LLMs hallucinate file paths in responses (e.g. ./system/generated/file.png)
when the actual file is at /app/workspace/teams/.../generated/file.png.
Backend: /v1/files/ now falls back to searching the workspace directory
tree by basename when exact path isn't found. Generated filenames
(goclaw_gen_*) include nanosecond timestamps and are globally unique.
Frontend: relative file paths now use just the basename in the /v1/files/
URL, letting the backend fallback find the actual file. Auth token
appended as ?token= query param for authenticated file access.
File preview dialog was showing {"error":"unauthorized"} because
fetch() calls to /v1/files/ lacked the Bearer token. Now appends
?token= query param (accepted by the files endpoint) to all resolved
file URLs including relative path detection.
* feat(discord): add writer management commands (!addwriter, !removewriter, !writers)
Port Telegram's file writer commands to Discord with guild-wide scope.
Uses wildcard scope (guild:{guildID}:*) to match per-user contexts.
Also fixes system prompt injection to find guild-wide writers.
* fix(discord): use per-user scope for writer self-check
CheckPermission with guild-wide wildcard scope (guild:{guildID}:*)
fails to match auto-bootstrapped per-user perms. Use per-user scope
(guild:{guildID}:user:{senderID}) which matches both per-user exact
and guild-wide wildcard stored patterns via matchWildcard.
* fix(discord): deduplicate writers in prompt injection, add context timeout
- Deduplicate writers by UserID after merging guild-wide and per-user
scope results in buildGroupWriterPrompt to prevent duplicates when a
user has both grant types.
- Add 10s context timeout to Discord writer command handlers instead of
using context.Background() to prevent DB calls from hanging.
---------
Co-authored-by: viettranx <viettranx@gmail.com>
- Add isBusy derived state (isRunning || teamTasks active) for stop button
and top bar status — reflects team task activity after main agent completes
- Align ActiveRunZone with MessageBubble layout: bubble wrapper for streaming
text, compact tool cards with divide-y borders
- Detect relative file paths (./path) in markdown and render as file chips
routed through /v1/files/ endpoint
- Fix auto-scroll jitter during streaming by using instant scroll and
ignoring programmatic scroll events in the near-bottom check
fix: Windows deployment — broken symlinks, missing packages, dep scanner false positives
- Dockerfile: add pdfplumber, pdf2image, anthropic, poppler-utils, bash to ENABLE_FULL_SKILLS
- Dockerfile: resolve broken git symlinks from Windows clone (core.symlinks=false)
- Dockerfile: strip CRLF from docker-entrypoint.sh at build time
- .gitattributes: enforce LF line endings for shell scripts
- dep_scanner.go: track nested subdirectories + scripts dir basename as local modules
Reorder team task reminders to inject before the user message instead of
after, ensuring the user's message is always the last in the messages
array. Proxy providers (LiteLLM, OpenRouter) reject requests ending with
an assistant-role message (unsupported prefill).
Also add a safety net in the OpenAI provider to strip any trailing
assistant message that might slip through in the future.
Remove leading tabs that wasted tokens in LLM system prompt.
Restore timezone example (Asia/Ho_Chi_Minh) and channel auto-fill info
that were dropped in #307.
Replace loose string matching ("502", "503", "504") with telegoapi.Error
type assertion — prevents false positives from error messages that
happen to contain these substrings. Also classifies 5xx as
post-connect errors since the server received the request.
- Replace sequential polling dispatch with semaphore-bounded goroutines
(max 20 concurrent handlers) to prevent polling hangs from slow media
downloads without risking unbounded goroutine growth
- Add sync.WaitGroup for graceful shutdown — Stop() waits for in-flight
handlers with a 15s timeout
- Add progressReader that aborts downloads stalled for 60s (no data)
- Return MediaError from resolveMedia so callers know what failed and why
- Annotate model content with error context when media is skipped
(e.g. "[sent media (image) — skipped: exceeds 20 MB limit]")
- Send Telegram reply to user when their file is too large or download fails
- Use errMediaTooLarge sentinel in all size-check paths for consistent
error classification
Add early detection and graceful degradation when the model exhausts its
iteration budget without producing a text response:
- Inject 75% budget nudge telling model to start summarizing
- Strip all tool schemas on final iteration to force text-only response
- Adaptive web_fetch maxChars (60K→20K→10K) via iteration progress context
- Bump DefaultMaxIterations 20→30 for web-research workloads
- Add userFriendlyError() helper to hide 5xx errors from users
- Add Loader2 animate-spin to all save/update/create/delete buttons
- Add useMinLoading(600ms) to StickySaveBar for visible feedback
- Move toast notifications from components into hooks (single source)
- Remove inline saved/saveError state in favor of toast system
- Add toast to all mutation hooks: agents, channels, skills, builtin-tools,
sessions, teams, storage, tts, cron, providers, config
- Add i18n toast keys for en/vi/zh across all namespaces
- Deduplicate toast calls in team board components
Model confused context-history media tags (<media:image>) with current
message media, triggering read_image on non-existent images. Change
lightweightMediaTags() to use descriptive brackets (e.g. [sent an image])
so LLMs can distinguish history descriptions from actionable media tags.
- Reduce quickClassify threshold from 60 bytes to 15 runes to prevent
false positives (e.g. "làm đơn giản thôi" no longer matches "thôi")
- Use utf8.RuneCountInString for accurate Unicode length check
- Replace substring match with whole-word boundary check to avoid
false positives like "nonstop" matching "stop"
- Remove statusKeywords; only exact "?" is fast-pathed as status query,
all other status queries go through LLM classification
- Increase LLM classify timeout from 5s to 10s
- Differentiate steer (inject into running loop) from newTask (queue
behind active run) instead of treating both as mid-run injection
- Fix TrimLeft → TrimPrefix to strip only one underscore separator
- Add registryName to indexedResult for parallel path cache
- Use registryName for bootstrapToolAllowlist and loopDetector checks
- Add server-side sanitization for ToolCallPrefix input
- Remove dead chatMessages alias and unrelated ParseStripAssistantPrefill
- Remove orphaned stripAssistantPrefill i18n keys
- Add 13 unit tests for StripToolPrefix
* feat(agent): support tool call prefix stripping for proxy providers
Proxy providers like LiteLLM and OpenRouter may prepend a prefix to
tool call names returned by the model (e.g. "proxy_exec" instead of
"exec"). This broke tool policy validation, registry lookup, and
hardcoded name checks for "team_tasks" and "spawn" in both serial
and parallel execution paths.
Add per-agent toolCallPrefix configuration that strips the configured
prefix from incoming tool call names before registry resolution. The
stripping is applied at resolveToolCallName() which is called before
permission checks, registry execution, and spawn/team_tasks detection.
- Add StripToolPrefix() supporting literal ("proxy_") and template
("{tool_name}") patterns
- Add toolCallPrefix to ToolPolicySpec with backward compat from old
"toolPrefix" JSON key
- Fix config save using spread operator to prevent dropping new fields
- Add UI input in Tool Policy section with i18n (en/vi/zh)
* chore: ignore AI tool config directories
Add .gemini/, .claude/, .opencode/ to .gitignore to prevent
committing user-specific AI tool configurations.
The scanner only tracked subdirectories as local modules, so cross-file
imports like `from extract_form_field_info import ...` were incorrectly
reported as missing pip packages, causing install failures.
On first start with a fresh Docker volume, mkdir for .runtime
subdirectories can fail due to a volume initialisation race condition,
causing the container to restart loop. The directories are only needed
for agent-installed packages (pip/npm), so the failure is non-fatal.
Address issues identified in PR #137 with a cleaner approach:
- Sandbox isolation: add SandboxCwd/ResolveSandboxPath helpers to map
filesystem tool paths to agent-scoped container subdirectories,
preventing cross-agent file access via read/write/edit/list tools
- DooD volume mounting: add resolveHostWorkspacePath with multi-strategy
container ID detection (/proc/self/mountinfo, HOSTNAME, os.Hostname)
and 5s timeout on docker inspect
- Sandbox hints: expand from 1 pattern (binary not found) to 6 patterns
(permission denied, network disabled, read-only FS, missing file,
resource limits) with MaybeFsBridgeHint for filesystem tools
- Nginx DNS: add Docker resolver (127.0.0.11) with dynamic upstream
variable to handle backend container IP changes
- MCP args: switch from comma-separated to space-separated parsing
with quote support for --flag="value with spaces" patterns
- Refactor: rename ExecTool.workingDir to workspace for consistency,
extract sandbox.DefaultContainerWorkdir constant
Replace single-value idRemap with queue-based idQueue so multiple tool
results sharing the same original ID pair correctly in encounter order.
Previously the map overwrote on duplicates, causing the second result to
be dropped and a synthetic placeholder to be emitted instead.
* fix(agent): handle duplicate tool call IDs with occurrence-aware sanitization
* refactor(agent): simplify duplicate tool call ID handling
Replace complex occurrence-aware queue system with simpler approach:
- uniquifyToolCallIDs: append runID+iteration+index to all IDs at
response time, guaranteeing cross-turn uniqueness via UUID
- sanitizeHistory: simple seen-set dedup for legacy sessions with
pre-existing duplicate IDs, no queue/counter logic needed
- Restore full problem description in sanitizeHistory docstring
- Add unit tests for both dedup paths and edge cases
---------
Co-authored-by: viettranx <viettranx@gmail.com>
The squash merge of #291 missed the parallel tool execution path.
Tool errors running in parallel still displayed as "Done" instead
of "Failed" when reloading chat history.
Lift file state from ChatInput to ChatPage so DropZone.onDrop can
append dropped files to the shared state. Previously drag-and-drop
showed the overlay but never passed files to the input.
Improves supportsThoughtSignature with providerType and model normalization.
Strengthens collapseToolCallsWithoutSig with TrimSpace and camelCase support
to handle 'Thinking=OFF' whitespace signatures and proxy key translations.
Simplified approach to fix duplicate tool_call_ids that cause HTTP 400 on OpenAI-compatible APIs (OpenRouter, vLLM, DeepSeek).
- uniquifyToolCallIDs: appends runID+iteration+index to all IDs at response time, guaranteeing cross-turn uniqueness via UUID
- sanitizeHistory: simple seen-set dedup for legacy sessions with pre-existing duplicate IDs
- Anthropic guard: skips ID rewriting when RawAssistantContent is present
- Unit tests for both paths + edge cases
Closes#283
Add "config" to channelInstanceAllowedFields so channel config changes
from the UI are actually persisted to the database instead of being
silently dropped by the allowlist filter.
- Add MiniMax-M2.7 to minimax_native provider model list
- Fix combobox dropdown UX: show all options on re-focus instead of filtering by current selection
- Add allowCustom prop to combobox for typing custom model names not in the list
- Add i18n key useCustomModel for EN/VI/ZH
- Audit all raw input/select/textarea elements across web UI to use text-base (16px) on mobile, preventing iOS Safari auto-zoom on focus
- Listen for session.updated event to update sidebar labels in-place
- Replace bg-muted/30 with bg-muted on chat cards so dotted background
pattern does not bleed through
After the first agent run completes, generate a short title from the
user's message using a 10s/50-token LLM call. Title is stored in the
session label column and pushed to clients via session.updated event.
Detect agents stuck in read-only loops (consecutive non-mutating tool
calls) and same-result patterns (identical outputs from different args).
Warns at threshold, force-stops at critical.
- Add deleteSession() to useChatSessions hook
- Trash icon on session rows (hover-reveal, always visible on mobile)
- Simple confirm dialog (no typing required)
- If active session deleted, switch to next or create new
- i18n strings for en/vi/zh
Root cause: ErrorBoundary in AppLayout used key={location.pathname},
causing React to unmount/remount the entire page on every URL change.
- Add stableErrorBoundaryKey() that strips dynamic segments from pathname
- Merge /chat and /chat/:sessionKey into single /chat/:sessionKey? route
- Derive sessionKey from useParams() instead of duplicating into useState
(dual state caused race condition: setState + navigate → B→A→B bounce)
- Move session reset from render-time setState to useEffect
- Only show loading spinner when messages array is empty
Detect exit code 127 and "command not found" patterns in sandbox
execution paths and append a [SANDBOX] hint to the error output,
so the model can inform the user about sandbox environment limitations.
Only applies to sandbox execution — host paths are untouched.
- buildNewSessionKey uses agent:{id}:ws:direct:{uuid} format
- sessions.list sends channel:"ws" to filter WS-only sessions
- isOwnSession supports both new and legacy key formats
- Read-only bar now floating with rounded corners
Runs idempotent regex migration at PGSessionStore init to convert
legacy WS session keys to canonical format. Handles multi-hyphen
userIDs (UUIDs) by matching last segment as timestamp.
viettranx