`ai test` for a coding agent names the program it ran and, when that was not the saved one, the saved path it replaced; `doctor` says a saved path has no program there instead of blaming PATH. Refs #721.
Codex chats and builds start the Craft Anything MCP server with `--all-tools`, because Codex does not refresh its callable tools after `enable_toolset`; the server still applies Clean room restrictions. Fixes#538.
The builder's plan card can be folded away, and the fold is kept in the project's feed until the builder makes a new plan (refs #504).
Supersedes #779.
Co-authored-by: Fábio Soares <fabiodev1123@gmail.com>
When the Claude desktop app is installed but Claude Code isn't, the setup card and ai detect say so and how to install Claude Code (refs #259).
Supersedes #781.
Co-authored-by: FuryNocturn <28540819+FuryNocturn@users.noreply.github.com>
A folder or app bundle can't be saved as an agent's program (on a Mac the file picker returns the .app), and one saved before is never started: the usual program is used (fixes#467).
Supersedes #778.
Co-authored-by: Fábio Soares <fabiodev1123@gmail.com>
An agent that is saved but switched off is offered again when it's found on the computer, and choosing it turns it on, keeping its program and folder, instead of No coding agent was found staying on screen (refs #423).
A build whose checks keep finding no working build stops after 10 in a row, across Try again presses, until a new request starts the count again (kept in craft.json as not_working_checks), and with app/ empty the builder is asked to write the project and show its build's output instead of fixing a build that doesn't exist (refs #736).
Supersedes #776.
Co-authored-by: Fábio Soares <fabiodev1123@gmail.com>
A person's own project, edited in place with nothing to recreate, is checked by its own build and then its own tests, in any language a toolchain table knows (Rust, Node.js, Python, Go, .NET, Maven, Gradle, Make), instead of the checks made for a recreation (no --screenshot option, self-test or MECHANICS.md added to their code). A project with no tests, or one Craft Anything doesn't know how to build, isn't a failure and the note says so.
A project edited in place (Work on my project, or rebuild --existing edit) keeps the stack its own files use (an engine's project file, then Cargo.toml or package.json dependencies, then other project files), unless the request names another; a Godot folder no longer gets Unity from a request that reads as a game.
Tool calls a local model writes as fenced JSON after a sentence, or several in one reply, are run when every one names an offered tool; words after the last call, other code or an unclosed fence keep the reply as text.
When a server reports no input-token count, as local servers often don't, the system prompt and the tool schemas are counted from their text, so a conversation past its budget is trimmed (refs #731).
When a website build leaves app/ empty, the build check says the builder hasn't written the project, instead of only that no built site was found (refs #736).
An agent's chosen program path that an update moved (Codex's bin/<build>/codex.exe) is found again in the newest build folder, else by its usual name on PATH, instead of failing every turn (refs #721).
* The build check says where a project made one folder down is, and the Rust steps say Cargo.toml goes in app/ itself
* Tick R5.4's nested project line and add the folder mix-up it showed
* Every missing-project note names a project one folder down; only app/ is told its files belong there
* A builder on a local model isn't told to call memory tools it isn't offered
* Record the first live local builds and what stops them
* Keep R5.3's API-key live check open: the stop was seen on a local build
Each Claude Code AI can have its own folder (CLAUDE_CONFIG_DIR) in Settings → AI or with craft-anything-cli ai folder, so each can use a different account; the sign-in check, runs and sessions use that folder.
What Detect found sits under two tabs with counts, Providers (an API key or a local model server) and Harnesses (an installed coding agent), so more providers don't push the coding agents off the screen.
Moonshot, DeepSeek, xAI, Groq, Mistral, Together, Fireworks and DeepInfra are offered by name when their key is set, from one table of base URLs, key variables and default models.
GitHub Copilot CLI can be a coding agent over ACP, with its models chosen through the shared agent model catalogue; options that could bypass the app's tool restrictions are refused, and agents share investigation findings safely.
OpenCode Go can be added as a backend: offered when OPENCODEGO_API_KEY (or OPENCODE_GO_API_KEY) is set, on its documented base URL, with a stable per-installation x-opencode-session header kept in the AI settings. Not run against the service.
Local chat discovery reads a generic GGUF file name by its model folder and treats task names with version numbers as specialised, so an embedding or reranking model isn't picked for chat (follow-up to #653).
A Google AI Studio (Gemini API) key is its own choice in Settings → AI and first run, through Gemini's OpenAI-compatible endpoint, read from GEMINI_API_KEY like the other providers (fixes#646).
Claude Code chats get no built-in tools and builders an explicit list (commands, files, search, todos), so tools that message other sessions, publish or schedule work never start; in Auto mode a tool without an approval rule is refused (fixes#635).
Replays #576, #577, #581, #589, #624, #625 and #628. Builder steps stop after 25 failed builds in a row, OpenRouter tool requests ask for a provider that supports tools, a start that never opened a window draws in software next time, spinners stop redrawing every frame, local builders get a lean tool list, lldb call observation works and reports failures, and LM Studio is found on other ports.
Co-authored-by: Fábio Soares <fabiodev1123@gmail.com>
Detect offers Kimi K3.1 Pro (moonshotai/kimi-k3.1-pro) on OpenRouter instead of the older Kimi K3, with OpenRouter's list price ($2 / $10 per Mtok, $0.20 cached).
Replays #616, #617, #619, #620 and #621. Aider isn't offered until it has an adapter, Antigravity is found under a redirected LOCALAPPDATA, servers with only specialised models aren't set up for chat, an SSE event's data is capped, and a retry keeps the original request when the agent's session can't be restored. Program lookup adds lower-case extensions on Windows.
Co-authored-by: Afloat <132809511+Afloat16@users.noreply.github.com>
When the website check finds no built site, it names every place a site is accepted (dist/, build/, out/, or a plain index.html without package.json) and where the app really is (fixes#494).
In Clean room, a chat or question refuses the agent's own file, shell and web tools outright, so it works only through our tool server; this no longer rests on the agent's defaults (fixes#564).
Refs #44, #110 and #481. A build stops after two builder turns in a row change no app files, counted across runs in a versioned inbox/build-progress.json; a new request or a real edit starts another attempt. File digests are cached by size and modification time, so unchanged files aren't read again.
Refs #377 and #142. Builder retries are decided by typed failures instead of English text: sign-in, billing, missing programs, configuration and failed required tools stop at once, while dropped connections, rate limits and plain crashes are retried up to three times with bounded waits. OpenCode's JSON output is read for errors, text, tools and usage.
Refs #498. A tool argument of the wrong type is answered with its JSON Pointer, the type error and how to correct the call, so a model can fix it; schema types stay strict.
Supersedes #216. A prompt brought back with Up keeps the pictures it was sent with. feed.json moves to version 3: a User item with pictures is saved as {text, pictures}, one without stays a plain string, so older files read as before. Picture paths are absolute, so they don't come back after the project is moved.
Co-authored-by: muraa-p <muraa-p@users.noreply.github.com>
Refs #417. A complete JSON tool call a model writes as its whole reply (bare or fenced, with name and arguments or parameters, or a function wrapper) runs as that tool when it was offered, including when native tool calling was offered; examples in prose, malformed or partial JSON stay text, and Action:/Action Input: steps still work.
Fixes#549. Gemini CLI gets its message on stdin with an empty -p, so a long message no longer passes cmd.exe's 8,191-character limit through gemini.cmd.
Fixes#540. Codex's codex exec can't ask for approval, so setup and the builder approve rebuild_start (which asks the person itself) on Codex's side, keeping the person's other settings in the entry; a refused call explains the fix instead of showing null.
Over ACP, session/update notifications that session/load replays before it answers are kept out of the new turn, so a continued session no longer shows the earlier conversation again.
Fixes#506. Under a Flatpak, where ~/.claude is read-only, Claude Code is started with a writable CLAUDE_CONFIG_DIR in the app's data folder so its conversations are saved and --resume works; the sign-in message says how to sign in inside the sandbox. A CLAUDE_CONFIG_DIR the person sets wins, and SSH builds are unchanged.
Refs #374. A backend can set its own conversation budget (context_tokens in [[ai.backends]]); the conversation is fitted to it before each model attempt, including a fallback with a smaller window, counting history added since the last measured request.
Native tool calls keep a reported output-limit or refusal stop instead of being run as tool use, and calls left unrun after a pause or other stop get matching non-execution results so the next request stays valid.
The rebuild releases its keep-awake helper before reporting Finished, and dropping the guard waits up to five seconds for the helper to stop (logging if it doesn't), so a CLI that exits right away no longer leaves it running.
Refs #369. API errors show the provider's own reason from error.metadata.raw (with the provider's name) instead of only "Provider returned error"; unknown or malformed payloads are not dumped, and the message is bounded.
Fixes#528. Codex gets its message on stdin (codex exec -) instead of as the last argument, so a long message no longer passes Windows' command-line limit (os error 206).
* Claim: count a plain HTML site as built (R2.7)
* A plain HTML site passes the build check as it is, with nothing to build
* Note the keep-awake helper seen outliving a CLI run, and a search that found itself
* A blank page, or a game engine's project beside it, doesn't pass as a static site
* A static site with an assets folder isn't sent to Unity
Fixes#469. craft.json records which coding agent saved the session (agent_session_by). A build or a question continues the session only with that agent; after the builder changes, the new one starts from the brief. A follow-up to an agent with no saved session also gets the brief.
Justin John