Merge-time fix for #666. `recordHookEvent()` ran for every `stop`, the run
summary files a `stop` as a warning, and the Away digest lists warnings under
Needs Attention, so each keep-alive nudge added a row there and to the
session's Summary timeline. The hook route now skips the run summary while
`keepAliveReason` is set (the keep-alive turn's `stop` and `idle_prompt`).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Merge-time fix for #666. The strict settings reader plus the retry after a
throw is right for a read that lands mid-write, but a settings.json that
stays broken (a hand edit with a trailing comma) made every session, keep-warm
off included, re-read the file and log a line every 60 seconds until it was
fixed (11 reads and 11 log lines in ten simulated minutes for one session).
The settings read now fails inside `evaluate()`: a given failure is logged
once until a read succeeds, and the retry is armed only for a session that
could nudge (its choice not `false`, the exporter `injected`, a warm report).
Other throws keep the generic retry in `schedule()`.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Merge-time fix for #666. A nudge only goes out inside the last five minutes
before `expiresAt`, so the check armed a whole `KEEP_WARM_LEAD_MS` after a
failed write landed at or past the expiry and stopped with "too close to
expiry": the retry the status promised never happened. The window
`beginKeepAliveTurn()` opened also stayed open for its ten-minute ceiling
with no turn running.
A failed write now ends the window (`endKeepAliveTurn()` joins
`KeepWarmSessionView`; `Session` already has it) and re-arms at
`min(now + KEEP_WARM_RETRY_MS, expiresAt - KEEP_WARM_LAST_CHANCE_MS)`. The
test advances the clock and sees the second write, and a second one pins
that the retry never lands past the last-chance line.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude Code draws its prompt suggestion in an empty composer, the first letter in the reverse-video cursor cell and the rest faint, so a plain capture read it as typed text and keep-warm never nudged those sessions. The keep-warm pane read now uses capture-pane -e and drops the suggestion before checking the composer.
Also from review: an off status carries no window, automation and an unreadable composer retry instead of stopping, a failed evaluation re-arms a retry, a trigger during an evaluation queues one follow-up that remove() cancels, and the cap's clock is seeded from lastSubmitAt for pre-upgrade state.
Only a human spends an idle alert, but the keep-alive turn resolved an unseen one three ways: its working transition, its stop hook, and its own idle prompt replacing the item with a pre-acknowledged one. While the turn runs, a pending idle item is now left exactly as found, the page's stop handler returns on the keep-alive reason before clearing hooks or raising Response Complete, and session:working/session:idle carry keepAlive so the stuck-detection timer neither restarts nor re-arms.
The window that marks the turn closes on its own idle prompt or any other submit instead of a fixed three minutes, which a slow turn's idle prompt (sent a minute after the turn ends) could outlast. The service keeps the default and cap in memory, read through a strict reader so a mid-write read is retried rather than kept, and the cap is clamped before the strict settings save.
Claude Code lets a subscription's one-hour prompt cache expire after an hour idle, and the first turn back re-writes the whole conversation at the cache-write rate; nearly every gap over an hour goes cold, while the saving is concentrated in gaps of one to four hours. Codeman now sends one short keep-alive turn five minutes before the expiry Claude Code reports on its statusline, so the next real prompt still reads the conversation from cache, and stops two hours (configurable) after the last real prompt.
Codeman owns the idle policy for these sessions, as agreed in Discussion #554: a launch with keep-warm on also carries idleCompaction: false in its one --settings JSON, so Claude Code's own idle compaction cannot replace the conversation with a summary a minute before the nudge. It is decided once per launch (keep-warm on, the statusLine exporter injected, Claude Code 2.1.290 or newer, an unknown version fails closed like --name), recorded on the mux record, and picked up by any relaunch; a launch with keep-warm off is byte-identical. A session switched on after launch is kept warm best-effort and the readout says so.
Every guard lives in one pure rulebook: the opt-in (an agent-spawned session does not inherit the app default), a launch that reports cache state, a warm 1h cache with enough tokens at stake and no compaction this gap, a live pane, no usage-limit pause, no respawn, Ralph, task or waiting agent, idle, no open dialog, an empty composer and no recent keystroke. The nudge is a submit and moves lastSubmitAt, which the response viewer's conversation correlation needs; the cap counts from its own persisted lastRealPromptAt instead. Read My Mind drops the exact nudge text, and the turn's stop hook and idle prompt open pre-acknowledged so nothing alerts. The last cache report and the clock survive a restart, so the timer re-arms.
A phone in landscape is wider than 599px, so the 44px targets now follow (pointer: coarse) instead of the width. A balanced title broke the zh-CN text inside 入侵; :lang(zh) keeps CJK runs whole and breaks at the title's spaces and colon.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A closing notice card stays in the DOM for its 200 ms fade, and its
link and "See what's new" buttons still worked there: a double click
opened the notice twice. close() now makes the card inert and every
button handler does nothing once it is closing.
- Notice links (and the combined card's "See what's new") open with
noopener,noreferrer. noopener alone still sent this instance's origin,
which can be a tailnet name or a tunnel URL, as the Referer. The plain
update toast is unchanged.
- The https rule is checked again at the button
(CodemanSecurityNotices.urlOk): a caller handing showSecurityNotice or
showUpdateNotice an unchecked URL gets no link button.
- A detached session window (/session/:id) is rendered without
__codemanCliAvailable, so it could never select the OpenCode notice,
yet it took the update toast and recorded the version, and the
dashboard then showed the notice as a separate card. While any notice
is pending, a solo window now shows and records nothing and leaves
both to the dashboard; with none pending it shows the plain toast as
before.
- The wiring tests pin the page's own clock (JSDOM gives the window its
own Date). On the real clock 18 of them would have failed on every push
from 2027-01-01, when the fixture notice expires. New tests pin the
once-only close directly (the old double-click comment claimed more
than it checked), the single open, the sink check, the solo windows,
the scroll cap and the light-skin ink.
- CLAUDE.md now says the public .json files are served no-cache, the
exception to the year-long immutable cache.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The card is capped to the room below the toast stack's top and
scrolls inside itself. The stack is fixed and cannot scroll, so on a
phone in landscape the update toast's buttons sat below the screen
with only the × left, which records the notice unread. The accent bar
becomes a background layer, which stays put while the content scrolls.
- Small amber text and the primary button use --notice-ink, darkened a
step on the four light skins, where --yellow measured 3.9 to 4.4:1 for
the 10.6px eyebrow (now 5.1 or better, the button label 5.7 or better).
- The update toast that carries a notice frames itself in the accent
colour and leaves the amber to the warning section inside it.
- The × uses --text-dim (2.4:1 on og before, 5.5 now).
- Phones: every button fills its row (a wrapped Dismiss no longer sits
stray at the left) and is 44px tall, and the × gets a 44px box.
- The title balances its lines, and the card matches the plain toast's
420px so a toast stacked under it lines up.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The message ran 66 words over five sentences, which pushed the update
toast's buttons off a 360x640 phone. It also repeated the card's own
eyebrow ("Security notice for OpenCode users:"), placed the breached
"legacy system" inside the Console (no source says so), and closed with
an unsourced line about the CLI not being affected. The new text (55
words) says what OpenCode said, including its own reassurance that a
user who got no email was not affected, which matters because the card
goes to every opencode user while about 1.4% of Console users were hit.
The zh-CN text follows it and uses 撤销 for revoking keys.
published moves to 2026-10-17, so the notice is already live in every
timezone when the release lands: a browser still on the 17th would
otherwise take the plain update toast and get the notice later as a
separate card. docs/releasing.md says both things for the next notice.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
src/web/CLAUDE.md gets the load-bearing rules (hand-maintained file,
strictly-true availability, one surface per notice per load, seen only
on a user action, no CLI-id branching, theme tokens only), and
docs/releasing.md a section on adding a notice, its fields, and letting
`expires` retire it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The pure select() and plan() loaded from the shipped constants.js, the
shipped JSON itself (unique ids, https links, real dates, every cli a
stock registry id, a zh-CN title wherever there is an English one), and
the wiring in a JSDOM page: fresh install and same-version reload get the
card, an existing install's update toast carries the notice, the update
toast without a notice matches master's DOM byte for byte, seen only on a
user action, storage blocked shows nothing, base path, zh-CN, and the
card's CSS contract (theme tokens, motion, phone width).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A hand-maintained src/web/public/security-notices.json carries warnings
about things outside Codeman that its users should act on. The first:
OpenCode's report of unauthorized access to its legacy Console, which
may have exposed provider API keys saved there.
- Pure core in constants.js (CodemanSecurityNotices): select() keeps the
notices that apply now (inclusive local-day dates, not yet seen, https
links, malformed entries skipped), and one naming a CLI only when
window.__codemanCliAvailable[cli] === true. Unknown means not
installed: a warning about software you do not have is noise. The CLI
id is data in the JSON; no code branches on it. plan() decides the
surface, so a load shows each notice once.
- An existing install on a new version gets the notice INSIDE the
release's update toast: the highlights as before, a warning section,
and the notice's link next to "See what's new". With no notice the
update toast is the plain one it always was.
- Everyone else (a fresh install, a same-version reload) gets its own
card in the toast stack: shield, "Security notice" eyebrow, title,
message, the link, Dismiss and the ×; amber-to-red accent bar, theme
tokens only (dark and light skins), full width on phones, an
opacity/transform entrance and none under prefers-reduced-motion,
role=alert.
- A notice is recorded as seen only when the user closes it, dismisses it
or follows its link, never when it appears. Storage blocked shows
nothing. The notice's own text is textContent and carries its own
zh-CN, so i18n.js leaves it alone; the card's own copy translates.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The static plugin gave every file under public/ a year of `immutable`,
including the JSON the page fetches at runtime: whats-new.json (rewritten
every release), the new security-notices.json (edited by hand) and
manifest.json. None of them is content-hashed or `?v=`-busted, so a
browser could keep last release's copy, and a security notice added in a
later release might never reach it. A `.json` file now revalidates
(`no-cache`); every other static asset keeps its immutable year, and a
route that set its own Cache-Control still wins.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
GitHub Copilot CLI (`copilot`) becomes a run mode: Run menu, welcome
screen, cron and CLI management once it is installed, tmux-backed like
the other agent CLIs, working/idle and the model read off its footer,
launch options through `copilotConfig` (model, resume, continue, the
multi-user-clamped allowAll), Custom Model Endpoints via Copilot's BYOK
variables, MCP sync of its mcp-config.json, a status probe, and the
`codeman agent` verbs (spawn waits for its composer, read returns its
last answer).
The plan-usage quota row that earlier rounds carried was split out on
review and is not part of this merge.
Merge-time commits on top: the wider COPILOT_* clamp and the zh-CN
README rows (maintainer), the Docker sign-in docs, divider rows inside
an answer kept by `agent read`, route tests for the screen-read
last-response branch, and doc drift (Past Sessions, wheelForward JSDoc,
COPILOT_/OMP_ in the env prefix lists).
Verified before merge on Copilot CLI 1.0.98: injected keystrokes,
bracketed paste and Enter reach the TUI in tmux (github/copilot-cli#4180
does not reproduce), a shell tool call completes inside tmux (#4223 does
not reproduce), and a session driven through this branch answered over
send-and-wait.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
docs/api-reference.md (the signal notes and the idle/working rows), and the
wiki's Notifications-And-Approvals, Keeping-Agents-Running and Troubleshooting
pages, plus docs/grok-integration.md, still called Antigravity's working and
idle status unreliable. It now declares a working line and prompt marker like
Codex, OpenCode, Gemini, Pi and OMP, so only Grok stays on generic cues. The
Troubleshooting entry for a session stuck showing busy also says that an
Antigravity session waiting on a shell-command approval shows busy until it is
answered, and the API reference says the same of its idle signal.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
_probePaneWorking() tests the pattern against the whole pane, so an answer
that quoted the spinner row (an agent explaining this entry: it matches
`⣯ Generating...`) kept the session busy for as long as the text was on
screen. Anchored like gemini's braille branch; every agy capture draws the
spinner at column 0. The prose test now covers the quoted row.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Master's #621 added tsv and zip to both file-link extension groups in
constants.js while this branch added diff|patch to the same lines. The
resolution keeps both, in master's order with diff|patch after sql where the
PR put it; the rooted and relative groups stay identical, and every upload
document extension plus .diff/.patch still links in both patterns.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Its prototype-only app had no _pendingDeliveries and only passed because the
removed try/catch swallowed the read. The real constructor always creates the
Map.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The parked-viewport test had no wrapped rows and the termination test runs at
baseY 0, so nothing pinned the walk of a multi-row composer through absolute
lines off the bottom. The new case fails on the pre-fix terminal-ui.js
('link' instead of 'history') and passes with the fix.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The new App Settings row had no zh-CN entry, so it read in English in a
Chinese UI, and docs/wiki/Settings-Reference.md did not list it. Adds the
label and description to i18n.js, a row in the Agents & CLIs table, and a
JSDOM test that runs the real translator over the real index.html.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- docs/cli-registry.md said Session history resume goes through
copilotConfig.resumeSessionId / continueSession. Past Sessions sends only
`continueSession: true` (terminal-ui.js, the modeConfigKey table);
resumeSessionId is for API callers.
- The wheelForward JSDoc in cli-registry/types.ts read as if the value had
effect. Nothing reads it (it is in the DECLARED-FOR-LATER list); the
browser gate is _shouldForwardWheelToApp in terminal-ui.js with its own
per-CLI rules, which the comment now says.
- The env-prefix lists in the CLAUDE.md Security table and the
allowedEnvPrefixes() comment in schemas.ts stopped at DEEPSEEK_, missing
both OMP_ and the new COPILOT_; the Multi-CLI prefix rule missed OMP_.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The suffix is part of the pill's text line, which the Tiles style hides (it
shows WS live) and phones hide too, so the page promised a suffix some users
never see.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The timer test only checked that a handle existed, and the other hold tests
re-rendered by hand, so deleting the re-render from the timer callback left
every test green. A new test waits out the hold for real and asserts the
suffix appears and the timer is spent with no further call.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The try/catch around _pendingBytes() could not keep _updateConnectionIndicator()
from throwing, because _computeConnectionDescriptor() read the queue again,
unguarded, right after; its test only passed because it stubbed the compute
out. The update now reads the queue once and hands the totals to the compute
(which still reads a fresh queue when called on its own), so the hot input
path walks the queue once per keystroke and ACK instead of twice. The stubbed
test is gone: the COD-133 combinatorial test already pins never-throw.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Now that the set is exported for the parity test, type it read-only like
VIDEO_ATTACHMENT_EXTENSIONS, so an importer cannot add to the live upload
allowlist.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
GET /api/sessions/:id/last-response reads a copilot pane through
capturePaneText, since a repainting full-screen TUI streams cursor moves
rather than lines, and nothing pinned that at the route: had the branch
stopped capturing, copilot would quietly have been segmented off the
fused stream again. Three app.inject cases with a stubbed mux: the
captured screen is what gets segmented (the stream text never shows),
a null capture falls back to the buffer, and a line-streaming mode
(opencode) never asks for the screen. The first two fail with the
capture switched off.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The reverse classifier check read only the one includes(ext) list, so a
second branch such as `if (ext === 'rtf') return 'document';` passed every
test; the method may now return 'document' from exactly one place. The
accept= checks kept only .ext tokens, so adding text/* to a picker passed;
the non-extension tokens must be exactly image/* and video/*. Both verified
by mutation (each fails the suite).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The pane segmenter ended the response block at every full-width rule, so
an answer that draws one (a markdown `---`, a passage quoted between two
rules) read as only its last section: `Summary / ──── / Details` came
back as "Details", and the existing test pinned that by expecting the
quoted passage without its "Here it is:" lead-in.
Inside an open response a rule is now held and written into the text as
`---` once more of the same answer follows, so a rule that only separates
the answer from the footer never trails it. The composer's own two rules
(the last rule/❯/rule triple) are still boundaries. The old test now
expects both parts, and a new one pins a Summary/rule/Details answer as
one block; both fail without the change.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The reflow around the new UTF-8 sentence left one 120-character line; the
paragraph is wrapped at about 90 columns like the rest of the page.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Pins the user-facing fix end to end: POST /api/cases with a path two levels
down returns 200, lands in the linked-cases registry and is scaffolded.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The one message read "That folder is in the cases folder (...), where it is
already listed as a case" for the cases folder itself too, which is not in
itself and is never listed, and "already listed" is not true for a direct
child that does not exist yet. Two messages now, both still pointing at plain
Create New.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
docs/api-reference.md and the case-path.ts @fileoverview still said a custom
folder must not be, or be inside, the cases directory. After #567 only the
cases directory itself and its direct children are refused (those are what
GET /api/cases lists as local cases); a folder further down is allowed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
CLAUDE.md (Prompt uploads) and architecture-invariants still named
_promptAttachKind() as the filter every surface reads. That stays true, but
the surfaces now admit through _admitPromptAttachments(), which is what says
how many files a pick skipped. A new surface that filtered on the classifier
directly would drop files silently.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Both Docker sections said "or set COPILOT_GITHUB_TOKEN for the session",
which cannot work: a docker quick-start refuses envOverrides
(session-routes.ts), and Copilot's dockerExecEnvNames is empty, so a token
in the server environment is not forwarded into the container either. The
sentence now says to sign in inside the container and why the token route
does not exist.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The PR names GitHub Copilot CLI throughout README.md (ten CLIs, the
install list, the run-mode table, the multi-CLI prefixes,
the agent-read note and the architecture diagram), and README.zh-CN.md,
which is kept in parity with it since 1a692f58, still said nine CLIs and
never mentioned Copilot. The same spots are translated here (not the plan-usage line: the quota
row was split out of the PR, so neither README names Copilot there).
Both READMEs' "Schema-validated inputs" bullet lists every admitted
env prefix and missed the new COPILOT_* one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The COPILOT_ prefix admits every variable Copilot reads, and the 1.0.98
bundle reads far more than `copilot help environment` lists. Only
COPILOT_HOME, COPILOT_ALLOW_ALL and the BYOK trio were privileged, so a
non-granted multi-user owner could still hand back what the allowAll
clamp removed:
- approval: COPILOT_PLAN_THEN_AUTOPILOT, COPILOT_ASSISTED_APPROVAL
- code with no approval row: COPILOT_CLI_DIST_DIR runs <dir>/index.js
in place of the CLI, the package caches choose the CLI build,
COPILOT_EDITOR, COPILOT_PROVIDER_API_KEY_COMMAND and
COPILOT_AHP_HOST_BIN are executed, skills, provider configs and hooks
are loaded from the named places (the DSH_HOME class)
- token redirects: COPILOT_API_URL and the other API bases receive the
CLI's sign-in token, which in multi-user mode is the operator's (all
users share the OS account and ~/.copilot), and the rest of the BYOK
provider family re-aims a Codeman custom endpoint (the
DEEPSEEK_BASE_URL / OMP_AUTH_BROKER_URL class)
All are now privilegedEnvKeys. The registry schema capped that list at 8;
the cap is now 48, since a longer deny list only takes more away from
non-granted owners. The test sends one of each class through the real
schema (they all pass it) and the real clamp. The docs say the list is
read from the 1.0.98 bundle and that a variable a later release adds is
not covered until it is listed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Add pinCliSessionName (synced, default OFF), re-read at each spawn and respawn.
Off keeps the #457 behaviour; on pins every non-blank tab name as --name.
Update the docs, JSDoc and changeset that stated the old rule.
Codeman maintainer