eframe makes a window icon the app's Dock icon on macOS, and the window icon was the full tile, not the bundle's icon padded to Apple's icon grid, so it looked bigger than other apps' (#102). On macOS the window gets an empty icon, which keeps the bundle's (no icon at all would bring eframe's own).
When UI Automation's button or check box proxy fails (it uses SendInput, which needs an interactive desktop), the walkthrough posts BM_CLICK to the control's window instead. PostMessage, so a click that opens a dialog doesn't block the walkthrough; only for buttons and check boxes.
Look for AI reads provider baseURLs from OpenCode's settings and probes those on this computer or its network, so LM Studio on another port or a server on another machine is offered. The JSONC reader moves from setup to formats so ai can use it. Fixes#83.
Taking over a replaced stand-in's port tries a busy port again for up to two seconds, so a restart no longer fails now and then with Address already in use.
The Settings tab column takes exactly what's left of the window with room for the scroll bar, and a path field leaves room for its button, so the Engines tab no longer widens the window. Fixes#68.
The conversation's history counts the item spacing and the prompt row is top-aligned, so the Agent console's panel no longer grows 17 points a frame until its tabs are cut off.
frida_devices_find runs frida-ls-devices in a pseudo console on Windows, where it crashed on pipes and hung; frida_targets_list falls back to plain processes; device columns split on two spaces. Fixes#58.
runtime_captures_compare's right input had no description, so its schema was the boolean true and Claude Code rejected the whole tool list in Deep-mode builds. It gets a description, and a test checks every MCP input property schema is an object. Fixes#90.
Linux packages put Ghidra in /opt/ghidra (Arch and derivatives) or /usr/share/ghidra (Debian, Kali) with only a start script on PATH. The engine table lists both folders and the Ghidra lookup falls back to them.
codex exec refuses to run outside a git repository or a folder it trusts unless it gets --skip-git-repo-check, and project folders aren't repositories, so every Codex build failed at once. The adapter passes the flag, once, even when the backend's own args already have it.
The DMG's app had only craft-anything, so the app couldn't start its MCP server and every Claude Code build failed with no tools. The CLI is now copied into Contents/MacOS, signed like its zip, and run from the finished DMG before release.
The PLT, prologue and pointer scans in disasm/program.rs use checked steps, so code near u64::MAX no longer wraps to 0 and walks the whole address space. Adds open roadmap lines for an A32 STRD/LDRD decoder panic and unchecked header offsets in dart.rs.
WinGet's Links folders, per user and machine-wide, join the folders searched after PATH, so an app started before a WinGet install of Claude Code still finds it.
engines_list, doctor, capabilities and Requires::Engine share Settings::engine_path, which honours the Ghidra folder setting, and locate() finds an engine unpacked into a versioned sub-folder.
A .utoc header naming billions of zero-length compression methods no longer collects empty names until memory runs out: only the 255 a block can point at are read, the rest are skipped by length.
va_to_off added a PT_LOAD segment's file offset to an address delta unchecked, so a segment whose p_offset is near u64::MAX panicked in debug builds and wrapped in release. It now uses checked_add, so such an address maps to nothing. Two regression tests, one with a crafted ELF64.
A group descriptor could point the inode table so far out that an inode started within 128 bytes of usize::MAX; the two checks that an inode fits in the image added 128 unchecked, which panicked in debug builds and wrapped in release. They now use checked_add, so such a root inode is refused and such a child skipped. Two regression tests, and a roadmap line for the same pattern in macho.rs method_list.
`ai test` on a coding agent only runs its `--version` and showed that line as the reply, so it
looked like a working answer while every real message could still fail (not signed in, no model,
no quota). The report now carries a typed `version_only` field (true for coding agents, false for
backends that answered a real request), and the doc comment says what the test does.
Fixes#48
An agent CLI's `model` names the agent (opencode) and the model it should use goes in cli_model,
which nothing in the app shows. People write the program in `command` and the model in `model`
instead (command = "opencode", model = "deepseek/deepseek-flash"), which ran the plain adapter
with the agent's own default model until its free quota ran out. Settings read that way now run the
agent the command names, asked for that model (OpenCode gets --model). A model that names an
agent, a cli_model already set, and unknown programs are left as they are.
Fixes#34
Doctor checked a local (OpenAI-compatible) AI server by its open port alone, so any program on
that port, such as a plain web server on 8080, was reported "ok" while every chat with it failed.
It now asks the server for its model list, as it already does for every other kind of AI, and
reports what came back ("answered 404: 404 page not found") when that fails.
Fixes#37
Builders driven by our own loop (an API model such as Ollama) get file tools whose paths are
relative to the project. Models often write the project's folders from the root (/app/src/…), which
was refused as outside the project, so they could neither list, read nor write app/ and went round
in circles while each check found nothing. A path from the root whose first folder the project has
is now read inside the project (/etc/… and folders it doesn't have are still refused), and a write
outside app/ says so in its result, so a model that leaves out app/ learns where the app goes.
Refs #24
In a VMware Windows VM the virtual GPU is reached through GL, which drew a black window, while
Windows' software renderer (WARP, over DX12) works. The desktop app now picks its graphics adapter
itself on Windows: GL comes after every other adapter, WARP included, and otherwise the fastest
device first, as before. GL stays the last resort when it's all there is, WGPU_BACKEND still
narrows the choice, and the adapter used is written to desktop.log.
Fixes#27
install checked for a pinned download before it looked at the sha256 it was given, so on FreeBSD
(no pinned adb) a malformed hash came back as "no pinned download" instead of "sha256 must be 64
hex characters", and rejects_bad_or_different_hash_before_downloading failed there on every run,
main included. The hash's form is now checked first; a well-formed hash is still compared with the
pin.
Local model servers were found by an open port alone, so any program listening on 8080 (a common
port for local web servers) was offered as "llama.cpp server (local)" with no model, and chatting
with it failed with "answered 404: 404 page not found". A local server is now offered only when
its model list answers; one that answers with an error or a page that isn't JSON is something
else. A server that doesn't answer in time is still offered, as before, since it may be loading.
Fixes#30
OpenCode had no adapter of its own, so it got the plain one: the prompt as its only argument.
OpenCode reads its first argument as the folder to open, so every chat and build stopped at once
with exit code 1. It now runs `opencode run [--auto] [--model …] <prompt>`; builds pass `--auto`
so what OpenCode would stop to ask is allowed unless its own settings deny it, as Codex builds get
a workspace-write sandbox. Flags checked against OpenCode's own `run` command.
Fixes#32Fixes#15
Release builds embed craft-fonts since the switch to the shared workflows, which turns the empty
boxes for Korean, Chinese, Japanese, Thai, Arabic, Hebrew, Hindi and arrows into text. A test now
checks every one of those in each of the app's font styles whenever craft-fonts is built in (as in
CI), so a font change can't bring the boxes back unnoticed. Packages now also ship craft-fonts'
OFL licences (OFL-<font>.txt), which the fonts require, and the release docs, ATTRIBUTION.md and
build.rs say how fonts are built in now (build.rs named the wrong app font).
Fixes#19
When a coding agent or tool crashed with a stack trace, the error's headline was the trace's last
frame ("OpenCode stopped with exit code 1: at <anonymous> (/$bunfs/root/chunk-….js:96:6417)"),
which says nothing about what went wrong. The headline is now the line above the trace (searched
in all of stderr, since a trace can outrun the excerpt), with Java's "Caused by" and Node's
version footer handled; output without a trace keeps its last line.
Refs #15
schemars writes `"items": true` for a list of any JSON (app_call_path_find's `functions`), which
JSON Schema allows but LM Studio refuses ("JSON schema error at #/properties/functions/items:
schema must be an object"), so every chat with it stopped. Tool schemas sent to a model now write
`true` as `{}` and `false` as `{"not": {}}` wherever a schema sits (to a bounded depth, since
run-time tools bring their own); `additionalProperties` keeps its boolean.
Fixes#17
npm installs Gemini CLI, Codex and often Claude Code on Windows as .cmd shims. They were found,
but Rust starts only .exe files by bare name, so every build stopped with "gemini isn't installed
or isn't on your PATH". An npm (or pnpm) shim's script now runs with Node directly, so a prompt
with line breaks or over cmd.exe's 8191 characters arrives whole; any other batch file starts by
its full path. The usual Windows install folders (~\.local\bin, npm, bun, cargo) are searched
after PATH.
Fixes#18
The status row names the command or file under way instead of only counting
actions. "More detail" opens the latest lines: stages, every command and file
with how it ended, what the builder said, and what programs printed outside the
conversation. Everything is kept in build-activity.log, with secrets hidden, and
the rebuild_activity_show tool and the CLI read the same.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Claude Code's figure is its own guess at API prices, which a subscription
isn't charged. The project page now says "AI cost (estimate)" with a note
saying so, and the title bar says "(estimate)". What an API key was charged
is kept apart in usage.json; older files count as estimates.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A run with no message was read as a first build, so Try again studied the
original again and gave the builder the first brief, even with its session
saved. A run now carries on when the builder has already worked, and only
runs the check when the builder had finished (awaiting_check in craft.json).
A check that returns an error couldn't look at the build at all, so it gets a
second go, and then the run says the build is finished but couldn't be checked.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Stop deadlocks as they form and watch every lock for one that never lets go
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Give the stuck-lock banner's test a program of its own
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Explain failed agent runs, keep a log to open, and stop offering git servers as computers
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Fix a lock-up: the build's agent setup took the session lock twice in one statement
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
SSH hosts that point at a git server (named after one, or with its HostName or User git) are left out of the AIs found on the computer and the remote-machine picker. Programs are looked up and started with PATH plus the usual install folders (~/.local/bin, Homebrew, npm, bun, volta, cargo), which a Mac app opened from the Dock doesn't get.
bflatastic