation committed.
Desktop: runRewindSubmit surfaces the field; restore/edit/reload on both the
primary chat and session tiles rebind surviving user bubbles positionally
(same visible-user filter the ordinal math uses) and clear any rowId they
cannot rebind — a cleared id degrades to the ordinal path instead of a 4018.
Absent field (older gateway) leaves state untouched.
Tests: consecutive-rewind regression on a real SessionDB (stale id 4018s,
returned id succeeds; mutation-checked) + vitest for survivorRowIdsFrom /
rebindSurvivorRowIds (rebind, null-clear, past-end clear, hidden skip,
identity preservation).
Address rewinds/edits via SQLite messages.id (truncate_before_row_id)
instead of shifting user ordinals. Resolve against in-memory stamps,
then durable session history when live turns drop _row_id; refuse
unknown durable targets with 4018 (no ordinal fallback) and 4030 on
ordinal/row_id mismatch. Stamp _row_id on insert, load row ids on
resume paths, send rowId from Desktop, filter renderer-synthetic ids,
and stop silently resending failed targeted edits without truncation.
Add production-shaped SessionDB tests for resolve and fail-closed paths.
Fixes#82959
The Project-grouping flag was one global bool while the grouping beneath it
was already stored per scope (workspace vs all-profiles). Picking Project
inside a workspace therefore dragged the all-profiles view into the project
tree and vice versa — "I have to re-set grouping every time I switch."
The flag now lives per scope like its sibling grouping atoms (the flat key
keeps its historical name so existing choices survive), setSidebarGrouping
writes to the scope it just switched INTO when Profile flips the view, and
reset clears both scopes.
scrollbar-overlay opts out of the themed thin scrollbar; on Windows there
are no native overlay scrollbars, so Chromium painted the classic
always-visible gutter instead — a permanent scrollbar next to the recents
list. The themed fade bar reserves its 4px on every platform but stays
invisible until hover, and the wrapper no longer stacks a second scroller,
which is what the overlay class was originally working around.
The card prop was gated off whenever Project grouping was active, so the
Inbox style toggle silently did nothing there. It is a render variant, not
a grouping: project lanes and overview previews now render the same card
the flat list does.
Also mirrors the section's real virtualization inputs (projectOverview /
entered-project content, not the persistent agentProjectTree cache) when
deciding the wrapper's scroll classes, and stops gating SCROLL_Y on that
parallel guess — the section is the single authority on which scroller
lives, so the recents pane can no longer end up with no scroller at all
(the "no sessions under Updated grouping after toggling settings" blank).
Third draft provider: recurring phrasing in the draft ("every morning",
"daily", "each week") offers a Schedule-this pill. Click prefixes the
draft with an explicit scheduling instruction and the agent creates the
job via its cronjob tool on send — the pill never schedules anything
itself. Proper-noun guard keeps titles like "the Daily Prophet" quiet;
hyphen-as-word-char keeps "weekly-report.pdf" quiet.
Two new sources on the suggestion bus, one per provider shape:
- skill (draft): the draft names an enabled skill (whole-word,
4+ chars), so offer to lead the message with its /command. Invoke
prefixes the draft via the new 'prefix' insert mode and stands down
once the draft starts with a slash; skill_manage invalidates the
cached index alongside the slash-completion cache.
- repair (event): an mcp__ tool call failing with auth/connection-
shaped output offers a one-click reconnect for that server, fed from
the gateway tool.complete handler. Reconnect runs the shared OAuth
flow with server-side cancel and reloads live tools before claiming
success; a later successful call to the same server withdraws the
offer on its own.
The bus now keeps a session-scoped declined ledger: a pill the user
watched appear and let die three times stops re-offering for the rest
of the session. Acting on a pill clears its count, so a suggestion
that was taken can come back for the next trigger. In-memory on
purpose — a fresh session is a fresh chance.
Desktop plugins reach the backend exclusively through the generic ws
JSON-RPC door (host.request), but profile enumeration/creation only
existed on the dashboard REST router, which plugins cannot reach — so
anything 'one chat per agent profile'-shaped (bot rosters, profile
pickers, team panes) was impossible to build as a plugin.
- tui_gateway/methods_profiles.py: new @method handlers
* profiles.list — profiles + optional last_session preview per profile
(mirrors session.list's kanban/tool deny-list; best-effort per-profile
state.db probe degrades to null instead of failing the call)
* profiles.create — ws twin of POST /api/profiles (clone_from/clone_all/
no_skills/description), plus optional SOUL.md content and a best-effort
model+provider pin; mirrors the CLI flow (seed skills, safe alias)
Both run on the RPC pool, not the WS reader thread (list_profiles walks
skill trees; create copies bundles).
- SDK: host.openSession(id, { profile, intent }) — open a stored session
the way core surfaces do, soft-swapping to the owning profile's backend
first (ensureGatewayProfile), and host.newChat(profile) — fresh draft in
a named profile (same door as the sidebar's per-profile '+').
- Docs: desktop-plugin-sdk.md gains both surfaces.
First consumer: a Grok Bot-style 'Bots' roster plugin (one persistent
chat per agent profile with a New Agent dialog) built on exactly these
four doors.
The pill strip from the inline-MCP work is worth more than one source, so
the MCP-specific store splits into two layers with the same UX contract
(session-scoped, capped, self-limiting, one-click with narrated
idle→working→done):
- store/composer-suggestions.ts — the bus. Draft providers register into
the existing debounced sampler; event providers push/withdraw directly.
Offerings merge (event before draft), dedupe by provider-namespaced key,
and keep reference identity on no-ops.
- store/suggestion-providers/mcp.ts — the founding provider, behavior
unchanged: directory keyword/host matching, configured-server exclusion,
one-click connect with OAuth cancel + config rollback.
- composer/suggestion-pills.tsx — the generic strip; phases and cancel
live here, action/rollback/toasts stay with the provider's invoke.
No new pills yet — this is the seam for them.
The card reads the labelled choices off the gateway request rather than
the raw tool args -- the backend applies the label there, and the card
only mounts once the request exists, so the args are a hydration-race
fallback. RECOMMENDED_LABEL and bareChoice live in the clarify store so
the component and the choice-length guard share one definition; without
the guard a long option could be dropped for length the label added.
A renderer-local directory of official hosted MCP remotes (URL-only,
vendor-documented endpoints — deliberately not the reviewed install
catalog) powers keyword and pasted-link suggestions: typing jira or
pasting a *.atlassian.net URL floats an 'Add Atlassian' pill in the
composer's micro-action strip. Matching is whole-word/phrase (unicode
boundaries) plus strict host-suffix on links, host hits outrank
keywords, capped at two, debounced 600ms, and excludes servers already
in mcp_servers. Pills are session-scoped like the micro-action badges
and self-limiting rather than dismissible — they exist only while a
trigger is in the draft. A click drafts the setup request; the agent's
setup_mcp card carries the consent. Brand glyphs extracted from the
mcp-tab into lib/mcp-brands (shared, monochrome marks follow the theme
so GitHub/Notion/Vercel survive dark mode).
The card follows the approval bar's consent vocabulary (primary-tinted
action + ghost decline, ⌘⏎/Esc with clarify's focus-stand-down rule) on
clarify's widget shell. Install prefers the reviewed catalog entry (env
prompts inline, background installs polled to completion) and falls back
to the desktop suggestion directory via the validated add-server POST +
OAuth; success reloads live MCP tools before unblocking the agent so it
resumes with the tools it was just promised. Esc stays live mid-flight
as cancel — the abandoned flow aborts at its next poll and a post-write
cancel rolls the config entry back. Typing while the card is pending
declines it and sends normally (skipClarifyRequest's pattern), and the
request/tool.start rows merge on the server arg so reconnects can't
double-render the card.
The one-line session row gets the exact treatment the inbox card's title
already has: hovering a truncated title glides the clipped tail into view —
one direction at constant speed, a short hold at each end, then a snap back.
Same armMarquee/disarmMarquee handlers, same CSS, so overflow is measured on
pointerenter, short titles never move, hover state lives in DOM attributes
(no re-render of the memoized row), and reduced motion disables it.
The helpers were tested; nothing proved main.ts called them. Reverting both
call sites and both imports in readDesktopConnectionConfig /
writeDesktopConnectionConfig left the whole suite green (947 passed / 2
skipped, tsc 0, eslint clean, e2e 1 passed 1 skipped) while connection.json
went back to 0644 — the user-visible fix this PR promises was untested.
The e2e spec could not catch it by construction: it asserts the ENCRYPTION
contract with a raw-bytes scan, and safeStorage keeps the token opaque
regardless of the file's mode, so a 0644 file passes that scan every time.
There was no mode assertion anywhere in e2e/.
Adds the missing third contract — unreadable by other local accounts — on all
three paths that can produce the file:
- write: assert the mode of the artifact test 1 already proves the app wrote.
- read, valid file: seed the app's own encrypted connection.json back to 0644
and assert launch tightens it. Scoped to the MODE only, so it is independent
of the still-deferred plaintext migration — the fixture's token is already
ciphertext, so nothing re-encrypts, no #62319 opt-in marker is involved, and
no rotation guidance is owed.
- read, corrupt file: a truncated file still holds the token bytes and throws
into the swallowing catch, so it would be the one file never tightened. This
is the only test that distinguishes the chmod's placement relative to the
parse.
Also moves the tighten above JSON.parse for exactly that reason, and pins the
cache invariant the placement depends on: the tighten must be a chmod, not a
rewrite, because it sits inside the function whose cache keys on mtimeMs.
Asserted as `mode & 0o077 === 0` rather than `=== 0o600` to avoid a
change-detector, and skipped on win32, where chmod maps to the read-only bit
and the fix deliberately no-ops (ACLs are PR #77527).
Every assertion was mutation-tested: reverting the full wiring fails all three;
reverting only the write path fails only the write test; deleting only the
tighten-on-read fails only the two read tests; moving the tighten below the
parse fails only the corrupt test; making the tighten a rewrite instead of a
chmod fails the mtime assertions. Bundle greps confirmed each mutation reached
dist/electron-main.mjs before the run.
(cherry picked from commit 99cfc16e7cdb759b674d890563f6a82113326547)
`connection.json` under the desktop app's Electron `userData` was written with no
file mode, so it landed at the `0644` umask default — while its two
credential-bearing neighbours in the same directory, `desktop-installation.json`
and `native-oauth-tokens.json`, were already `0600`. That file holds the
safeStorage-encrypted gateway token plus the fields that are NOT encrypted: the
gateway URL and the SSH host, user, and key path.
- Route the single write choke point through a helper that creates the file
owner-only and atomically.
- Tighten an already-existing `0644` file once per launch on the read path, so
installs that already have one do not stay world-readable until the next save.
- Refuse to act on a path that is a symlink or not owned by the current user,
matching the guards `desktop-installation.ts` already applies to its sibling.
The symlink guard alone turned out to be insufficient, and that is worth
recording: `writeSecretFileAtomic` tightens its *temp* path, so a symlink planted
at `connection.json.tmp` meant `writeFileSync` followed it, the guard correctly
bailed, and `renameSync` then moved the link onto `connection.json` permanently.
Measured, guard-only vs. as-landed:
guards only token leaked: true config is a symlink: true 755
guards + temp unlink token leaked: false config is a symlink: false 600
So the temp path is unlinked before the write.
Issue #77486's headline claim — that a dashboard session token is persisted in
plaintext — does not hold against main. The token has been safeStorage-encrypted
since the desktop app reached mainline in 51c68d4ab, and `encryptDesktopSecret`
aborts with an actionable message rather than degrading to plaintext when
safeStorage is unavailable. The `{ encoding: 'plain', value }` literal does exist
at main.ts:7084, but only on the `persistToken: false` branch, whose sole caller
is the connection-test handler, which never writes. So no mainline path *writes*
a plaintext token. The commits that did contain a plaintext-writing fallback
(d3d177283, d208f2c2c) are not ancestors of main — they live only on
upstream/bb/gui-* and the desktop-pr20059-installers pre-release tag.
At-rest migration of legacy non-safeStorage payloads is deliberately NOT included.
An earlier revision of this branch implemented it and it was removed after review
reproduced two token-loss paths: it force-converts the opt-in plaintext choice
PR #62319 adds (silently reverting the user's decision, then destroying the token
on the next launch without the `--password-store=basic` flag), and it converts a
portable credential into a keychain-bound one with no consent — destroying the
only recoverable copy while not remediating the real exposure, since every
existing backup still holds the plaintext and the true remedy is rotation. It also
persisted raw `parsed`, bypassing `sanitizeConnectionProfiles`. A comment at the
read path records the three preconditions any future attempt needs.
`decryptDesktopSecret`'s non-safeStorage read fallback is untouched — it is what
lets a pre-release or hand-edited config work at all.
Windows still inherits the userData directory ACL rather than an explicit
owner-only one; mode bits are advisory there, so that half is deferred to
PR #77527 rather than growing a second ACL implementation here.
e2e: `at-rest-connection-token.spec.ts` asserts the at-rest contract
implementation-independently — the token's plaintext value (and its base64 form)
must not appear in a raw-bytes scan of any file under userData or HERMES_HOME,
AND the app must still put the exact original token on the wire after a restart,
so a fix that simply drops the token cannot pass. Proven non-vacuous by mutation:
writing `{ encoding: 'plain', value }` still fails the scan while the
file-exists and gateway-URL guards pass. The migration case is a documented
`test.fixme` naming its three blockers.
Electron project 928 -> 924 tests (-9 migration, +5 new guard and
mechanism-isolation). Two of those five exist because reverting either owner-only
mechanism alone initially scored zero failures — they were masking each other, so
either could have been deleted green.
(cherry picked from commit 6e01add6578f08f015a567d3a7a7378f2ec3e768)
Typing while the turn was parked on a blocking prompt routed the text
through steer (session.redirect), which sat undelivered behind the
blocked tool batch — nothing rendered, and stopping the turn to force it
through resolved the prompt to empty and ended the turn as the literal
"Operation interrupted." row, eating the message.
Clarify already had a carve-out (typing skips the question and steers)
because a real message IS an answer to a clarify. Approval/sudo/secret
have no such answer path, so the busy submit now queues the words as the
next turn instead: the prompt stays answerable, the queue drains on
settle, and the busy button advertises queue rather than steer while one
is pending. Slash commands still execute inline, and another session's
prompt never affects this one.
Right-clicking anywhere the app owns no menu of its own — the titlebar
gutter, an empty pane body, the sidebar background — did nothing at all.
Electron's native handler bails on non-editable, non-selected content by
design, so those surfaces had no menu to fall back to.
Wrap the shell in a fallback context menu carrying the verbs that belong
to the window rather than to a row: new session, new window, command
palette, toggle the status bar, settings, update Hermes. Every row reuses
the store action and the copy its Cmd+K twin already uses, so the two
can't drift.
A guard on an inner element keeps it a fallback: a right-click that lands
inside a surface with its own context menu, on an editable, or on a live
selection stops propagating before Radix's trigger sees it, leaving that
surface's menu — or Electron's native edit menu — in charge.
A pasted GitHub PR comment deep link (#discussion_r… / #issuecomment-…)
now lands as a typed review attachment instead of a bare url chip. The
card attaches optimistically and resolves through gh in the background —
author, file:line anchor, body, and the diff hunk — expanding at send
into an anchored fenced block, so "address this" carries exactly what
"this" is. When gh can't answer (offline, unauthenticated, foreign repo,
remote gateway) the card downgrades to the plain url ref and nothing is
lost.
Queued turns could only wait for the settle or interrupt the turn to jump
the line. Text-only queue entries now carry a steer action while the agent
is busy: the entry rides the existing mid-turn redirect (no interrupt, no
drain lock), is consumed only when the gateway accepts it, and lifts a
park so the rest of the queue keeps flowing. Slash commands and entries
with attachments keep their existing semantics.
An inbox card whose session is working through a todo list shows its "X/Y"
fraction at the far right of the footer line, updating live as items
complete. The count projects the runtime-keyed todo map to stored session
ids through the same lineage-alias fallback the working/attention
projections use, emits pre-rendered strings so unchanged rows never repaint,
and skips cancelled items on both sides of the fraction.
Hovering a truncated inbox-card title glides the clipped tail into view —
one direction at constant speed, a short hold at each end, then a snap back
to the start. Overflow is measured on pointerenter and the animation arms
only when the text actually clips, so short titles never wiggle. State lives
in DOM attributes and CSS variables, so a hover never re-renders the
memoized row, and the blanket reduced-motion override already disables it.
A new "Inbox style" toggle in the sidebar filter menu renders the flat
recents list as cards: a workspace header line (project when it resolves,
else the cwd leaf, else Home) with the age at its right edge, the title
grouped with a one-line last-message preview, and a model + size footer.
The preview line ships on by default and has its own Show-menu toggle,
offered only while Inbox style is active — the one-line row has nowhere
to put it.
A render variant, deliberately not a grouping — it composes with whichever
grouping is active and only the flat recents list opts in; pinned, project,
and messaging surfaces keep the one-line row. Spacing hangs off a single
--card-gap variable; the title/preview pair is one grouped cell with its own
tighter internal gap. The age and kebab sit in flow inside the header line
rather than a full-height side column, so title, preview, and footer span
the card's entire width.
The card's project label reads through a selector that resolves the label
string, so tree polls with fresh atom identity repaint only rows whose label
actually changed.
The virtualized recents list sat inside a wrapper that was itself a themed
scroller, so two 4px classic-scrollbar gutters stacked and every row ended
8px short of the sidebar edge the pinned list reaches. Drop the wrapper's
scroller when the virtual list owns scrolling, use the overlay scrollbar on
the virtual scroller (zero gutter, native fade), and neutralize both axes of
the wrapper overflow — `overflow-y-visible` next to `overflow-x-hidden`
computes to `auto` and still reserved a gutter.
The idle dot variant had no background of its own, so a settled session
outside any project rendered a literally invisible dot — the row read as
missing its status indicator next to its neighbours. Fall back to the
faintest filled grey; a project color still wins when there is one.
Todo, subagent, background, and preview titles in the composer status
stack were capped at an arbitrary max-w-[18rem], ellipsizing long items
way before the row ran out of space. The spans already live in the
shared StatusRow's min-w-0 flex-1 content slot, so plain truncate gives
correct overflow at the actual row edge — drop the cap.
The branch's PR chip carried its own pull-request icon next to the row's
branch glyph, so the strip opened with two git marks in a row. The chip now
takes showIcon, and the coding row renders it glyph-less and ahead of the
branch name — the leading branch icon covers both, and the row reads
icon → #number → branch. Sidebar rows keep the full chip.
The background review's summary fell through SystemMessage's generic branch: centered, 60% width, muted grey — while the memory tool row it reports on wears the gold-to-purple legendary chrome. Type the event with a review: marker at the gateway handler (same convention as steer: / slash:) and give it the brain glyph, gradient label and purple detail, left-aligned in the reading column.
