Sign each search request with HMAC-SHA256 using Hub ID, a per-request
nonce (UUIDv7), and unix timestamp. The signed reqId is sent in the
request body to prevent unauthorized API usage.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add localApprovalHandlers map so exec approval requests can be routed
to the Electron renderer via IPC instead of requiring a Gateway
connection. Expose setLocalApprovalHandler/removeLocalApprovalHandler
and resolveExecApproval on Hub for the IPC layer to use.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Rename CopilotHub references to Devv Search (constants, types, function
names, error message). Remove unused resolveTimeoutSeconds/resolveCacheTtlMs
imports and use constants directly.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Remove "requires API key" wording and rebrand to Devv Search across
tool definition, desktop UI, system prompt, and README.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Remove Brave and Perplexity providers in favor of a single CopilotHub
search endpoint (api-dev.copilothub.ai/web-search). Simplifies schema
to query-only, removes credential dependencies, retains caching.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Keep needsCompaction() method from compaction-feedback PR
- Keep tool result pruning from this branch
- Add "pruning" to CompactionEndEvent.reason type in events.ts
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adding explicit return type to help TypeScript resolve the type
correctly across different build configurations.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The CompactionResult type includes "pruning" as a reason but SessionEntry
did not, causing type errors when writing compaction entries.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two-phase pruning approach based on OpenClaw's microcompact-style:
- Soft Trim (30% utilization): Keep head 1500 + tail 1500 chars of large tool results
- Hard Clear (50% utilization): Replace old tool results with placeholder
Protections:
- Never prunes before first real user message (bootstrap protection)
- Protects last 3 assistant messages and their corresponding tool results
- Skips image-containing tool results
- Respects tool allow/deny lists
Enabled by default in tokens/summary compaction modes.
- Wrap maybeCompact() in try/catch to ensure compaction_end always fires
- Widen multicaListeners type to match subscribeAll() callback signature
- Import CompactionEndEvent from SDK instead of inline type casts
- Add doc comment explaining reason field type difference (agent vs SDK)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Profile: add usage hint for base path
- Memory Recall: clear step-by-step instructions for search and update
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Profile: concise directory path with single usage hint
- Memory Recall: one-liner instruction matching OpenClaw style
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements a simple memory_search tool for searching memory files:
- Searches memory.md and memory/*.md files by keyword
- Returns matching lines with context (2 lines before/after)
- Supports case-sensitive/insensitive search
- Respects maxResults limit
Tool is only available when a profile is active (has profileDir).
System prompt includes memory usage guidance when tool is present.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Introduce MulticaEvent type system parallel to pi-agent-core's AgentEvent,
with compaction_start and compaction_end events. Agent.subscribeAll() merges
both event streams. maybeCompact() now emits events bracketing the compaction
work, gated by a new SessionManager.needsCompaction() pre-check.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Memory is now managed through profile files (memory.md, memory/*.md) using
standard read/edit tools, following OpenClaw's file-first approach.
Changes:
- Remove memory/ folder with KV-based memory tools
- Remove group:memory from tool groups
- Update system prompt to remove memory tool references
- Update README docs to reflect file-based memory approach
Agents use workspace.md instructions to manage memory via file operations.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Log storage write errors in SessionManager.enqueue() instead of
silently swallowing them
- Remove existsSync check in ensureSessionDir to avoid TOCTOU race;
mkdirSync with recursive is idempotent when dir already exists
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The subagent announce flow reads the child's session file to extract
results, but saveMessage() is fire-and-forget (void enqueue). When
agent.run() completes, pending appendEntry writes may not have flushed
to disk yet, causing readLatestAssistantReply() to find no entries.
Add SessionManager.flush() and call it in AsyncAgent.write() after
agent.run() returns. This ensures waitForIdle() won't resolve until
all session data is on disk, fixing the "no return content" issue.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The mkdir call with { recursive: true } can throw ENOENT on macOS APFS
due to filesystem race conditions. This caused SpawnSession child agents
to silently fail — the session data was never written, so the parent
read back empty results ("子任务完成了，但没有返回内容").
Two fixes:
- Add retry logic to mkdir calls in acquireSessionWriteLock and
ensureSessionDir to handle transient ENOENT
- Add .catch() to SessionManager.enqueue() to prevent unhandled
promise rejections from fire-and-forget storage operations
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Update workspace.md template to prevent agents from claiming to
"remember" things without actually editing files. The new guidance:
- Uses "No Mental Notes!" as a strong warning
- Lists specific trigger phrases (记住, Remember, I prefer, etc.)
- Maps each trigger to the appropriate file to edit
- Explicitly forbids saying "I'll remember" without file edits
This addresses the issue where agents would acknowledge user
preferences verbally but not persist them to profile files.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Change `multica dev` default from gateway+console+web to desktop app
- Remove console from dev command options (use embedded Hub in desktop)
- Update package.json scripts to reflect new defaults
- Update README.md and CLAUDE.md architecture documentation
The desktop app has an embedded Hub, so no separate console/gateway is
needed for local development. Gateway is now optional, only needed for
remote client access.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Simplify 4-layer policy to 3-layer:
- Layer 1: Global allow/deny (user config)
- Layer 2: Provider-specific rules
- Layer 3: Subagent restrictions
Removed:
- ToolProfileId type (minimal/coding/web/full)
- TOOL_PROFILES constant
- getProfilePolicy function
- profile field from ToolsConfig
Users can achieve the same effect using allow/deny with group:* syntax.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Resolve tools before building subagent system prompt so the
"## Tooling" section is included, matching OpenClaw's pattern.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Only inject workspace.md into system prompt (not soul, user, memory)
- Add profile directory path to workspace section for on-demand file access
- Agent now reads soul.md, user.md, memory.md on first session using tools
- Reduces system prompt size and improves token efficiency
- Aligns with workspace.md template instructions ("Read soul.md first")
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add getProviderInfo() and setProvider() methods to Agent class
- Expose provider methods via AsyncAgent
- Add setLlmProviderOAuthToken() for storing OAuth credentials
- Extend ProviderConfig type with OAuth fields (oauthToken, oauthRefreshToken, oauthExpiresAt)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add optional onApprovalNeeded callback to exec tool (backward compatible)
- Thread callback through CreateToolsOptions → AgentOptions → resolveTools
- Add ExecApprovalConfig to ProfileConfig for per-profile configuration
- Create CLI terminal approval callback (readline-based) for non-Hub mode
- Export all exec approval types and functions from tools index
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add getMessages() to Agent and AsyncAgent for retrieving session history
- Fix type annotations in hub.ts for AgentMessage handling
- Remove duplicate type export in preload.ts
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add AsyncAgent.subscribe() that allows multiple subscribers to receive
the same agent events, enabling local IPC chat to coexist with other
event consumers.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Read reasoningMode from profile config and storedMeta when not
explicitly set via options (matching thinkingLevel pattern)
- Skip extractThinking() call when reasoningMode is "off"
- Clean up redundant ?? undefined casts in CLI entry points
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Wire reasoningMode through Agent runner, CLI output, and all CLI
entry points (run, interactive, non-interactive). In stream mode,
thinking content is printed to stderr in real-time. In on mode,
thinking is shown after message completion. Default is stream.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add ReasoningMode type (off/on/stream) to AgentOptions and related
config types. Add extractThinking() for extracting thinking content
blocks from LLM responses, mirroring the existing extractText() pattern.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- reloadSystemPrompt() now updates both agent and session prompt
- reloadSystemPrompt() uses buildSkillsPrompt() consistently (not buildModelSkillsPrompt)
- Extract rebuildSystemPrompt() to eliminate duplicated builder logic
- Preserve original tool name casing in tooling summary (matching OpenClaw)
- Fix report line count to use actual newline count
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
ProfileManager.buildSystemPrompt() now delegates to the structured builder.
Runner assembles prompt after tool resolution with safety, tooling summary,
and runtime info. Subagent prompts use minimal mode with safety constitution.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Structured, mode-aware system prompt builder inspired by OpenClaw.
Supports three modes (full/minimal/none), safety constitution,
conditional tool sections, runtime info, and prompt telemetry.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add AgentStyle type with 6 preset options (professional, friendly, etc.)
- Add getStyle/setStyle methods to ProfileManager
- Update soul.md template to include style section
- Add reloadSystemPrompt support for style changes
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add getSoulContent/updateSoulContent to ProfileManager
- Create soul.md with agent name when updating profile name
- Add reloadSystemPrompt() method to Agent and AsyncAgent
- Include soul.md content in system prompt build
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add methods to ProfileManager, Agent, and AsyncAgent for:
- Getting/setting agent display name (stored in config.json)
- Getting/setting user.md content
This enables the desktop app to manage agent settings.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Disable noUnusedLocals/noUnusedParameters in desktop tsconfig
(external src/ files have unused imports that fail strict checking)
- Add TSchema constraint to wrapTool generic to satisfy AgentTool type
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix originalToolsConfig assignment to handle undefined properly
- Fix devNull type cast for WritableStream compatibility
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
