* fix(desktop): decode CLI output across pipe chunks so split UTF-8 characters survive
* fix(desktop): cap serve answers by decoded size and fall back instead of killing the resident
* perf(desktop): scan only the new chunk for a serve line break
* fix(desktop): an over-cap answer fails its request without a cold retry
* fix(desktop): an oversized serve frame fails its request without a cold retry
* fix(desktop): give the serve frame room for an answer that escaping doubles
* test(desktop): feed serve frame boundaries and split UTF-8 as explicit chunks, count the newline scan
The Sessions drill-through asks for contribution segments over the whole
population, and over all time that answer is 28MB, past the 16MiB cap:
the request failed and the resident child was killed with it. The CLI
now answers one of n fixed slices with --part i/n, keyed by session
identity so a re-derived report never duplicates or drops a row, and the
bridge fetches eight slices in turn and restores the report order. The
renderer receives the same rows in the same order as before.
The resident serve child answers one request at a time. A background
compare (about a minute) or optimize (10 to 70 s) made any click queue
behind it. Both payloads are small and persist, so a revisit still
paints from the snapshot; the first visit pays its own time only.
Every subagent's step list ships with the session since the agents view,
so a session with 863 helpers grew from 10.8 MiB to 42.4 MiB and the
desktop refused it. Steps now leave out default fields (isError false,
answer false, empty detail, calculated basis), the unused context delta,
and round the cached share. Past a 12 MiB budget the JSON drops whole step
lists, cheapest helper first and then prompts, with a truncatedSteps count
the Inspect, Agents and Story views show.
* feat(desktop): inspect view for a prompt with request cards and insights
An open prompt now reads as one card per model request, newest first, with
the tool calls each request asked for nested under it. Each card shows the
request number and model, tokens in, out and cached share, what the request
did (asked for tool calls or wrote the answer), its duration against the
prompt's longest step, and its output tokens and cost. Tool rows show the
command or target, duration, the context the result added, a parallel label
when one message asked for several calls, and a muted failed state. A search
field and an All / Model / Tools switch narrow the list, and any card or tool
opens the existing step detail.
Above the cards a summary line and an insights strip name the slowest and the
costliest request, the tool that added the most context, how the tool calls
ran, and failed steps when there are any. Rows that cross the existing
slow-call and carried-result thresholds lead. The two-lane waterfall is gone:
the cards carry the same timing.
The session payload gains the fields this needs, computed at serve time:
each tool step names the model step that asked for it, a parallel position
and its result size; each model step carries its cached share, the context
delta to the next request, its tool call count, its cost basis and whether
it wrote the answer.
* fix(desktop): one place for out tokens on a request card
The request card showed output tokens twice, in the meta line and as a pill. The pill is gone; the meta line keeps the count and the dollar pill stays.
* feat(desktop): agents view with each subagent's lifecycle and cost by agent
* feat(desktop): story view that tells a prompt in plain sentences
* fix(desktop): story wording and copy feedback
* feat(desktop): inspect view for a prompt with request cards and insights
An open prompt now reads as one card per model request, newest first, with
the tool calls each request asked for nested under it. Each card shows the
request number and model, tokens in, out and cached share, what the request
did (asked for tool calls or wrote the answer), its duration against the
prompt's longest step, and its output tokens and cost. Tool rows show the
command or target, duration, the context the result added, a parallel label
when one message asked for several calls, and a muted failed state. A search
field and an All / Model / Tools switch narrow the list, and any card or tool
opens the existing step detail.
Above the cards a summary line and an insights strip name the slowest and the
costliest request, the tool that added the most context, how the tool calls
ran, and failed steps when there are any. Rows that cross the existing
slow-call and carried-result thresholds lead. The two-lane waterfall is gone:
the cards carry the same timing.
The session payload gains the fields this needs, computed at serve time:
each tool step names the model step that asked for it, a parallel position
and its result size; each model step carries its cached share, the context
delta to the next request, its tool call count, its cost basis and whether
it wrote the answer.
* fix(desktop): one place for out tokens on a request card
The request card showed output tokens twice, in the meta line and as a pill. The pill is gone; the meta line keeps the count and the dollar pill stays.
* feat(desktop): agents view with each subagent's lifecycle and cost by agent
* feat(desktop): inspect view for a prompt with request cards and insights
An open prompt now reads as one card per model request, newest first, with
the tool calls each request asked for nested under it. Each card shows the
request number and model, tokens in, out and cached share, what the request
did (asked for tool calls or wrote the answer), its duration against the
prompt's longest step, and its output tokens and cost. Tool rows show the
command or target, duration, the context the result added, a parallel label
when one message asked for several calls, and a muted failed state. A search
field and an All / Model / Tools switch narrow the list, and any card or tool
opens the existing step detail.
Above the cards a summary line and an insights strip name the slowest and the
costliest request, the tool that added the most context, how the tool calls
ran, and failed steps when there are any. Rows that cross the existing
slow-call and carried-result thresholds lead. The two-lane waterfall is gone:
the cards carry the same timing.
The session payload gains the fields this needs, computed at serve time:
each tool step names the model step that asked for it, a parallel position
and its result size; each model step carries its cached share, the context
delta to the next request, its tool call count, its cost basis and whether
it wrote the answer.
* fix(desktop): one place for out tokens on a request card
The request card showed output tokens twice, in the meta line and as a pill. The pill is gone; the meta line keeps the count and the dollar pill stays.
* feat(claude): attribute usage to the account that was active
* fix(claude): identify by organization and keep the identity field under a filter
* test(claude): check the log file mode on POSIX only
* chore(telemetry): default on in every region, consent screen and toggle unchanged
The desktop app, the macOS menu bar and the Windows tray no longer pick the
telemetry default from the country. A fresh install starts with telemetry on
everywhere, including the EU, EEA, UK, Switzerland and an unknown region.
The country lists and the helpers that read them are gone.
A stored choice, on or off, is kept as before. The first-launch flow and the
Settings toggle are unchanged. The README and the code comments now say the
default is on in every region and that Settings turns it off.
* docs(telemetry): say what each surface does
With only the desktop app installed from the Homebrew cask, the menu bar
failed to load Today with "nonZeroExit (code: 127, stderr: env: codeburn:
No such file or directory)". The menu bar looked for a codeburn in the
path recorded for it and in package-manager folders, then fell back to the
bare name. A desktop-only machine has nothing in those folders, so when the
record was missing or no longer ran, nothing was found and the error named
nothing.
The packaged desktop app now carries a plain launcher at
CodeBurn.app/Contents/Resources/cli/codeburn that runs its bundled CLI
through its own executable and finds its own bundle. The menu bar checks
it, in /Applications and ~/Applications, after the recorded path and the
package-manager folders. When none of them runs, Today says which paths
were tried. The CLI's menubar installer records the same launcher instead
of refusing the install when nothing named codeburn is on PATH.
* feat(optimize): flag an autoCompactWindow that blocks idle compaction
* fix(optimize): render a finding without a dollar figure cleanly and gate the idle-compaction rule on the one-hour cache
* fix(optimize): group headers without a figure when the group has none
Anthropic renamed the Max 20x plan to Ultra, but the credential's rate-limit tier id is still default_claude_max_20x. A parsed multiplier of 20 now labels Ultra in the CLI, desktop, Windows and menubar twins; other multipliers keep Max Nx.
* feat(sessions): mark sessions kept after their transcript was deleted
When a write refresh keeps a cached session whose source file is gone
(a Claude session with PR links, or a copilot or quickdesk entry inside
its 90-day window), the cache entry now records retainedSince, the local
date of the first refresh that found the file missing. It is set once and
kept on later refreshes. Offline WSL roots are not deletions and stay
unmarked.
Sessions served from such an entry carry retained.since. The sessions
JSON rows include it, the sessions table puts * after the start time with
a legend line, and the export adds a last retained column to records and
sessions in JSON and CSV. The desktop Sessions list and drawer show a
muted note naming the tool and the date.
Behaviour does not change: the same entries are kept and evicted, and
every total and the by-PR numbers are unchanged. The field is optional,
so the cache version stays.
* fix(sessions): say when the transcript was found missing, not deleted
The desktop note, the sessions table legend, the export readme and the docs no longer give a deletion date. retainedSince is the day CodeBurn first found the transcript missing, so the note now reads that the transcript has been missing since that date.
* fix(export): give the JSON export stable keys
The JSON export reused the CSV row builders, so its keys were display
headers with spaces and parentheses, and the money keys changed name
with the configured currency. A program had to reshape the file before
it could read it.
The builders now produce one camelCase row, and a single key to header
map writes the CSV, so every CSV header line stays byte-identical. The
JSON schema moves to codeburn.export.v3; the currency is stated once in
currency.code and the money values stay converted to it.
Before (v2, EUR):
{"Period":"7 Days","Date":"2026-11-02","Cost (EUR)":1.1,"Saved (EUR)":0,"API Calls":1,"Sessions":1,"Input Tokens":100000,"Output Tokens":50000,"Cache Read Tokens":200000,"Cache Write Tokens":30000,"costBasis":{"calculated":1}}
After (v3, EUR):
{"period":"7 Days","date":"2026-11-02","cost":1.1,"saved":0,"apiCalls":1,"sessions":1,"inputTokens":100000,"outputTokens":50000,"cacheReadTokens":200000,"cacheWriteTokens":30000,"costBasis":{"calculated":1}}
v2 consumers must rename keys. New keys can still appear under v3.
* docs: changelog entry for the v3 export JSON
* test: read the v3 export keys
The sessions table, the interactive sessions view, the JSON rows and the
desktop Sessions page now show a cache hit rate per session. It is cache
reads divided by input plus cache reads plus cache writes, the share of the
prompt that came from cache. A session with no cache reads or writes is
unmeasured and shows a dash instead of 0%.
The session drawer, the compare views, the menubar payload and the JSON
report now use the same formula through one helper, so every surface
reports the same number. Optimize already used it, so its finding does not
move.
* chore: drop the no-zstd fallbacks the 22.15 floor made unreachable
The dsh, zed and openclaw providers still guarded against a Node whose
node:zlib has no zstd. The floor is 22.15 now, so they call
zlib.zstdDecompressSync directly like the shared helpers do. The tests that
exercised the missing-zstd path and the guards that skipped zstd cases on old
Nodes are gone.
* docs: state the 22.15 zstd floor plainly and name the tools degraded mode cannot read
* fix(quota): label every Claude Max multiplier and the Ultra tier
The plan label only knew max_5x and max_20x, so any other Max multiplier showed as Max 5x and an Ultra tier showed as Subscription. The multiplier is now parsed from the tier id, a tier containing ultra is labelled Ultra, and an unknown tier falls back to its readable text. Applied to the CLI quota code, the desktop app copy, the Windows tray and the macOS menubar.
* fix(quota): keep a bare Max subscription labelled Max
A Max subscription whose tier carries no multiplier fell through to the readable fallback or Subscription. It now labels Max in all four copies.
* chore: raise the Node floor to 22.15 and make zstd unconditional
Node's built-in zstd in node:zlib arrived in 22.15. With the floor at 22.13
every reader of Codex's compressed .jsonl.zst rollouts had to check for it,
and the daily cache kept a separate generation for Nodes without it so a later
Node upgrade would still re-derive their days. That split could not survive
another cache bump.
The engines field, the launcher's version check, .nvmrc, the README and the
CI pins now say 22.15. The VS Code extension applies the same floor when it
picks a Node to run the CLI, and its runtime notice names it.
Removed: the optional zstd exports, zstdSupported and requireZstd in
fs-utils, the skipped-compressed-rollouts notice in launcher-homes, and the
76/77 split in the daily cache, which is now 77 for every Node.
* docs: say 22.15 wherever the Node floor is stated
vitest 5 declares vite as a peer dependency only, so root and vscode now
name vite ^7.3.7 themselves. Without that, a fresh lock resolves the peer to
vite 8, which vscode's esbuild pin rejects and which pulls rolldown and
lightningcss into the root install.
The desktop app and the VS Code extension read quota through their own
copy in app/electron/quota, which only knew the unsuffixed "Claude
Code-credentials" Keychain item and ~/.claude/.credentials.json. Claude
Code keeps the login of any other config dir under "Claude
Code-credentials-" plus the first 8 hex chars of the SHA-256 of the
dir's NFC path, so on macOS that login read as not connected.
The Claude row now reads the first configured dir, in the CLI's order:
CLAUDE_CONFIG_DIRS, CLAUDE_CONFIG_DIR, the claudeConfigDirs the menubar
writes to config.json, then ~/.claude. An expired credential file no
longer hides a live Keychain login; a denied Keychain still reports
accessDenied, and with no Keychain item the expired file is still the
answer. A background poll, which never reads the Keychain, reports an
expired file on macOS as not checked, so the login a forced check found
is held instead of flipping to expired.
Keychain reads still happen only on a forced refresh.
From the third launch of the desktop app on, once the splash has gone, a small centered dialog asks to star CodeBurn on GitHub. It shows the GitHub mark the About window already uses, the heading "Enjoying CodeBurn?", one muted line, "If it has been helping you, a star on GitHub would mean a lot. It keeps the project free and open.", a primary Star on GitHub button that opens the repository through the existing open-external bridge, and Not now. Either button marks it done and it never shows again. Escape or a click outside the dialog only closes it for this start, so it comes back next launch, as does ignoring it.
Launches are counted in the renderer's localStorage (codeburn.launchCount), where the other shell settings already live, and the done flag is codeburn.starNudgeDone. Nothing checks whether a star was given. The VS Code extension host does not show it; the sidebar there already has its own star line.
The dialog reuses the About modal's backdrop, panel and animation classes and the app's button styles, with three CSS rules for its width, heading and button row. Splash gets an optional onDone callback so the dialog opens only after the fade-out. Strings are in all six catalogues because the catalogue test requires it; the five translations have not been reviewed by a native speaker. No telemetry event was added, so the measure is the star count on GitHub itself.
Typecheck clean. StarNudge test 8 passed, catalog test 8, App test 72, Splash test 7. Screenshot taken from the renderer served through demo-bridge with the launch counter set to three.
The Windows tray's Capacity Dock re-seeded its provider rail and resting
provider from whichever providers answered the last quota refresh, so one
missed Codex read made Cursor the resting provider for good. autoSeed is
now additive: providers already in the rail stay, a saved preferred
provider is kept while the CLI still knows it, and every resting pick
(ring click, Resting menu, Electron relay) is latched in the shared dock
prefs. A separate-rings Claude profile row id never enters the rail,
because autoSeed only pins ids the CLI reports.
The mac menu bar keeps its own seeding and is unchanged; it likely has the
same eviction and gets a parity follow-up. The new tests cover the seed
and latch rules; four of them fail on main.
Cursor 3.27 can store its app session as cursor-safe-storage:v1 ciphertext
(Chromium v10, key in the "Cursor Safe Storage" keychain item), behind
Cursor's own cursor.storage.authTokenEncryptionAtRest flag. CodeBurn read it
as an expired login. The CLI and the menu bar now unlock it with that item
and run the usual JWT checks; the newline `security -w` appends is trimmed
before the KDF.
Keychain consent is user-initiated, as for Claude: the first read happens on
`codeburn import cursor --sync`, `codeburn quota`, or Retry/Reconnect on the
menu bar's Cursor quota. Background syncs and dock refreshes read only after
one such read succeeded, reuse the opened token per blob, and the first
refused read switches that off again. Status never reads the keychain. No
service-only keychain lookup, since after a Deny it asks again.
Windows and Linux say CodeBurn cannot unlock the session there yet; the
desktop app, VS Code and the Windows tray show the new `locked` line in all
locales. Docs describe the prompt and what Always Allow grants.
Closes#1801
Today's day row comes from the live parse with session identities, but
the session-count provenance treated it like a cache row, so a Today-only
period reported its exact count as partial on the headline, the provider
rows and the project rows. Today is now left out of the cache-provenance
days at every call site (money still uses the complete day set) and the
status snapshot render version moves to 13 so stale labels are rejected.
The Overview hero count-up wrote textContent into a div whose children
React owns; when the ~ estimate marker appeared, React hit a detached text
node (insertBefore NotFoundError) and the error boundary took over. GSAP
now writes only into an empty span and React keeps the marker; a currency
change with the same raw value refreshes the text.
Fixes#1760Fixes#1785
The desktop app starts the tray with CODEBURN_BIN set to its own
codeburn-cli.cmd launcher, which the tray took as final. When that launcher
failed (#1750: exit 1, "The system cannot find the path specified"), the
Volta install under a relocated VOLTA_HOME was never tried.
The setup check now probes CODEBURN_BIN, every install the PATH and prefix
search finds, then the recorded launcher, and keeps the first whose
--version answers. When none does, the error lists each path with its own
failure. Export and terminal actions use the probed CLI. The desktop app no
longer overwrites a CODEBURN_BIN the person set themselves.
The desktop app runs codeburn update for the CLI and menu bar on launch and after it updates itself, and the update banner installs and restarts in one click with a Try again path if the install fails.
* fix(pr): push evidence beats the carried pr-link, show every PR fact
Claude Code's pr-link tags a session with its latest PR and CodeBurn carried
that tag to every later turn and subagent run. A run or turn that pushes the
head branch of a PR opened in the same data (paired gh pr create --head) now
goes to that PR, same repo only; spend left only on the tag is carriedUSD.
Rows also get runList (why each run counts), a locally evidenced state and
follow-up round times, and a call opening several PRs pairs its pr-link with
the last create. The PR card states every unknown and the expanded view shows
timeline, runs, cost by phase and work as compact tables.
* fix(app): full text on hover for cropped PR cells, plainer wording for same-session spend
Truncated PR labels, run descriptions, table labels and summary figures now show their full text in the shared tooltip surface while cut off. The carried-from-session-tag wording becomes 'from other work in the same session' in all six locales.
* fix(pr): explain carried spend without the pr-link jargon
* fix(pr): a subagent whose description names a PR folds into that PR
A subagent run folded into whatever PR its parent's last pr-link carried, so
review runs for #1703, #1712, #1728 and #1729 landed on a stale closed#24.
The Agent description ("Review PR 1729 ...") is now read from the subagent's
.meta.json at report time; a single PR it names outranks the carried PR.
* fix(pr): date each PR by its own turns and count its agent runs
Every PR row carried the whole session's first and last timestamp, so one
long orchestrator session stamped the same span and "1 session" on all of
them. Rows now span their first to last attributed turn or folded run, and
a new `runs` field counts contributing sessions plus folded subagent runs.
* feat(pr): local one-shot, follow-up rounds, review runs, merge time and cost split per PR
Each PR row now carries signals read from cached commands, no GitHub calls:
openedAt (gh pr create paired with the PR's first pr-link), follow-up rounds
(pushes naming its --head branch, or from a run whose description names it,
in 10-minute bursts), oneShot only when the open and its branch were seen,
reviewRuns and reviewCostUSD, mergedAt and timeToMergeMs, a build/review/fixes
cost split that sums to the row, and linkEvidence to flag pr-link-only rows.
* feat(session-why): PRs the session opened, their follow-up rounds and review runs
The why view now carries prWork: each gh pr create whose output printed a PR
URL, the pushes after it on its --head branch (or from a helper whose
description names that PR) in 10-minute rounds, and the helpers that read a
PR without committing or pushing, with their cost.
* perf(pr): only parse commands that can name a push or a PR
* feat(app): local PR facts on PR cards and PR work in the session view
Each PR card gets one line of facts from the CLI's local signals: the review
run and its cost, one shot or follow-up rounds, and merge time after open.
Unknowns are left out; a pr-link-only row says so. The count shows agent runs
instead of sessions. The expanded card splits spend into building, review runs
and after first review, and lists the review runs. The session view lists the
PRs it opened and the review runs it spawned.
* perf(pr): read subagent descriptions only for PR attribution
Every report, status and menubar refresh read each subagent's .meta.json
(about 2,000 files in 30 days) and the cross-provider correlation scanned
every folded run's commands for PR signals it never uses. Sessions now carry
the sidecar path; the description is read on demand, memoized by mtime, only
for folded runs and runs that push, and correlation skips the command scan.
The by-pr footer now counts runs to match the Runs column.
A preset subscription plan (SuperGrok, Cursor Pro, Google AI) past its price
showed "$X over" in red and "On pace to exceed", but a flat subscription is
not billed overage. Those cards now read "13.6x your plan's price in API
value" with a neutral bar and projection; custom plans keep the cap wording.
#1657 put the ~ mark on estimated rows, but the headline total stayed bare on
the desktop hero, the mac hero, the Windows hero and tray text, and GNOME.
Each now applies the same 1% rule to the provider split's estimated portion
(the desktop helper moves from the VS Code summary into lib/format). Combined
device totals carry no estimated split and stay unmarked.
Switching period or provider kept the previous model pair in state, so the
report query fired once with the new period and the old models before the
facets effect reset them, and the CLI exited 1 with "model not found". Both
Cohorts and Classic now request a report only for models listed in the
current period's facets.
cli_error gains duration, exit, a locally classified reason and the scoped provider. New provider_read_fail (desktop) and update_result (desktop, menubar, tray) events, enums only.
For a bundled CLI the not-found panel now tells desktop users to reopen or reinstall the app instead of installing from npm. Strings in all six locales.
Lockfile-only updates: @hono/node-server 1.19.17 in the CLI; xmldom, postcss, browserslist, baseline-browser-mapping and shell-quote in the desktop app. No major bumps.