feat(shortcuts): rebindable previous/next session, duplicate-chord refusal, rebinds stay out of the PTY
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
fix(deepseek): find pnpm when installing the terminal profile, and install it with dsh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
fix(skill): keep the §0 bootstrap intact when Claude Code loads the skill with arguments
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
fix(uploads): refuse an unsupported video before uploading it
Conflict resolved against #621 in src/web/public/image-input.js (the
_promptAttachKind doc comment only): #621's comment is kept with #622's
clause that a video MIME counts only for the containers the server
admits; #622's video MIME check and #621's document branch both stay.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
fix(telemetry): gate the prompt cache like the model report, and keep the first 20 miss causes
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The DeepSeek install command asked npm for dsh and pnpm in one go. On a
machine that already has pnpm through corepack (`corepack enable` puts a
pnpm shim in node's own bin directory, which is npm's global bin for nvm
and Homebrew installs), `npm install -g pnpm` refuses to overwrite that
shim with EEXIST, and because both packages were in the same command the
whole install failed: dsh was not installed either. Measured with npm 10
against a scratch prefix holding a corepack-style shim. On master the same
machine installed dsh fine and the profile install used the existing pnpm.
The command is now `npm install -g @deepseek-ai/dsh && (command -v pnpm
>/dev/null || npm install -g pnpm)`, so pnpm is installed only when the
install environment has none, and a failed dsh install still fails the
command. The two hints that embed it follow (the launch-table drift test
requires that), and the catalogue is regenerated; install.sh withholds
the command for a launcher-only CLI, so its block is unchanged.
The test now runs the real stock string through /bin/sh with a fake npm
on a PATH it controls, for the three cases: no pnpm, an existing pnpm,
and a failing dsh install.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The wiki said .codeman-uploads/ is the one path "a local agent, a container
and a remote host all see the same way", two sentences after saying uploads
are refused for remote (SSH) sessions. Only a local agent and a Docker case
(workspace bind-mounted at the same absolute path) share it, which is why the
route refuses remote sessions; say that.
The same paragraph said every image goes in after a re-encode on the device,
but _normalizeImageForUpload passes a GIF through untouched to keep its
frames.
CLAUDE.md and a test comment described the image fallback as behaving
"as before 1.36", a release that never existed (1.35.0 was followed by
1.40.0). It means the image path as it was before this change.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude Code rewrites a skill body that it loads with arguments: every bare
$1, $2 ... becomes the matching word of the request (0-based). Nine lines of
the agent preamble used bare positional parameters, so an agent that pasted
the §0 bootstrap (every session without a pre-seeded preamble: Agent Skill
injection off, the default, or a remote or Docker session) wrote helpers
that queried session "5" with timeout=claude and looked up "5=" instead of
CODEMAN_PASSWORD= in the .env, sending every call unauthenticated.
- $1/$2 -> ${1}/${2} in preamble.sh and the identical §0 heredoc in
SKILL.md; same bash, and Claude Code's pattern (\$\d+(?!\w)) skips it.
- CODEMAN_PREAMBLE 1.33.4 -> 1.41.2, so a cache file written by the broken
bootstrap (which carries the old stamp) is rewritten on the next run.
- A comment at the top of the preamble states the rule.
- New test: Claude Code 2.1.300's substitution applied to SKILL.md with
sample arguments must leave the body, and the pasted heredoc, unchanged.
It fails on the old file with the exact sessions/5 and timeout=claude lines.
- plugins/codeman re-synced; patch changeset.
Verified with real Claude Code 2.1.300 (the delivered heredoc is now
byte-identical to preamble.sh; before, 9 lines differed) and by running the
pasted helpers against a local recording server.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The skill only had a section inside "Driving Codeman From An Agent". It now
has its own page, docs/wiki/Codeman-Skill.md, built around three real runs:
- five Claude Code workers powering on in the tile grid (2026-10-15)
- eight worker tabs with lineage lines back to the lead (2026-10-15)
- a DeepSeek Harness worker and a Claude Code worker from one prompt
(2026-10-10)
plus every install route, prompts that work as written, a run start to
finish, which workers it can drive, what it will and will not do on its
own, the Codex notes and what is in the skill. The page says that the grid
recording put each worker on the grid as it started, since sessions an
agent starts do not join the grid by themselves.
The agent page keeps a short skill section (the heading anchor stays) and
links to the new page; the sidebar, Home, FAQ and Settings Reference link
to it too. The FAQ answer no longer says the skill is enabled per case only.
docs/media-assets.md records the new embeds.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Two live captures of a Claude Code lead driving the codeman skill, added to
the agent skill section of both READMEs:
- Step 2: `use codeman to start 5 claude workers, one per TODO item`, five
Claude Code workers power on as tiles in the grid (1120x630, 17 s, 10 MB).
- Step 3: the same prompt with 8 workers in the normal view, eight tabs
power on with lineage lines back to the lead, then a click into a live
worker (1120x630, 16 s, 8 MB).
Real sessions only, cut from the same raw takes as the promo videos; the
time-lapse stretches are labelled on screen. CRT beats play at 50 fps, the
busy time-lapses at 25 fps, one global palette, no lossy pass. Privacy OCR
on every second frame: clean. docs/media-assets.md lists both.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The dialog's own prompt glyph armed idle confirmation and the quiet dialog passed it, so a fresh case reported idle about 4 s in. An agent that waited for idle and then prompted typed into the dialog, and its Enter confirmed No, exit and killed the pane.
The scan must schedule its own next read rather than wait for PTY output, but a call inside the throttle was simply dropped. The dialog's own paint is often the last output a launching pane produces and lands inside that throttle, so the session sat on the dialog with no keystroke ever sent. A stopped session no longer scans, since the kill's own output kept it answering a dialog in a dead pane.
Claude Code's and Codex's selection dialogs ignore a bracketed paste, so
after the previous commit a pasted `3` + Enter confirmed the highlighted
option instead of option 3. sendInput() now types a single character with
`send-keys -l` as master did (submitsAsKeys()). One character cannot form
the burst Codex reads as a paste, so the Codex fix is unaffected.
The trailing-`;` escape moves into escapeTmuxTrailingSemicolon() and now
also covers every `send-keys -l` call in sendInput() and
sendInputToPane(); tmux 3.5a types `x;` for `send-keys -l 'x\;'`.
Docs: the @fileoverview lines and orchestrator-loop-architecture.md still
described `send-keys -l`; CLAUDE.md, architecture-invariants.md and
cron-guide.md name the single-character exception.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
sendInput() typed the text of a text+Enter pair with `send-keys -l` and
sent Enter 50 ms later. Codex detects an unmarked paste from a fast
character burst and folds an Enter inside that window into the paste as a
newline, so the prompt stays in the composer unsubmitted. Measured with
codex-cli 0.162.1 in a throwaway tmux pane: 1505 chars stuck 3/3 as
`[Pasted Content 1506 chars]`, a 98-char line stuck 1/4. SubmitVerifier
re-presses Enter for the short case, but it cannot match the long one: the
composer shows the placeholder instead of the prompt's head.
The text now goes in via set-buffer + `paste-buffer -p -d` (markers only
when the pane requested bracketed paste), followed by the separate Enter as
before. Same pane, same texts: 4/4 and 3/3 submitted. Each send gets its
own named buffer; a paste that fails (pane or server gone) deletes it,
since tmux never trims named buffers on its own.
Also escapes a trailing `;` in the text, which tmux treats as a command
separator (`send-keys -l 'x;'` typed `x`, a lone `;` typed nothing).
sendInputToPane() has no callers in src/ and is left as it was.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
last_miss_cause.causes was z.array(z.unknown()).max(20), so a statusline
payload carrying more than 20 causes failed validation and lost the
whole post: plan usage, model, cache and the footer with it. Truncate
to the first 20 with a preprocess that slices before the array
validates, so an over-long array costs one bounded slice rather than
per-element work across every entry.
The status-telemetry route gated the model report on
capabilities.statusLineTelemetry but stored the prompt-cache report for
any session mode, even though both ride the same statusline payload the
exporter POSTs. Gate the prompt-cache store on the same capability so a
CLI with no statusline exporter no longer carries prompt-cache state it
never actually reported.
_promptAttachKind returned 'video' for any video/* MIME, so an Android
3gp or an mkv streamed up to 1GB before the ?kind=video route refused it
with a 400. Mirror the server's own video-container set so those are
filtered client-side and get the existing unsupported-file toast at once.
Follows the maintainer's follow-up suggestion on #606.
The Attach button, the Compose and paste-dialog pickers, terminal drop and the
clipboard now take documents beside images and videos: the office formats
pdf/docx/pptx/xlsx/zip and the text family txt/md/csv/tsv/json/log/yaml/yml/xml.
A document is admitted by its name's extension (the MIME a phone attaches to a
.md is anyone's guess), buffered under the same 50MB image cap, and judged on
its bytes by the same fail-closed sniffer: a magic-number signature for the
binary formats, UTF-8 with no NUL byte for text. It is saved under a sanitized
stem of its name, so Q3 report.pdf is findable as paste-<ts>-<hex>-Q3-report.pdf.
sh/py/js/html are refused by name with a clear 400. That is a scope choice, not
a security property: file-raw already serves non-media text inert, so this keeps
the server from storing what it would serve or an agent could run, and html and
svg are the ones deliberately kept out. One client classifier (_promptAttachKind)
now returns image, video or document, and the server's dispatch mirrors it: an
image MIME wins whatever the name says, a document extension takes the document
branch, and a foreign extension with no image MIME is refused only after the
image sniff fails, so nothing the image path accepted by its bytes is lost.
Only GIFs that a public page embeds stay in docs/images/: the five README
GIFs plus subagent-demo-20260724.gif, which the published "mission control"
article hotlinks from master. Seven GIFs nothing embeds leave the tree
(their masters are kept in the maintainer's local media library).
.gitignore now blocks *.mp4/*.webm/*.mov/*.m4v/*.mkv/*.avi/*.gif outside
docs/images/*.gif (and the promo branch's Remotion inputs), so a video can
only be committed with an explicit `git add -f`.
docs/media-assets.md records the rule: what is tracked and where it is
embedded, website media (pr/website/assets, deployed separately), release
clips on the design-assets branch, and the local library layout.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The installer's three up-front questions, the :443 fallback, the
optional tailnet rename, command-line flags and install.sh status; the
installer's closing QR note in the Tailscale tip; Diagnostics in
Settings, the bottom-bar git status and creating a case in a custom
folder; docker load in the Docker export bullet; and the installer
note on the loopback default. The Chinese README now carries the same
content as the English one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Mirrors the English README update: tile grid ranking, saved layouts,
tile animations, wheel scrolling in tiles, Detach Tiles and gesture
moves; rail search, tab groups and Focus; HTML pages, the folder grid
and clickable ~/ and relative paths; .codeman-uploads and the phone
Attach button; OSC 52, the prompt-cache readout, MCP server sync with
Copilot CLI, Settings Apply and notification display times; remote SSH
scrollback; the sandboxed HTML page view in Security; Ctrl+W and the
tile shortcuts in the shortcut table.
Adds what the Chinese README never had: the `codeman agent` and
`codeman mcp` sections, the agent skill for every CLI (install table,
--mode, the Codeman Agent Skill setting, the skill GIF), and the
refreshed dashboard tour and tab-states images.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Tile grid (session ranking, saved layouts, tile animations, wheel
scrolling in tiles, Detach Tiles, gesture moves), rail search, tab
groups and Focus, HTML pages and the folder grid, clickable ~/ and
relative paths, .codeman-uploads and the phone Attach button with
videos, OSC 52, the prompt-cache readout, MCP server sync with Copilot
CLI, Settings Apply, notification display times and remote SSH
scrollback.
Also corrects three stale claims: HTML now renders in the sandboxed
page view, Ctrl+W has not closed a session since 1.40.0, and the tile
shortcuts were missing from the table.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The root CLAUDE.md had grown to 202 KB, all of it loaded into every
session. This lands the split drafted on 2026-10-10 (it sat uncommitted
in the shared checkout through two releases), redone on top of today's
master so nothing that landed since is lost:
- src/web/CLAUDE.md: the frontend invariants (every src/web/public/ UI
subsystem, the module notes, the XLSX preview limits, the z-index
stack). Claude Code loads it when a file under src/web/ is read; the
root file tells sessions to Read it before editing src/web/public/.
Its link targets are relative to the file so they resolve on GitHub.
- docs/releasing.md: the COM procedure (now nine steps, with the MCP
registry publish), behind a pointer in the root COM section. The
Version line stays in the root file.
- Root CLAUDE.md (130 KB): everything else, plus the split's trims: the
Core Files table, the reference-only rows of the Additional Commands
table, Data Flow and the respawn presets, whose rules already live in
the root's other paragraphs (the "Read before touching a subsystem"
line keeps the table's doc pointers).
Every other paragraph of master's CLAUDE.md lands in exactly one of the
three files, checked line by line; the paragraphs master changed since
the draft (tile grid, OSC 52, prompt uploads, file-path links and the
HTML view, markdown preview, z-index, MCP server, agent CLI and others)
are in their current form. Both new files join CLAUDE.md in
.prettierignore. Four code comments and docs/architecture-invariants.md
now point at src/web/CLAUDE.md for the moved sections.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
`codeman mcp` is listed in the official MCP registry as
io.github.Ark0N/codeman (first published for 1.41.1). The registry does
not follow npm, so without a step in the COM procedure the listing stays
on the version it was last published with. The step also records the
two traps hit on the first publish: the Snap Store mcp-publisher (1.1.0)
rejects the current server.json schema with a false "deprecated schema
detected", and older publishers save live login tokens into the current
directory as .mcpregistry_* files.
The announcement step becomes step 9.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- isUserBoundShortcutEvent skips toggleSessionSidebar (it has its own gate) and an
entry re-captured onto its own default, so ^L still reaches the CLI.
- Re-enabling a shortcut and Reset run the duplicate check and refuse when the
chord was taken meanwhile; Ctrl/Shift+Enter and Ctrl/Cmd+V are reserved with
Alt+1-9, and the Alt-digit check no longer excludes Cmd.
- Alt+[ / Alt+] keep their Alt/Option+[ labels via displayBindings.
- Behavioural tests (vm harness) replace the source-string test.
- Docs: wiki Rebinding / Why-chords-behave-oddly, architecture-invariants note.
- Small: _shortcutActions indentation, stale Alt-block comment, typeof check
instead of Object.hasOwn, full-width parentheses in zh-CN.
The 800px GIF rendered the header stats strip (CPU, MEM, 5H/7D) at ~5px
text height. The 1600px source shown at width=800 stays sharp on high-DPI
screens. New filename so GitHub's image cache serves the new file.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
One consolidated patch changeset for the 24 PRs in this release (the PRs'
own changesets folded in), CHANGELOG, package and plugin manifest versions,
and the CLAUDE.md version line.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
One stored grid per browser shared by every Codeman window, no cross-profile
notice (BroadcastChannel is per profile), and the user-activation window that
blocks a pop-out from a drag-out held too long.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
_dockForeignTile posted close-request only when this window knew the session
was popped out. A pop-out opened from another dashboard window was not in this
window's detachedSessions, so it stayed open beside the newly docked tile and
two views sized one PTY. The request now goes out for every docked id;
dashboards ignore it and only that session's pop-out reacts.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
_onTileAdoptedElsewhere acted only on a tear-off, a pending tear-off or a drag
that just ended. A header drag carries the tile id even when this window has
Detach Tiles off (no tear-off), so another window could dock the tile and its
claim could arrive before this drag ended; the tile then stayed here as well.
The header drag (_draggedTileId) now counts as dragging too.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
There is one stored grid per browser (codeman:tile-grid), and the window that
docked the tile has just written it. When this window then hears the claim and
drops the tile, removeTile (or, for its last tile, closeTileGrid with
keepStored: false) overwrote or forgot that stored grid. The removal now runs
under _tilePersistHold, and the last-tile close keeps the stored grid.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
#604 makes `~/` links clickable and passes them through unchanged, since only
the server knows the session host's home. The html-view mint read `~/x.html`
as workspace-relative and answered 404, so the preview fell back to the
source view. It now expands `~/` against this host's home, as the attachment
route does, and refuses it in a Docker case (that `~` is the container's)
with the same message as the attachment and folder routes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Pins the merge resolution: `~/`, relative (`./`, bare) and hard-wrapped
links to .md/.markdown and .html/.htm open the file preview (rendered
markdown, HTML page) exactly like rooted ones, while other text such as a
relative .json keeps the log viewer. Fails if the terminal's activate path
drops rendersAsPage().
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Conflicts resolved against #604, #606 and #601:
- constants.js: the rooted link pattern takes html-md's `markdown` and
`html?` with #606's m4v/ogv; #604's relative pattern gets the same list
(link-provider-regex pins the two equal); #604's findFilePathLinks /
folder helpers and html-md's rendersAsPage() are both kept, and
window.CodemanFilePaths exports all of them.
- terminal-ui.js: #604's resolved `target` (relative, ~/ and folder links)
now also goes through rendersAsPage(), so every link kind routes .html to
the page preview; .md routes there through FILE_PREVIEW_EXTENSIONS.
- CLAUDE.md, api-reference, architecture-invariants: both sides' text kept.
- test/folder-grid-ui.test.ts (#604) stubs rendersAsPage() like the other
panels-ui harnesses.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
format:check failed on src/web/server.ts and src/web/public/approvals-ui.js;
both assignments fit on one line at printWidth 120.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Three comments said `altPath` is never written, but run() makes the alt file
the write target when the primary is absent. The behaviour is fine (a JSONC
file with comments fails the parse guard and is never written); the
comments now say so, so nobody loosens the path rule on that belief.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
mcpTable took any record at `mcp.servers` as OpenCode's V2 container.
OpenCode itself does so only when the value is not a V1 server (`type`
local or remote). So a server named "servers" copied from another CLI was
written as V1, re-parsed as a container, failed the read-back guard and
failed the whole OpenCode target, and an existing V1 "servers" stopped
being read. The V2 entry now also wins a name collision, as in OpenCode;
sync never writes a name already defined, so the read-back is unaffected.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
_sendBracketedPaste wrapped the text as is, so clipboard text carrying the
end marker (a pastejacking page can write one) closed the bracket early and
the rest was typed as keys, its CR running a command before review. The
paste markers inside the text are now dropped first; this also hardens the
Compose path, which shares the helper.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Codeman maintainer