Repository navigation
Releases
Fuigo 1.0.45
Fuigo 1.0.45 release notes
Folder trust and consent
- A nested repository no longer borrows its parent's trust. A folder whose own git metadata points at a trusted
ancestor (throughcore.worktree) is asked about like any other folder instead of being treated as trusted with no
question. The same holds for the folder a session works in when acore.worktreeancestor has no.git. - Task children are judged in the folder they were launched from. A subagent started with a
cwdis checked
against its parent's launch folder as well, so it cannot move a session from an untrusted folder under a trusted
folder's shell rules. A relativecwdis resolved against the parent'scwd, not the process's. - No git status scans in untrusted folders. The in-process change tracker waits for the folder-trust decision, so a
repository-configured content filter cannot run before you have answered. - No language servers in untrusted folders. User- and plugin-configured language servers start only after the
folder is trusted.
MCP servers and plugins
- Untrusted folders cannot start MCP servers under a different name. A project server whose name needed
sanitising, or that was forwarded under a renamed copy, is now dropped in an untrusted folder like any other project
server. - Project plugin settings in an untrusted folder cannot touch your plugins. A project's
[plugins].disabledno
longer switches off your own plugins, and a project plugin with the same name no longer displaces yours. - Marketplace plugins do not refresh themselves at startup. The spawn-time local refresh skips installs that came
from the marketplace cache, so a plugin changes only through an update command. - Bounded
tools/listpagination. A server that returns endless pages, endless tools or a repeating cursor stops
at a fixed cap instead of looping.
Sandbox
- Custom sandbox profiles work in the workspace server. The workspace server resolves your custom profiles the
same way the main binary does instead of replacing them withworkspaceand possibly continuing unsandboxed. - A bad
[sandbox]field is reported.sandbox.profileis read on its own, and a type error in another
[sandbox]field prints a message on stderr instead of silently turning the sandbox off. - Project profiles in untrusted folders are ignored, and built-in names ignore letter case. A project
sandbox.tomlin an untrusted folder cannot select the profile, andMinimalorSTRICTare treated as the
built-in names. - The workspace
.mcp.jsonand imported MCP config files are protected from sandboxed writes underenforce. - Missing protected entries are denied in every letter case on macOS, not only the spelling you wrote.
Editing and files
- Edit tools never follow a look-alike filename. A file name that only resembles an existing one (for example with
look-alike Unicode characters) is no longer silently redirected onto a symlink or another file. - Non-UTF-8 files are no longer corrupted.
apply_patchand the other edit tools refuse to
rewrite a non-UTF-8 file. Deleting such a file still works, reported without a snapshot. apply_patchis safer with moves and adds. A Move onto the same file updates it in place instead of deleting
it, and Add and Move report the destination's prior state.- Rewind restores the right content.
writeandsearch_replacereport the true prior state of the file, so a
rewind brings back what was there. - PPTX reads are bounded. Extraction has a total decompressed-size budget and a slide cap.
web_fetch
- Hostile pages no longer hang or crash a fetch. Converting a page to text is bounded in time and memory: deeply
nested markup, huge attribute counts, quadratic inline merges, long runs of newlines and nested tables are handled
within fixed limits, and a page that would cost too much falls back to plain text. - Tables render without padding, with correct headers. Header rows come from the table head, cells are
normalised, and captions are not pipe-escaped.
Sign-in and sharing
- A huge
expires_inorintervalno longer aborts. A token response with an oversized value no longer stops
sign-in or refresh. - Logout reports a real error. When the stored login cannot be read,
fuigo logoutsays so and keeps the file
instead of reporting that there was nothing to log out of. - Saved API key helpers follow
FUIGO_AUTH_PATH. - Shared sessions are scrubbed before upload. The session body sent to the share backend gets the same secret
scrub as the copy kept in storage.
Terminal, links and media
- File links reveal launcher-type and executable files instead of opening them, so a modifier-click on a path in
repository content cannot run it. - Network-share paths are not probed. Paths found in model or tool text are never stat'ed or read when they name a
network share. - Image reads are capped. Inline images over the size cap are skipped and remembered as too large, and
ffmpeg,
ffprobeandsipscalls have time limits. fuigo wrap: one host-image read at a time, at most one per interval, with a bounded write queue.- Image paste through
fuigo wrapover SSH: a new localfuigo wrapnow authenticates image-paste requests with a
per-session nonce, so pasting an image into an olderfuigoon the remote side no longer works until that remote is
updated. Commands Fuigo runs no longer inherit the nonce. - Permission cards opened while you type hold the next Enter and digit, so a keystroke meant for the composer does
not answer the card.
Release, update and serving
agent servewarns on a non-loopback address. Binding beyond loopback over plainws://prints a loud warning.- Updater temp file is private. The temporary
.npmrcis created exclusively, mode 0600, in an owner-only
directory. - Install scan skips empty and relative
PATHentries, sofuigo --versionis not run from the current folder. - Release workflow is stricter. A real publish is refused off a tag or without macOS signing, and the in-tree
leftover upstream install scripts, which downloaded from hosts Fuigo does not control, are removed with their README
line.
Campaigns
-
Campaign patches cannot move trusted API origins. The session bearer and
FUIGO_API_KEYstay with the origins
you configured. -
Campaign patches cannot set where a model connects or what it sends. Hosts, headers, query parameters, providers
and keys of every model stay the ones you configured. -
Campaign-added MCP servers are untrusted and carry no secrets: no saved-key references, and no environment,
header or OAuth secret fields. -
Stored MCP tokens survive a damaged or unreadable
mcp_credentials.json. A file that cannot be read is left
alone and the save fails; a damaged one is copied aside first. Every write, including removing a server, is locked
on every platform. -
Campaign patches cannot set execution policy or where data and code come from.
[sandbox], web fetch local
access and proxy, the shell environment policy and command prefix, the hub, telemetry destinations and exporters,
web search routing, session trace upload and the telemetry switch it follows, the agent definition and agent
name, which plugins are enabled, whether language servers start, which MCP servers you disabled, subagent role
and persona files, and the sources of updates, plugins, skills and rules stay as you configured them. Campaign
values are never expanded from environment variables. Administrators set these in their own managed or
requirements layer, not inside a campaign. -
Campaigns can now set only what campaign features use. A campaign can set the default model and the helper
models (a nudge your own/modelpick always overrides), a model's display name and context window, add new
MCP servers (without secrets), and turn managed MCP servers off. Everything else in a campaign is ignored, and
Fuigo shows a note naming the settings it ignored. -
Campaign model nudges cannot reach your own endpoints or keys. A campaign may name a catalog model or one with
no route of its own, not only a built-in one. It cannot point the default, image-description or
session-summary model at a model whose endpoint, provider or credential your own configuration sets; such a setting
is ignored with a note. A context window outside 1,024 to 10,000,000 is ignored, and a change to a helper model is
announced once. -
A program-running MCP server added by a campaign waits for you. It is listed as off until you turn it on in
/mcpor withfuigo mcp enable <name>, and if the campaign later changes its program, arguments, folder or the setup choice that picks the program, it needs
your approval again. This wait covers program-running (stdio) servers only: a campaign HTTP server does not wait
for approval, and your own servers are unchanged. -
Campaign notices are cleaned and capped, name only the sub-settings that were ignored, and show the program a
campaign MCP server would run.
More fixes
Shell and tools
- When a model sends several JSON objects in one tool call, Fuigo runs the one that fits the tool.
- Output a command prints after it exits or is stopped no longer grows memory or the log file without limit; the
short drain after a background task exits keeps at most 8 MiB. - Two shell calls that share one call id no longer share a log file or overwrite each other's background task.
- In the persistent shell, programs a command starts (bash 4+ and zsh) can no longer write to the channel Fuigo uses
to save shell state. Two limits remain by design: builtins running in the same shell can still write to it, and
bash 3.2 (the stock macOS/bin/bash) hands the saved copy of the channel to the programs it s...
Fuigo 1.0.44
Fuigo 1.0.44 release notes
Sandbox: a read-only launch folder
- New built-in
minimalprofile.fuigo --sandbox minimalruns with the strict profile's reads and network
policy, keeps session state, temp files and devices writable, and makes the folder Fuigo was launched in read-only.
Use it for a checkout you do not trust. - New
workspace_writesetting. Any custom profile can setworkspace_write = falseto make the launch folder
read-only. Leaving it out keeps the base profile's behaviour, so existing profiles are unchanged. The read-only
launch folder wins over any write grant inside it, including relative paths, which are never created there. - Enforced, or Fuigo does not start. On Linux a read-only launch folder needs bwrap: if bwrap cannot run, Fuigo
and the workspace server refuse to start with a clear message instead of falling back to a weaker sandbox. A Linux
build without sandbox enforcement refusesminimalthe same way. macOS uses Seatbelt. Windows warns and continues. - No new aliases. The sandboxed process cannot create a hard link to a launch-folder file in a writable
directory, on the same filesystem or another one, or move a launch-folder file out. - Profile names you did not write cannot weaken it. A project's own
.fuigo/sandbox.tomlcannot redefine a
built-in name (minimal,workspace,devbox,read-only,strict,off): the built-in applies and a warning
names the file. Your own~/.fuigo/sandbox.tomlcan still define[profiles.minimal], which keeps working as before
and prints a warning that it shadows the built-in. - Warnings instead of silent surprises. Unknown keys inside
[profiles.*]and session storage that would sit inside
a read-only launch folder each print one warning; Fuigo keeps starting. - Deny entries written with
./now work on macOS. Adeny = ["./.env"]entry (or any path containing/./)
was not enforced on macOS, so the file stayed readable. It is now denied like.env. Linux was not affected. ~in a deny entry means your home everywhere.deny = ["~/.aws"]already protected the home path for
sandboxed commands; Fuigo's own process now applies it to the home path too, instead of a~folder inside the
workspace.- Known limit. Like any path-based sandbox, it cannot see links that already exist before launch: a hard link or a
writable bind mount of a launch-folder file created outside the sandbox keeps that path's own access.
Folder trust and consent
- No answer before the question is on screen. A follow-up click after accepting the consent notice or trusting a
folder no longer acts on the next screen before it is drawn, and keys typed before the trust question appears are
ignored. "On screen" means the terminal has finished writing the frame that shows the question. Ctrl+C and Ctrl+D
always work. - Short terminals. The trust question accepts answers whenever its line is visible, even when the menu has no room.
- Tiny terminals. Welcome, consent and trust screens and MCP elicitation cards draw without crashing or overflowing
at any size down to 1x1.
Sign-in
- Sign-in finishes against the login stored now. If another window logs out, signs in again or switches account
while a browser or device-code sign-in is open, completing it writes nothing and says the login changed in another
window. Token refreshes and account details loading during the sign-in do not count as a change. - Logout is reliable. A logout that cannot remove the stored login reports an error and keeps the session instead
of reporting success. Logout no longer freezes the screen while it waits for another window. - An unreadable
auth.jsondoes not block sign-in. If the file cannot be read (for example a permission error), the
new login works for this session only and is not saved, with a warning; that session leavesauth.jsonalone
until Fuigo restarts, so an older stored login cannot take its place. A corrupt file is backed up and replaced, as
before. If the auth file stays locked by another window, sign-in stops instead of keeping an unchecked login. - Token refreshes never overwrite another window's login. A refresh that finishes after another window logged out,
signed in again or switched account keeps its token in memory and leaves the stored login alone. This also covers
the workspace server's refresh. - Sign-in screen starts fresh after logout or a cancelled login.
- One retry in rare cases. With a 1.0.43 window open, or when a failed token refresh removes the stored login while
a sign-in is open, that sign-in can report a change once; signing in again works.
Startup, paste and headless
- Faster startup when the Linux system bus is slow. The power listener connects in the background; startup no longer
waits for it. If it cannot connect within its deadline, power events are reported unavailable with one warning. - Faster startup when the terminal never answers the keyboard query. The probe has one 200 ms deadline, then falls
back to classic key handling. - Large pastes without bracketed paste no longer stall. A 200 KB paste reaches the prompt in under a second instead
of more than ten. A burst of 256 or more single-line key events arriving within 1 ms of each other is treated as one
paste; a trailing Enter stays a key. fuigo -p --resumeneeds an id or title. An empty or missing value is refused up front;--continueresumes the
most recent session.
Display and input
- Narrow and wide terminals: the usage modal, dashboard (compact rows at 40-43 columns), dashboard peek, plan review
buttons, pickers, list panes, numbered lists past 10, terminal titles and CLI tables fit by display width, including
CJK and emoji, and use one ellipsis style. - Failed web fetches show the real error when expanded; grep results expand tabs; open folds point down; empty pickers
say what is empty. - Copying a table keeps inline-code backticks; Shift+drag beside the scrollbar copies no scrollbar glyph (also with
NO_COLOR); drag selection scrolls straight away at the pane edge; Up on the first line of a question's text answer
returns to the options unless a dropdown is open. - Legacy Windows consoles get plain fallbacks for the pending todo glyph and scrollback glyphs.
- The composer draft is kept when a queued row being edited disappears under an open question; labels say "1 hook",
"1 page" and so on.
Fuigo 1.0.43
Fuigo 1.0.43 release notes
Rate limits and provider errors
- Provider errors keep their code. In agent mode, a failed request's JSON-RPC error and every retry notice now carry
a typeddata.upstreamobject. It holds the HTTP status, the provider's errortypeandstate, the Fluxcode
andcodes,hold_reason,limit,retry_after_sandresets_at. Before, only the message text survived, cut to
280 characters. Existing fields keep their meaning. The agent-mode guide documentsdata.upstreamas a stable wire
format. - Only known values are reported.
type,stateandcodecarry values from a fixed vocabulary: Flux codes,
supported provider error types and LiteLLM exception names. Anything else is reported as"other", so a value a
provider echoes back, such as a key, never reaches the wire or a saved session. - Errors rewrapped by a gateway keep their code. A provider error quoted inside another error's message, for
example by LiteLLM, is read for its code and reset time, including on a rejected sign-in and on errors that arrive
mid-stream. - A Flux refusal that retrying cannot fix ends the prompt. A revoked or unknown key, a used-up allowance, budget or
plan cap, or a Flux rate limit with no known wait now stops the turn with its reason instead of retrying in a loop. A
rate limit with a known wait still backs off. Other providers keep their existing rate-limit retries, including
Retry-After. - One sign-in refresh, then stop. A bare 401 on a session that can refresh its token refreshes once and resends
once. A second refusal ends the prompt. A new prompt starts fresh. - A failed session-summary call no longer ends the main prompt.
Fuigo 1.0.42
Fuigo 1.0.42 release notes
Editing the prompt
- Vim editing in the prompt.
/vim-mode all(or[keys] vim = "all") gives the prompt NORMAL and INSERT modes,
with motions, operators, counts, undo and redo. Enter still submits from either mode. The shortcuts sheet lists the
NORMAL-mode keys while it is on. In NORMAL modeCtrl+Ris redo, so open the sessions list from INSERT mode. The
/settingsswitch "Disable vim input mode" ([ui] simple_mode) now turns this same prompt vim on and off. - More ways to mention a file with
@.@~/reaches files in your home folder,@!reaches hidden files, and a
:Nsuffix points at one line. The picker also lists files outside the workspace
(ui.at_picker_outside_workspace, on by default). Typing at the end of a file chip edits it in place. - Quit with one
Ctrl+C. Withui.single_ctrl_c_quiton, one press quits when nothing would be lost: no turn
running, no draft, no queued prompt and no open question. Off by default. /doctorwarns about a slow tmux escape time. Inside tmux, anescape-timeabove 50 ms is reported with the
command that fixes it.
Transcript
- Dragging selects text across blocks. A drag that starts on text now runs on into the next blocks, and copy gives
plain text without colours, borders or buttons. A drag that starts outside text still selects whole blocks. - Copy buttons work while a menu is open, and while a queued prompt is being edited.
- Text stays readable in an 8-colour tmux session. Bright colours fold onto the eight normal ones there.
- Markdown links that contain
$or LaTeX characters render as links instead of switching into math. A$inside
a link's address is shown as%24, which is the same address. - Mermaid diagrams break lines on
\nand<br>instead of printing them. - Warp opens the whole of a wrapped URL.
Keys and input
Ctrl+Ein minimal mode moves to the end of the line while you are editing a draft.Ctrl+Yon a question card no longer closes it unanswered.- Esc on an unanswered
/btwcancels it, and/btwstarts a side question only at the start of a message. - A slash menu closed with Esc stays closed when the command list refreshes.
- Typing while the transcript has focus goes to the prompt, including non-ASCII text, and leaves an open card alone.
- Windows: Shift-held symbols and AZERTY digits work in shortcuts.
- Double-clicking a view's close button no longer presses what was underneath it, such as a stop button.
- Long pastes delivered as keystrokes no longer freeze while the plan-mode tip is shown.
Background work and the dashboard
- Dock stop buttons work while that run's approval card is open.
- Dashboard rows stay Working while background subagents and workflows run, and stopping a row ends them.
- The footer's context use is right after a rewind.
Sessions
- Delete, rename and copy refusals show their full text, and session load errors give the whole reason.
- A deleted or renamed session no longer appears in resume lists or title searches.
- A session with an unknown reasoning effort resumes with the default effort.
--resume <id>reads only that session, so it starts faster with many sessions.- A resumed session shows a background command's real run time in the turn that ran it.
fuigo exportis faster and uses less memory on sessions with large shell output.
Hooks
- Opt-in fail-closed hooks. A
PreToolUseorUserPromptSubmithook entry can set"failClosed": true. If that
hook times out, crashes, fails to start, exits non-zero without a decision, or prints output that is cut off,
unreadable or carries a key Fuigo does not recognise, the tool call is denied or the prompt is blocked. Hooks without
the flag still fail open, as before. A prompt block applies only to a prompt you typed, as for any block. - A
!shell line now honours a hook's stop ("continue": false), as tool calls already did.
Updates and startup
- Automatic update stops retrying a version that fails to start. Only a real launch failure counts: a slow first
run or a busy file does not. fuigo updateruns when a sandbox profile is set, unless Fuigo is already confined. A confined session never
auto-updates and tells you to runfuigo updateinstead.--restore-codeis safer. A failure leaves your uncommitted and untracked files in the tree, it restores its own
stash when another worktree changed the stash list, and it keeps changes stashed if a post-checkout hook fails.- Rejecting a change keeps edits you saved, even when the save was missed.
- Non-UTF-8 environment variables no longer stop startup.
- macOS: the full-screen app no longer freezes in a sandbox that blocks the window server.
- Typing does not lag when ffmpeg is missing and
PATHis slow: the check runs off the drawing path, once. - Exiting is reliable.
Ctrl+Uupdate-and-quit exits within its deadline, and on Windows Fuigo exits fully after a
startup error,/minimal,/fullscreenor a signal. - Opening links and files no longer leaves a zombie process per click.
- Short terminals: the cancel-turn prompt and rewind confirmation no longer crash the full-screen app.
- Kitty graphics: hidden images are freed in the terminal.
Other fixes
- MCP stdio servers no longer lose part of a long line when a read is interrupted.
!mode Tab completion offers symlinked commands.- Custom agent files with a UTF-8 byte order mark load.
- Sign-in error messages no longer repeat the underlying cause.
- The always-approve help text says the setting is saved to config. The macOS docs recommend
Ctrl+Vto paste
screenshots. Auto-review mode tints the whole prompt bar, like plan and ask. Editing a queued prompt labels and dims
its row.
From 1.0.41
- The strict sandbox starts for a non-root user when a container socket sits in a root-only folder (for example
/run/podman). Such a socket is skipped only when the folder that blocks it belongs to another user. This shipped in
1.0.41 but was missing from its notes.
Fuigo 1.0.41
Fuigo 1.0.41 release notes
Folder trust
- Trusting a folder mid-session now reloads the session. After you trust a folder, the session loads that folder's
hooks, plugins and MCP servers. MCP servers whose settings did not change keep running. /trustopens the trust screens in the full-screen app. This folder shows the session folder, its trust key
and whether it is trusted:ttrusts it andustops trusting it. Trusted folders (Tab) lists every folder
you trusted, andustops trusting the selected one. Each change asks you to confirm.- A resumed or forked session asks about its own folder. The trust question now names the folder the session was
recorded in, not the folder you launched from.
Sessions
--no-session-persistencekeeps a headless run off disk. Nothing is saved under~/.fuigo/sessions: no session
folder, prompt history or search index entry. The JSON output still names a session ID, but it cannot be resumed.
Combined with--resumeor--continueit exits 2. An interactive session warns and is saved as usual.- One process at a time writes a session. Deleting, renaming or copying a session that is open in another window
is refused with a clear message instead of racing the other process. /deletestarts on Cancel. Pressing Enter once no longer deletes a session.- A headless
--resumethat fails says why. With JSON output, it prints one JSON error that gives the reason.
Hooks
SessionEndcommand hooks run when a session closes. The end reason is available to the hook's matcher.- Project hooks can be disabled per repository. A new disable applies only to the repository you set it in. A
disable saved by an earlier version keeps applying everywhere, as before. Enabling the hook in a repository removes
both.
Permissions
- A permission file that does not load is reported at session start. If a
.claudesettings file (yours, or a
trusted project's) exists but cannot be read or is not valid JSON, Fuigo names the file and says its rules did not
load. Rules that do not parse are listed by file. The session still opens, and every rule that did load applies.
Startup
- A non-UTF-8 argument no longer crashes startup. This includes a non-UTF-8
--cwdpath and the Linux sandbox
re-launch.
Fuigo 1.0.40
Fuigo 1.0.40 release notes
Folder trust
- An ACP client can ask for a notice after trust is granted mid-session. Set
"reopenNotice": truein the
fuigo/folderTrustclient capability. The client then gets one liveconfig_notice: "Folder trust is on. Reopen
this session to use it." It goes only to the connection that was asked and is not replayed on load. Clients that do
not opt in see no change.
Pasted images
- Images pasted with Send Now come back with their placeholder. This holds after cancel (Ctrl+C), after a failed
send, and after/rewind. After/rewindon a resumed session, the images come from the image stored in the
session. - A missing or invalid stored image is now reported. Before, it was dropped without a word. Now Fuigo says the
image is no longer available. - The transcript shows which images were sent. An
[Image #N]placeholder whose image was not actually sent shows
dim and struck through. This works live and on resume.
Releases
- The Linux release build now frees runner disk first.
Fuigo 1.0.39
Fuigo 1.0.39 release notes
Subagents
- Forked subagents no longer get the project instructions twice. A child forked from a session that already holds
the folder'sAGENTS.mdand other instruction files now recognises them and does not add a second copy. - Subagents get the plugins of the folder they work in. A subagent in another folder (a worktree, or a session
opened elsewhere) now gets that folder's plugins, not those of the folder Fuigo was launched in, and a plugin edited
since the last/plugins reloadshows up. The check of which agent types exist, and the skills a subagent inherits,
now use the parent session's own plugins, so they match the agent types the parent offers. - A subagent worktree that cannot be created now says why. When a tracked file fails to copy into a new subagent
worktree, or the disk is full, the subagent stops with that error instead of quietly running in the shared folder.
Other worktree errors behave as before.
Scheduled tasks
- Scheduled tasks keep working after a model switch made before the first turn. Before, that switch left the session
talking to a stopped scheduler, so parking the session and the scheduled tasks of new subagents did not reach it. - A scheduled task fires once. Two schedulers sharing one session's state, for example around a model switch, could
both fire the same task. Each fire is now admitted once. A loop task whose subagent is still running is not started
again, and that subagent keeps running across a model switch; the loop's next iteration after it starts a fresh
chain. - Shutdown behaves as in 1.0.38.
Settings and commands
- New
[keys] vimsetting.vim = "nav"turns on vim-style navigation andvim = "off"turns it off. If
[ui].vim_modeis also set, it wins, so existing configs behave exactly as before./vim-modestill changes the
current session only. fuigo worktree listandshowmark a worktree whose folder is missing.
Releases
- The macOS release build may now run for up to 6 hours, GitHub's limit for hosted runners.
Fuigo 1.0.38
Fuigo 1.0.38 release notes
Folder trust
- Trusting a folder during a session now brings in its project instructions. Before, a folder trusted after a
session started left that session without the folder'sAGENTS.mdand other project instruction files, even after
a later load, while a new session had them. Now the files arrive at the session's next turn and stay there through
later loads, resumes, compaction, rewinds and forks, with any agent. Each file reaches the model once. - A trust answer that arrives just after the first prompt still counts. When a client's trust request is
outstanding, the first prompt of a session waits up to 2 seconds for the answer, so the instructions land in that
first request. Sessions in a folder that is already trusted never wait. - A client can trust a folder for this run only. A client that shows the folder trust request can reply
{"outcome":"trust","scope":"configAndCommands","duration":"process"}. The trust lasts until Fuigo exits and nothing
is written to disk. It grants no more than a normal answer, and a saved or earlier "no" still wins. Any other
durationvalue is a rejection. Older Fuigo versions reject this reply, so they stay untrusted.
Two limits remain:
- Sessions saved by an earlier version are not repaired. If one is missing its project instructions, start a new
session. - A session that gets trust part way through receives the top-level project instruction files. Nested
AGENTS.md
files and project skills arrive with the next new session.
Rate limit messages
- The retry count now matches what Fuigo does. A rate-limited request now shows "(1/2)" while it waits, and "failed after 2
attempts" when it gives up, instead of counting against a limit it never reaches. A subagent that runs out of
retries also reports the number of calls it really made. Retry behaviour itself is unchanged.
Fuigo 1.0.37
Fuigo 1.0.37 release notes
Windows
- On Windows, a PATH ending in
;no longer makes every shell command ask. EmptyPATHentries (a trailing
;, or;;) mean nothing on Windows, so Fuigo now drops them from thePATHof the shell commands it runs, both
in its own shell and in an editor's terminal (it sends thatPATHwith the command when the editor runs it), and
no longer counts them as an entry that can steer which program runs. A relative entry,., or an entry inside the
project still makes commands ask, and macOS and Linux are unchanged (there an empty entry is the current folder). - Editor terminals now get the same environment check as Fuigo's own shell. When an editor runs a command for
Fuigo in its own terminal, a command approved without asking (a read-only command, or one Auto mode approved) is
now checked against the environment Fuigo sends with it, as Fuigo's own shell already did. If that environment
can change which program runs (for example a sessionPATHthat puts.or a folder of the project first, from
your settings or a trusted.envrc), the command does not run and Fuigo says why; an allow rule, your answer or
"allow all commands" still runs it.
Fuigo 1.0.36
Fuigo 1.0.36 release notes
A permission policy for one launch
A program that launches Fuigo (an editor, a CI job, an embedding app) can now hand it a permission policy for that
run. See Launch Policy.
- Set it with two variables.
FUIGO_POLICY_PATHnames the policy file (an absolute path) and
FUIGO_POLICY_SHA256must give its sha256. The file holds only a[permission]section inconfig.tomlsyntax
(allow,deny,ask,rules). - Per session too. An ACP client can add a policy to one session with
_meta["fuigo/permissionPolicyPath"]and
_meta["fuigo/permissionPolicySha256"]onsession/new, and sends both again on every load, resume and fork it
wants governed. The response confirms it in_meta["fuigo/permissionPolicy"]. - Its deny and ask win in every mode, always-approve included. An
askfrom the policy always reaches your client
as a permission request; remembered approvals don't answer it and answering it remembers nothing. Anallowacts
like your own config: it can remove a default prompt, never a deny or an ask. - A bad policy stops Fuigo before it starts. Unknown keys, malformed rules, a relative path, a symlink, a file over
256 KiB, or a missing or wrong hash: Fuigo prints one JSON line on stderr ({"policy_error": ...}) and exits with
code 78.fuigo policy check <file> [--json]tests a file the same way and prints its sha256. - The file is pinned for the life of the process, and the agent cannot edit it. To change it, start a new process.
- A governed process runs standalone: no shared leader, and hub routing is refused.
- Nothing changes when the variables are unset.
Browser mode gives the session back when you leave it
- Leaving
browser_userestores the session's own instructions. Pickingdefault,askorplanafter
browser_usenow gives back the session's own system prompt, its session rules and the client's system prompt,
exactly as they were. Before, browser mode replaced them and leaving it did not put them back. - Browser mode adds to the prompt instead of replacing it. While it is on, the browsing instructions are added
after the session's own prompt on each request; nothing is stored, and approvals are unchanged. browser_useis reported truthfully.session/set_mode,current_mode_updateandmodes.currentModeIdon a
live session now saybrowser_useinstead ofdefault.- A reloaded session starts in its normal mode. Browser mode lasts for the running session; after a restart the
session loads in its normal mode (its history replay still shows the earlier switch to Browser). - Sessions saved in browser mode by 1.0.35 load with their normal instructions. Session rules that 1.0.35 dropped
when it entered browser mode were never stored anywhere and cannot be recovered.
Session rules and memory survive mode and model changes
- Switching model keeps your remembered context. A model switch, or anything else that rebuilds the session's
instructions, now keeps the memory context and your session rules. Before, the memory context could be lost until
a restart. - Picking
defaultoraskkeeps the session's instructions. Adefault.mdorask.mdfile now supplies only
the mode's template: your session rules, the client's system prompt and the memory context stay.
A bare Read rule means file reads
Readinallow,denyoraskcovers file reads and listings (reading files, listing folders, grep, LSP,
memory files). It no longer reaches the pure bookkeeping tools (todo_write,ask_user_question,
scheduler_list,search_tool,refuse_peer_message), soask = ["Read"]no longer asks on every to-do update
anddeny = ["Read"]no longer blocks them. Every tool that can touch files is still covered; a rule with
tool = "any"still reaches everything.
Worktree copies report what they could not copy
- A tracked file that fails to copy stops the worktree. Creating or forking a worktree used to drop such a file
silently and report success; now the create stops with an error (a full disk says so) and the partial worktree is
removed. - Untracked and ignored files that fail to copy are a warning. One short line ("Worktree created, but 1 file had a
copy problem: ...") gives the count and the first file; the details are in the log. The same line shows when a
session is forked or resumed into a worktree, and headless runs print it on stderr.
MCP tools work with more Chat Completions providers
- Fuigo fills in what MCP tool schemas leave out when it sends them to a custom model over Chat Completions: a
missingtype, an object withoutproperties, an array withoutitems. Some providers rejected those tools.
Complete schemas are sent unchanged, other API paths and built-in tools are untouched, and a tool is never dropped.