- Add hasKG flag to MemorySearchTool, inject hint in results when KG is enabled
- Wire SetHasKG(true) in gateway when KG store is available
- Improve knowledge_graph_search tool description with concrete use cases
- Update system prompt KG guidance to be more actionable
Two related fixes:
1. Memory interceptor now resolves workspace from request context
(per-user workspace) instead of using the static global workspace.
This fixes memory writes with absolute paths under per-user
workspaces (e.g. workspace/channel/userID/memory/) being bypassed
and written to disk instead of the database, which also prevented
KG extraction, memory indexing, and cross-session recall.
2. KG extractor: increase max_tokens 4096→8192, add retry on
truncation (finish_reason=length), and support chunking for
long inputs with deduplication on merge.
Inject user follow-up messages into the running agent loop at turn
boundaries instead of queueing them for a new run. This preserves
context so the LLM sees both tool results and user follow-ups together.
- Add InjectedMessage type and drainInjectChannel helper
- Add InjectCh to ActiveRun with buffered channel (cap=5)
- Drain injection channel at two points in agent loop (after tool
results and before no-tool-calls exit)
- Route steer/new_task intents to InjectMessage with scheduler fallback
- WebSocket: inject into running loop when session is busy
- Remove IntentClassify config toggle (always on)
- Web UI: show send + stop buttons side by side during agent run
- i18n: add injection acknowledgment messages (en/vi/zh)
BuildContext() and GetEntries() only read from RAM, so after gateway
restart or LRU eviction the agent never saw pending group messages
despite them existing in the database. Add lazy loadFromDB() that
queries ListByKey on cache miss and populates RAM for subsequent calls.
Agent presets in the setup wizard and create dialog were hardcoded
in Vietnamese. This change moves them into the i18n catalog so they
follow the user's selected language.
- Replace static AGENT_PRESETS array with useAgentPresets() hook
that reads from the i18n catalog via useTranslation
- Add presets.foxSpirit / artisan / astrologer entries to
en/agents.json, vi/agents.json, and zh/agents.json
- Update step-agent.tsx and agent-create-dialog.tsx to use the hook
Co-authored-by: weirguo <weir_guo@foxmial.com>
Tasks can be created with require_approval flag, starting in
pending_approval status. Users approve/reject via tool actions or
WS methods. Approval respects blocked_by dependencies — tasks with
unresolved blockers transition to blocked instead of pending.
Delegate agents are restricted from approving/rejecting.
MessageTool.parseMediaPath() was hardcoded to only allow files in /tmp/,
while all other filesystem tools (read_file, write_file, edit, exec) use
workspace-aware resolvePath() with restrict_to_workspace enforcement.
This meant agents could create files in their workspace via write_file
but couldn't send them as attachments — only /tmp/ files from
create_image/create_audio worked.
Replace parseMediaPath() with resolveMediaPath() that:
- Reuses resolvePath() for consistent security (symlink, hardlink, traversal)
- Honors per-agent workspace + restrict_to_workspace from context
- Still allows /tmp/ as fallback (for create_image, create_audio, etc.)
- Supports relative paths resolved against workspace
- Updates tool description so LLM knows about MEDIA: prefix
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
Add domain context, coreference rules, controlled relation types (15 predefined),
few-shot example, and dynamic entity count (3-15). Increase max input from 6000
to 12000 chars, reduce max output tokens from 8192 to 4096.
Replace direct ActivityStore injection with event-driven audit system.
Handlers emit audit events via msgBus.Broadcast(), a single subscriber
with buffered channel persists to activity_logs table.
Coverage expanded from 3 agent CRUD actions to ~65 audit points across
all HTTP handlers and WebSocket RPC methods including agents, providers,
skills, MCP servers, cron, sessions, teams, pairing, and more.
Add independent `share_memory` config flag to control memory and
knowledge graph sharing separately from workspace folder isolation.
- Add ShareMemory field to WorkspaceSharingConfig
- Decouple WithSharedMemory(ctx) from shouldShareWorkspace() in loop.go
- Add shouldShareMemory() helper independent of workspace sharing
- Fix KG Traverse CTE to scope user_id in recursive step (pre-existing bug)
- Add memory toggle UI with violet styling in workspace sharing section
- Add i18n translations (en/vi/zh) for new memory sharing controls
- Add unit tests for shouldShareMemory() independence
Recharts Brush component crashes when data length changes after
cross-filtering because its internal indices go out of bounds.
Add key={chartData.length} to force re-mount on data change.
Add ErrorBoundary wrapper for resilience against future chart crashes.
* docs: add brainstorm report for discord guild-user memory
* docs: update brainstorm report with corrected root cause analysis
* feat(discord): per-user memory scope in guild channels
Fixes shared USER.md between guild members by scoping userID to
"guild:{guildID}:user:{senderID}" for Discord group messages.
Updates all group-context prefix checks (write permissions, writer
cache, cron peer kind, history filter) to include the new guild: prefix.
Closes#165
* fix(cron): deliver reminders directly as bot messages instead of routing through agent
When deliver=true, send Payload.Message straight to the channel without
agent processing. This fixes reminders appearing as fake user messages
followed by an unnecessary agent reply.
Also auto-set deliver=true when cron jobs are created from chat context
(channel + chatID available), so reminders always deliver correctly
without requiring the LLM to explicitly set the deliver flag.
* feat(docker): add ENABLE_NODE build arg for MCP stdio servers
Optional Node.js/npm in Alpine runtime, default false.
Required for running MCP servers via npx inside container.
* feat(ui): add back button to setup wizard steps
Allow users to navigate back to previous steps during onboard setup.
Back button appears on steps 2-4, positioned left-aligned in the
action row with secondary (gray) styling.
* revert(cron): restore cron delivery to main version
* fix(ui): remove duplicate onBack in setup step interfaces
Memory (MEMORY.md, memory/*) and knowledge graph are now shared when
workspace sharing is active, matching the filesystem sharing behavior.
Previously memory was always per-user isolated even with shared workspace,
causing inconsistencies when collaborating on the same files.
Adds MemoryUserID(ctx) helper that returns empty userID (global scope)
when shared memory flag is set, used by memory interceptor, memory tools,
and KG search. UI warning updated to note data is not migrated on toggle.
Add workspace_sharing config in other_config JSONB to control per-user
workspace isolation. When enabled, users share the base workspace directory
instead of isolated subfolders — configurable separately for DMs and groups,
with a per-user allowlist override.
Backend: WorkspaceSharingConfig struct, ParseWorkspaceSharing(), conditional
isolation in loop.go/loop_history.go, 7 unit tests.
Frontend: prominent always-visible config section with contact search
combobox, sticky save bar layout fix, i18n (en/vi/zh).
Steps 2-4 now have a Back button. Steps with DB resources
(provider, agent) support edit mode — pre-fill form and use
UPDATE instead of INSERT on re-submit.
loadExistingCreds now distinguishes sql.ErrNoRows (safe empty map) from
real DB errors (propagated to caller). Prevents transient DB failures
from silently wiping existing credentials during partial updates.
1. Media tag enrichment (audio/video/document):
- Add enrichVideoIDs() — video media_id was never injected into
<media:video> tags, causing LLM to hallucinate UUIDs
- Fix all enrich functions to replace the LAST bare tag instead of
the first. When group history prepends older media tags, the first
occurrence belongs to history — injecting the current turn's ID
there causes the LLM to reference the wrong file
2. Gemini File API polling:
- Upload response returns fileURI immediately but file may still be
in PROCESSING state. Check state field; only skip polling when
file is already ACTIVE. Fixes "not in an ACTIVE state" errors
3. Channel instance credential merge:
- Partial credential updates (e.g. updating just token) now merge
with existing credentials instead of wiping other fields
- Loads, decrypts, merges, re-encrypts in a single Update() call
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
* feat(telegram): support custom Bot API server URL
Add api_server credential field for Telegram channels to allow
using a self-hosted Telegram Bot API server (e.g. for 2GB upload
limits). Uses telego.WithAPIServer() option.
* feat(telegram): local Bot API media download + config improvements
- Support local Bot API --local mode: detect absolute file paths and
copy directly from filesystem (requires shared volume mount)
- Fall back to HTTP download via custom API server URL for non-local mode
- Move api_server/proxy from credentials to config (non-secret fields);
credentials still accepted for backward compat
- Add media_max_mb config field (friendlier than media_max_bytes);
deprecated media_max_bytes still works
- Default to 200 MB max when local Bot API is configured (vs 20 MB cloud)
- UI: move api_server/proxy to config section, use MB for media size
* fix: use 127.0.0.1 instead of localhost in API server placeholder
Docker containers cannot resolve localhost to the host machine.
127.0.0.1 works reliably with Docker's default bridge network.
---------
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
* feat(infra): add runtime package support for skills
Install nodejs, npm, pandoc, github-cli + pre-install Python packages
(openpyxl, pandas, python-pptx, markitdown) and Node packages
(docx, pptxgenjs). Configure runtime dirs for agent pip/npm installs
with PIP_TARGET, NPM_CONFIG_PREFIX, NODE_PATH to enable dynamic
package installation in read-only container environment.
* feat(infra): add bundled skills with runtime package support
- Add 5 bundled skills: docx, pdf, pptx, xlsx, skill-creator from container skills-store
- Wire GOCLAW_BUILTIN_SKILLS_DIR env var in gateway and CLI
- Support optional runtime packages alongside dynamic skill loading
- Update Dockerfile to COPY bundled-skills at /app/bundled-skills/
- Add PIP_CACHE_DIR in docker-entrypoint.sh for clean pip installs
- Document bundled skills in 14-skills-runtime.md section 6
* feat(infra): remove ai-multimodal skill directory from bundled skills
Remove the ai-multimodal skill package as part of consolidating runtime
package support for bundled skills. This directory is no longer needed
in the bundled skills structure.
* feat(ci): add semantic release and Docker Hub publishing
Add go-semantic-release workflow to auto-create semver tags on merge to
main. Extend docker-publish to push all variants to both GHCR and
Docker Hub (digitop/goclaw).
* feat(skills): add system skills infrastructure with is_system column, dep scanning, and seeder
- Migration 000017: add is_system boolean column with partial index
- Store layer: UpsertSystemSkill, delete protection, IsSystemSkill
- ListAccessible auto-includes system skills (no grants needed)
- ListWithGrantStatus returns is_system field
- Dependency scanner: auto-detect deps from scripts/ or skill-manifest.json
- Dependency checker: verify system binaries, Python/Node packages
- Seeder: seed bundled skills into DB on startup (idempotent via hash)
- Gateway wiring: GOCLAW_BUNDLED_SKILLS_DIR env for bundled skills
- HTTP: delete guard (403), slug conflict check (409), rescan-deps endpoint
- UI: System badge, hide delete for system skills, rescan deps button
- Agent skills tab: "Always available" for system skills
- i18n: en/vi/zh keys for system skills, deps scanning
* feat(skills): conditional system prompt, skill manifests, and Zip Slip fix
- System prompt: only show package list when python3/node are available
- Add skill-manifest.json for pdf, docx, xlsx, pptx bundled skills
- Fix Zip Slip vulnerability in office/unpack.py (all 3 copies)
* refactor(skills): extract shared office code to _shared/ and deduplicate
Move office scripts (pack, unpack, validate, schemas, validators) from
duplicated copies in docx/xlsx/pptx to skills/_shared/office/ with
symlinks. Remove soffice.py (non-functional in containers) and update
SKILL.md references to use soffice binary directly. Update seeder
copyDir to follow symlinks.
Removes ~45K lines of duplicate code across 3 skills.
* fix(skills): address code review findings for system skills integration
- H1: Remove dead symlink branch in copyDir (filepath.Walk follows symlinks)
- H3: Fix rescan-deps to query ALL skills (including archived) and re-activate
when deps become available; add ListAllSkills() + Status field to SkillInfo
- H4: Add Status field to SkillCreateParams, stop overloading Visibility
- M1: Batch Python/Node dep checks into single subprocess per runtime
- M4: Add rows.Err() check in ListSkills to prevent caching partial results
* feat(skills): async dep checking with realtime WS events
Split Seed() into sync DB upsert + async CheckDepsAsync() goroutine.
Gateway startup no longer blocks on Python/Node subprocess dep checks.
- Seed() returns seeded skills list, all initially status="active"
- CheckDepsAsync() runs in background, emits skill.deps.checked per-skill
- skill.deps.complete event emitted when all checks finish
- Each failed dep check: archives skill + BumpVersion() for immediate
cache invalidation so next agent turn picks up the change
- UI: use-query-invalidation listens to skill.deps.* events → auto-refresh
skills list in realtime
* feat(skills): system skills integration with toggle, dep checking, and per-item install
- Add is_system, deps, enabled columns to skills table (migration 017)
- Seed bundled core skills (pdf, docx, pptx, xlsx, skill-creator) on startup
- PYTHONPATH-based dep detection — eliminates false positives from local modules
- Per-item dep install UI with individual status (installing/success/error)
- Enable/disable toggle for core and custom skills (independent of dep status)
- Re-run dep check when skill is toggled back on
- Inline skill thresholds: 40 skills / 5000 tokens before switching to search mode
- Fix UpsertSystemSkill: backfill null file_hash without bumping DB version
- Remove redundant skill-manifest.json files (replaced by deps JSONB column)
- Show author from frontmatter in custom skills tab
- Runtime checker for python3/pip3/node/npm availability
- WS events for dep checking/installing progress
- docs: add 15-core-skills-system.md, 16-skill-publishing.md
---------
Co-authored-by: Goon <duy@wearetopgroup.com>
The read_audio tool errors with "audio not found" because:
1. BuildMediaTags() creates bare <media:audio> tags without IDs
2. persistMedia() generates UUIDs stored in context, not in message content
3. LLM sees <media:audio> and passes it literally as media_id parameter
4. resolveAudioFile() tries to match "<media:audio>" against UUIDs — fails
Two-part fix:
- enrichAudioIDs(): embed persisted UUIDs into <media:audio> tags (like
enrichDocumentPaths does for documents), so LLM sees actual IDs
- resolveAudioFile(): sanitize tag-like media_id values and fallback to
most recent audio instead of hard error on unmatched IDs
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
When users create cron jobs from channel chats (Zalo, Telegram, etc.),
the LLM often omits the `deliver` flag (defaults to false). This causes
cron jobs to run successfully but silently discard results — the response
never reaches the chat where the user requested it.
Auto-detect channel context and default deliver=true for real channels,
skipping internal channels (cli, system, subagent, cron, delegate).
The existing auto-fill logic then populates channel and chatID from
context, ensuring delivery info is always captured.
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
Add GET /v1/traces/{traceID}/export endpoint that recursively collects
trace, spans, and child traces (delegations) up to depth 10, returning
gzip-compressed JSON. Add download button with icon+text in trace detail
dialog header.
- Set providerType on Suno case in registerProvidersFromDB (was missing,
causing misrouting for custom-named Suno providers)
- Fix dbTypeToMediaType: replace dead "openai" key with "openai_compat"
to match actual store constant, add "bailian" → "dashscope" mapping
- Clone entry.Params before injecting _provider_type to avoid mutating
the original chain config
Media tools (create_image, create_video, create_audio, read_audio,
read_video, read_document) routed API calls based on provider name
pattern matching (e.g. strings.HasPrefix(name, "gemini")). This breaks
when users give custom names to DB providers — a Gemini provider named
"chatgpt-sap-het" would be misrouted to the OpenAI-compat endpoint,
causing 404 errors.
Fix: carry the DB provider_type through OpenAIProvider, resolve it via
typedProvider interface in ExecuteWithChain, and inject as _provider_type
param for callProvider routing. Name-based heuristic kept as fallback
for config-file providers that don't have a DB type.
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
Provider and Model donut charts showed 0 because they used request_count,
which is only populated on channel-level snapshot rows. LLM-level rows
store counts in llm_call_count instead.
The PR #153 whitelist stripped byProvider on agent config save,
but byProvider is actively used in policy.go for per-provider
tool overrides. Add it to the whitelist and ToolPolicyConfig type.
The per-agent `imageGen` and `vision` fields in `ToolPolicySpec` (stored
in agents.tools_config JSONB) were added in d5cc5a7 (Feb 26) as the
original way to configure image/vision providers. When the media provider
chain system was introduced in 5815437 (Mar 8), these fields were kept
"for backward compat" but became dead code with no UI to manage them.
This causes a hard-to-debug issue: if an agent's tools_config contains
stale imageGen/vision data (set via API or leftover from DB), it silently
overrides the global provider chain configured in the builtin tools UI.
Users see the correct chain in the UI but the tool calls a completely
different provider/model, with no indication of why.
Changes:
- Remove Vision and ImageGen fields + struct definitions from ToolPolicySpec
- Remove associated context helpers (WithVisionConfig, WithImageGenConfig, etc.)
- Remove per-agent override injection in agent loop
- Simplify create_image and read_image to use chain as sole source of truth
- UI: whitelist known tools_config fields on save to clean stale DB data
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
Add alias support to the tool Registry so Claude Code skills can reference
Anthropic tool names (Read, Write, Edit, Bash, etc.) and have them resolve
to GoClaw canonical tools (read_file, write_file, edit, exec, etc.).
- Registry: aliases map, RegisterAlias(), resolve(), Aliases()
- Policy engine: auto-includes alias defs when canonical tool passes filter
- System prompt: alias entries in coreToolSummaries + missing use_skill
- Legacy toolAliases migrated to Registry.RegisterAlias() at startup
* feat(ui): multi-skill upload with client-side validation
Allow uploading multiple skill ZIP files at once with pre-upload
validation. JSZip parses each ZIP client-side to verify SKILL.md
presence, frontmatter format, and slug validity before upload.
- Add JSZip dependency (lazy-loaded, code-split ~30KB gzip)
- Create validate-skill-zip.ts mirroring server-side checks
- Rewrite skill-upload-dialog for multi-file with status badges
- Add concurrent validation, sequential upload with per-file progress
- Add empty SKILL.md check to backend upload handler
- Add i18n keys for all new UI strings (en/vi/zh)
* fix(ui): duplicate entries and validation hang in multi-skill upload
- Move pending list construction to assignment inside updater return
to prevent StrictMode double-invoke from pushing duplicates
- Wrap per-file validateSkillZip in try/catch so one failure doesn't
block Promise.all and leave entries stuck in "validating" state
* fix(ui): use static import for JSZip instead of dynamic import
Dynamic import("jszip") fails in browser - bare module specifiers
don't resolve at runtime. Use static import which Vite handles
via its module graph and code-splits automatically.
* feat(ui): add inline visibility toggle on skills table
Click the visibility badge on managed skills to cycle through
private → internal → public. File-based skills stay read-only.
* fix(ui): move dedup logic outside state updater in upload dialog
Avoids reading stale entries inside functional updater. Builds
pending list from current entries state before calling setEntries.
* fix(ui): auto-select first active agent when current agent unavailable
When agents load from API, if the current selected agent is not in the active agents list, automatically select the first available active agent instead of remaining unset. Prevents chat page from being unable to send messages when default agent selection is invalid.
* feat(ui): make agent display name editable in setup wizard
Allow users to customize the agent display name during onboarding instead of keeping it hardcoded to "GoClaw". Removed read-only state from the display name input and added a placeholder for guidance.
* feat: add document path enrichment and media filename support
Backend changes:
- enrichDocumentPaths() in agent/media.go: injects persisted file paths into <media:document> tags
- Document paths allow skills (e.g. pdf skill via exec) to access files directly
- chat.go: support new media format {path, filename} alongside legacy string paths
- Updated read_document tool description to guide agent on using path attribute
- Docker: add pypdf to Python dependencies for PDF processing
- Softened MUST language in read_* tool descriptions (changed to Call this)
Frontend changes:
- chat-input.tsx: attach filename with each uploaded file in media payload
- use-chat-send.ts: send media as {path, filename} objects instead of just paths
- i18n: add "uploaded_files" text in en, vi, zh locales
- chat-page.tsx: minor adjustment for media handling
Enables skills to process uploaded documents directly without intermediate copying.
Previously the spinner only cleared after a 120s timeout. Now a
useEffect watches the groups data and immediately clears the spinner
and polling when the compacting group's has_summary transitions to true.
ListMembers() now filters by a.status = 'active' so inactive agents
no longer appear in TEAM.md or delegation prompts, preventing leaders
from attempting to delegate to unavailable agents.
handleCreate() only invalidated agentRouter but missed emitting the
pub/sub event for TeamToolManager. Reuse invalidateTeamCaches() which
does both, matching the pattern used by all other team mutations.
ExecuteWithChain previously required all providers to implement
credentialProvider (APIKey/APIBase). OAuth-based providers like
CodexProvider (ChatGPT OAuth) don't expose static credentials,
causing all media tools to fail with "does not expose API credentials".
Make credentialProvider optional (nil when unsupported). Each
callProvider gracefully falls back to the provider's Chat() API
when credentials are unavailable. Generation tools (create_image,
create_video, create_audio) return a clear error since they require
direct API access with no Chat fallback.
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
Four per-agent settings stored in the database (and configurable via UI)
were silently ignored at runtime because the tool/system layer always
used the global config defaults instead.
**restrict_to_workspace**: Tools used the global config default baked at
startup. Fix: pass per-agent value through context; tools check context
override before falling back to constructor default.
**subagents_config**: ParseSubagentsConfig() existed but was never called.
All agents shared one SubagentManager with global limits. Fix: resolve
per-agent config in the agent resolver, store it on each spawned task,
and use it for limit checks, deny lists, and system prompt generation.
**memory_config**: Only the enabled toggle was read per-agent; search
weights (vector_weight, text_weight, max_results, min_score) were
hardcoded from PGMemoryStore defaults. Fix: extend MemorySearchOptions
with weight overrides, read per-agent config from context in the
memory_search tool.
**sandbox_config**: Only workspace_access was extracted per-agent; mode,
image, memory, CPU, timeout, network settings were discarded. Fix: pass
full sandbox.Config through context; Manager.Get() accepts an optional
config override for new containers.
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
The embedding provider resolution only matched 3 hardcoded names
(openai, openrouter, gemini), silently failing for DB-stored providers
like "openai-embedding". This caused memory chunks to be stored
without vectors even when a valid embedding provider was configured.
Changes:
- resolveEmbeddingProvider: fallback to provider registry for DB-stored
provider names when hardcoded match fails
- gateway startup: read per-agent memory config from DB (priority over
config file defaults) for embedding provider resolution
- memory IndexDocument: log embedding errors instead of swallowing them
- memory admin ListChunks: return full chunk text instead of truncating
to 200 chars, avoiding confusing partial content in the UI
Co-authored-by: Luvu182 <208665161+Luvu182@users.noreply.github.com>
- Replace h-screen with h-dvh for correct mobile viewport height
- Fix input font-size to 16px on mobile preventing iOS Safari auto-zoom
- Add safe area padding for notched devices (Dynamic Island, home indicator)
- Expand touch targets for icon buttons to ≥44px on touch devices
- Make dialogs full-screen on mobile with slide-up animation
- Add virtual keyboard detection for chat input positioning
- Smooth auto-scroll for incoming messages, instant scroll on user send
- Add overflow-x-auto to table pages for mobile horizontal scroll
- Fix grid layouts to stack on mobile (grid-cols-1 sm:grid-cols-2)
- Safe-area-aware toast positioning
- Landscape compact mode for phone landscape orientation
viettranx