* 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.
A picture over 20 MB is refused when it is added (prompt box, drag and drop, Make's picker) with a message, and copy_pictures refuses the whole list before copying anything, instead of dropping it silently when the message is sent.
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.
Each paragraph of an agent's answer is laid out as one width-limited text block, keeping bold, inline code and glossary underlines, so a long file link after a glossary term no longer pushes messages under the sidebar (fixes#523).
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.
CRC cpio checksum work shares one byte allowance across a firmware scan and stops when the scan is cancelled; checksum fields must be eight hex digits, and skipped verification is reported (R3.12). Replays #676.
Co-authored-by: Afloat <132809511+Afloat16@users.noreply.github.com>
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.
Auto mode's delete check compares paths as written and reads quoted or escaped paths with spaces as one argument, so it neither refuses deletes inside a project with capitals in its path nor lets a delete of the folder beside the project through. Replays #357.
Co-authored-by: samurai <samuraideng@gmail.com>
Older stand-in recordings keep their empty query pairs when current credential rules are reapplied, so replay matches the right page (follow-up to #648).
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).
On Linux a switched-off screen (a closed laptop lid) no longer counts as the one the GPU drives, and PCI ids pick a card only when no other card shares them (follow-up to #659).
The unterminated DER OID regression test ends with 0x81, so it reaches the arc-completion guard instead of the non-minimal-encoding check (follow-up to #652).
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).
A ZIP's end-of-central-directory record is the one whose comment length reaches the end of the file, so a signature inside the comment can't hide the real entries; a file with bytes after its comment still opens.
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).
On Linux the window is drawn on the GPU that drives the screen (the built-in panel's, or the only one with displays), so laptops with two GPUs open fast and don't lag; WGPU_ADAPTER_NAME and WGPU_POWER_PREF still decide (fixes#571). Replays #578.
Co-authored-by: Thân Minh Nghĩa <thanminhnghia3@gmail.com>
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).
The command palette follows its selection only when the arrow keys or typing move it, so the wheel's scroll isn't pulled back to the top on the next frame (fixes#584).
Pictures can be pasted into the prompt box with Ctrl+V (Cmd+V, Shift+Insert): a copied image is saved as a PNG in the person's own cache folder and attached, copied picture files are attached as they are, and attachments show as picture tiles.
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>
JavaScript analysis reads a source map comment from the last eight lines of the whole script and keeps its full URL, so originals are restored from large inline source maps.
"Always for this project" on a consent card is remembered in the project, and a tool allowed that way isn't asked about again (fixes#563). Replays #575.
Co-authored-by: Fábio Soares <fabiodev1123@gmail.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).
Replays #606–#615. Binary plists, base64, OIDs, string offsets, tar, cpio and source maps refuse or flag input they used to half-read.
Co-authored-by: Afloat <132809511+Afloat16@users.noreply.github.com>
Replays #602–#605. HAR reports redact percent-encoded credential names and declared secrets in every kept field, count decoded body bytes, and template only real dates in API paths; the stand-in backend matches older recordings with the current query rules.
Co-authored-by: Afloat <132809511+Afloat16@users.noreply.github.com>
Builder file tools and the approval policy follow links the way the file system does, so a symlink or junction inside a project can't lead a builder's read or write outside it (CWE-59); build_files_list shows links without walking into them.
The Ghidra client refuses a reply line over 128 MiB and closes the connection, so a hostile binary can't make one reply exhaust memory; a long reply split across read ticks is still read whole.
web_scripts_save copies only the scripts that fit its max_scripts and max_bytes; scripts past them are listed by URL only, so a large or hostile capture costs no more than the export's own limits.
A program that is on disk but can't start (an npm script whose node is missing) says where it is and what to check, instead of "isn't installed"; on Windows an .exe is started by the path it was found at (refs #152, #95).
A native call graph lists at most 50,000 call edges and says when it stopped there, so one function with a huge number of call sites can't blow up the report.
Fixes a flaky platform test: on macOS an accepted socket inherits the listener's non-blocking mode, so the keep-awake stand-in helper dropped a stop that arrived after accept.
Replays #597, #598, #600 and #601. The preview server answers only requests addressed to this computer, streams files, sends no body for HEAD on a missing file, caps open connections and gives a request head one deadline.
Co-authored-by: Afloat <132809511+Afloat16@users.noreply.github.com>
Replays #591–#596. The HTML scanner reads rel token lists, decodes asset URLs once, keeps script and title text through lookalike closing tags, and skips raw-text elements, markup inside attribute values and custom tag names.
Co-authored-by: Afloat <132809511+Afloat16@users.noreply.github.com>
BFlat