Memory: add shared OpenAI embeddings rate limiter

… and command:new (M4)

Add three new hook capture points to the experiential-capture handler:

1. PreCompact (CRITICAL priority):
   - Always captures, bypasses rate limiting
   - Saves buffer snapshot with recent captures and evaluations
   - Records assistant text count and tool meta count

2. SessionEnd (command:stop):
   - Captures session summary with full buffer state
   - Includes recent captures and evaluation history
   - Records session lifecycle timestamps

3. Session Transition (command:new):
   - Captures session transition moment
   - Includes accumulated buffer state from ending session
   - Records both recent captures and evaluations

Also extends MeridiaExperienceRecord with 'kind' field (tool_result | precompact | session_end | session_transition) and adds new MeridiaTraceEvent variants for the new capture types.

Refactored handler to use shared helpers: resolveSessionContext(), resolveBufferPath(), loadBuffer().

…reflect MCP tools (M3)

- experience_capture: manually record significant experiences to JSONL+SQLite
- experience_search: FTS5 search with filters (session, tool, date range, score)
- experience_reflect: pattern analysis + reflection prompts on experiential records
- Registered in openclaw-tools.ts as Meridia experiential continuity tools

…fail

Codex/gateway connect error fastfail

feat(meridia): experience-capture, experience-search, experience-reflect MCP tools (M3)

…ions (M5)

- Add src/meridia/reconstitute.ts — core reconstitution engine
  - Queries Meridia SQLite for recent significant experiences (last 48h)
  - Formats compact briefing with key experiences, tool patterns, sessions
  - Configurable token budget (default 2000), lookback hours, min score
  - Gracefully handles missing SQLite or empty databases

- Add meridia-reconstitution bundled hook
  - Hooks into agent:bootstrap to inject MERIDIA-CONTEXT.md
  - Reads config from hooks.internal.entries.meridia-reconstitution
  - Supports maxTokens, lookbackHours, minScore config options
  - Non-fatal: sessions start normally if reconstitution fails

- Add src/meridia/reconstitute.test.ts — unit tests (6 passing)

Every new session now starts with experiential continuity awareness.

feat(meridia): expand experiential capture hooks — PreCompact, SessionEnd, command:new (M4)

Slack: add rich messaging payloads

feat(meridia): session reconstitution — morning briefing for new sessions (M5)

Introduces MeridiaDbBackend interface that abstracts away the underlying
database implementation. The SQLite-specific code is now isolated in
backends/sqlite.ts while consumers can program against the interface.

New files:
- src/meridia/backend.ts — MeridiaDbBackend interface definition
  with all DB-agnostic types (RecordQueryResult, SessionSummary, etc.)
- src/meridia/backends/sqlite.ts — SQLite implementation (SqliteBackend)
- src/meridia/backends/index.ts — Factory: createBackend(), closeBackend()
- src/meridia/backends/sqlite.test.ts — 25 tests for the backend interface

Changes:
- src/meridia/db.ts — Now a backward-compat shim, re-exports types from
  backend.ts. All functions still work but are marked @deprecated.
- src/meridia/query.ts — Types now re-exported from backend.ts for
  backward compat. Functions still work with DatabaseSync handles.

Key design decisions:
- Interface is DB-agnostic: no DatabaseSync or SQLite types leak through
- getMeta()/setMeta() added for key-value metadata (migration tracking etc.)
- SqliteBackend exposes rawDb getter for backward-compat shims only
- Factory caches singleton by path, like the existing openMeridiaDb()
- All 96 existing tests pass unchanged

Future backends (PostgreSQL, MySQL) can implement MeridiaDbBackend
with ~1 day of work per backend.

feat(meridia): pluggable DB backend interface (M2.5)

Codex/slack reasoning trace logging

Upstream sync 02 04 26

feat: dashboard fixes

…te w/cron

* refactor(tools): create shared tool execution module

* refactor(pi): use shared tool execution in pi-tool-definition-adapter

* refactor(sdk): use shared tool execution in tool-bridge

* test(tools): add tests for shared tool execution module

* docs: add unified tool execution coverage gap analysis

* fix(auto-reply): use unified tool execution for inline skill commands

* docs(tools): document wrapper reliance on runtime adapters

* test(auto-reply): add inline skill command error handling tests

* docs: update gap analysis with completed fixes

* fix: missing stuff

- Consolidate /agents/dashboard and /agent-status into a single route with grid/list layout toggle
- Reduce summary cards from 7-8 to 4 primary + conditional stalled/errored alert chips
- Declutter agent cards: combined session counts, cron badge in header, no redundant health metric
- Fix detail panel on mobile: use Sheet overlay on sub-lg screens instead of hidden panel
- Lighten activity timeline: borderless rows for completed items, full Card only for live tasks
- Add inline metadata stats (duration, tokens, cost) on timeline rows
- Batch consecutive search/code activities into collapsible groups
- Remove hardcoded mock data fallback in AgentActivityTab; show empty state instead
- Add keyboard navigation (ArrowUp/Down, Escape) and aria-live for screen readers
- Group developer actions (Copy ID, Copy JSON) into overflow dropdown menu
- Promote session section above generic actions in detail panel
- Extract shared format utilities (formatTokenCount, formatRelativeTime, etc.) into lib/format.ts
- Improve Lit dashboard: CSS grid metrics, channel badges, right-aligned tokens, hover highlights

Co-authored-by: Claude <noreply@anthropic.com>

Evaluates the original Meridia V2 architecture proposal against the
actual codebase state and existing architecture docs (ARCH.md,
COMPONENT-MAP.md, 7 component deep-dives). Produces a refined plan
that builds on the designed 14-component system rather than
reinventing it.

Key refinements over the original proposal:
- Two-pass evaluation (cheap gate, expensive phenomenology only on
  captured events) instead of full LLM extraction on every tool result
- Compaction-based graph sync with session-end mini-compaction instead
  of per-event Graphiti writes
- Vector search adapter interface (Graphiti vector first, pgvector
  optional) instead of hard PostgreSQL dependency
- Includes missing components (normalizer, gates, sanitization) the
  original proposal omitted
- Migrates to ExperienceKit type from COMPONENT-MAP rather than
  extending MeridiaExperienceRecord
- 5-phase implementation plan with dependency ordering

https://claude.ai/code/session_01Ko4xrCYytJBNW5VAxzfemE

…l, and reconstitution

Implements the Meridia V2 plan across all 14 components from COMPONENT-MAP.md:

**New modules (26 files):**
- event/normalizer: Component 1 — typed MeridiaEvent envelopes from hook payloads
- gates/budget: Component 2 — rate limits and capture budgets (extracted from hook)
- phenomenology/{taxonomy,prompt,heuristic,extractor}: Component 4 — two-pass
  phenomenological extraction with controlled vocabularies (17 emotions, 5 engagement
  qualities, 12 texture metaphors, 5 sensory channels from experiential-engine docs)
- artifacts/collector: Component 5 — file/link/media reference extraction
- kit/{builder,resolver}: Component 6 + Phase 5 — ExperienceKit assembly and
  meridia://<id> URI resolution
- sanitize/redact: Component 12 — secret redaction and payload truncation
- retrieve/{intent,ranker,hybrid,vector-adapter}: Component 10 — hybrid retrieval
  with blended ranking (significance, recency, diversity, source score)
- reconstitution/{engine,pack-builder}: Component 11 — structured ReconstitutionPack
  replacing bullet-list format with narrative, anchors, uncertainties, approach guidance
- fanout/dispatcher: Component 9 — fire-and-forget async dispatch with error isolation

**Updated existing files:**
- types.ts: Added Phenomenology, CaptureDecision, ArtifactRef, ExperienceKit,
  ReconstitutionPack types; added phenomenology field to legacy record
- hooks/experiential-capture: Uses normalizer, gates, phenomenology extractor,
  artifact collector, kit builder, and sanitizer
- hooks/meridia-reconstitution: Uses enhanced reconstitution engine with legacy fallback
- db/backends/sqlite.ts: Schema v2 migration (phenomenology columns), updated FTS
  searchable text, v1→v2 migration is idempotent
- meridia-search-adapter: Implements readFile() for meridia:// URI resolution,
  includes phenomenology in search snippets
- tools: Fixed pre-existing async/await issues across all three experience tools
- tests: Fixed async/await in sqlite and reflect tool tests, updated schema assertions

https://claude.ai/code/session_01Ko4xrCYytJBNW5VAxzfemE

* feat: added work-queue workflow workers and a worker manager, integrate w/cron

* fix: address 12 logic flaws in workflow engine, adapter, phases, and cron types

- engine: capture phase before clobbering in catch handler (#1)
- engine: remove dead `state.plan ?? plan` fallback (#2)
- engine: mark workflow failed when all execution nodes fail (#6)
- adapter: fix retry count off-by-one (attemptNumber vs retryCount) (#3)
- adapter: clean up abort listener in sleep on timeout (#10)
- discover: fix batch/entries index mismatch when spawns fail (#4)
- execute: add cycle detection to topological sort (#5)
- review: add autoApproved flag to distinguish fallback approvals (#7)
- plan: add comment clarifying sessionKey reuse across repair attempts (#8)
- decompose: remove unused model/maxPhases/maxTasksPerPhase/maxSubtasksPerTask opts (#9)
- types: add autoApproved to ReviewIteration (#7)
- cron/state: use discriminated union for CronEvent (CronJobEvent | CronHealthEvent) (#12)
- tests: add WorkflowWorkerAdapter test suite (8 tests) (#11)

https://claude.ai/code/session_01L8kquwpmUh5zmU9S4MHgPu

---------

Co-authored-by: Claude <noreply@anthropic.com>

…aSearchAdapter

Fixes P1 and P2 issues from code review:
- search() now properly awaits backend.searchRecords() (was calling .map on a Promise)
- status() now async with await on backend.getStats() (was reading undefined properties)

Resolved conflicts by accepting origin/main versions which include:
- Type-safe discriminated unions for CronEvent (job vs health events)
- Better workflow execution semantics (attemptNumber vs retryCount)
- Improved error handling and phase tracking in workflow engine
- Bug fixes for discovery question-to-result alignment
- Cycle detection in topological sort
- Auto-approval tracking in review iterations

All conflicts resolved in favor of main's improvements.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

…ntial backoff

- RetryableError: composable error boundary with visual retry feedback
- useRetry: React hook with exponential backoff, jitter, countdown timer
- withRetry: standalone retry utility for non-React contexts
- Comprehensive test suite with 10 test cases
- Card and inline variants, auto-retry option, cancel/reset controls

Closes P4 Error Handling UX: retry patterns for failed operations

- GlobalSearchDialog: full-featured search overlay powered by useGlobalSearch hook
- Category filter chips for narrowing results by type (agents, sessions, goals, etc.)
- Recent searches persisted to localStorage
- Keyboard shortcut Cmd+Shift+F / Ctrl+Shift+F to open
- Search button added to sidebar with keyboard hint
- Score-based ranking with grouped category results
- Comprehensive test suite for keyboard shortcut hook

Wires up the existing useGlobalSearch hook into a production-ready UI.
Addresses P4 Search & Filtering from BACKLOG.md.

* [autodev] Command Palette: Add nested sub-command menus and category filtering

Task 1 - Nested/sub-command menus:
- Replaced single page state with a page stack (PageEntry[]) for arbitrary nesting depth
- Added per-agent sub-menus: Agents > {AgentName} > View Details/Chat/Sessions/Edit Config
- Multi-level breadcrumb navigation with clickable intermediate crumbs
- Back navigation via Backspace (empty search) or clicking breadcrumb entries
- Agent sub-pages show agent info (role, status, current task)

Task 2 - Category filtering in search:
- Added filter chips bar that appears when searching on root page
- Categories: All, Quick Actions, Navigation, Agents, Appearance, Settings
- Active filter highlighted with primary color, dismissible via X button
- When filtering, shows all items in that category (not just top 3/5)
- Appearance and Settings items surface in search results when filtering

* [autodev] Mark command palette nested menus and category filtering as complete

* [autodev] Add progress indicators for long-running operations

New components:
- useAsyncOperation hook: State machine (idle → running → success/error)
  with progress tracking, auto-reset, and callbacks
- OperationProgress component: Visual indicator with three variants
  (inline, banner, compact) showing progress bar, step info, error/retry

Features:
- Step-by-step progress reporting (current/total/message/percent)
- Indeterminate progress bar for operations without step count
- Auto-reset to idle after configurable timeout
- Error state with retry button
- Success state with dismiss option
- Three display variants: inline (card), banner (full-width), compact (single-line)

Test coverage: 10 tests (all passing) covering state transitions,
progress tracking, auto-reset, callbacks, error handling

Barrel exports added to composed/index.ts and hooks/index.ts

* [autodev] Mark progress indicators task as complete

…cron types (openclaw#217)

- engine: capture phase before clobbering in catch handler (#1)
- engine: remove dead `state.plan ?? plan` fallback (#2)
- engine: mark workflow failed when all execution nodes fail (#6)
- adapter: fix retry count off-by-one (attemptNumber vs retryCount) (#3)
- adapter: clean up abort listener in sleep on timeout (#10)
- discover: fix batch/entries index mismatch when spawns fail (#4)
- execute: add cycle detection to topological sort (#5)
- review: add autoApproved flag to distinguish fallback approvals (#7)
- plan: add comment clarifying sessionKey reuse across repair attempts (#8)
- decompose: remove unused model/maxPhases/maxTasksPerPhase/maxSubtasksPerTask opts (#9)
- types: add autoApproved to ReviewIteration (#7)
- cron/state: use discriminated union for CronEvent (CronJobEvent | CronHealthEvent) (#12)
- tests: add WorkflowWorkerAdapter test suite (8 tests) (#11)

https://claude.ai/code/session_01L8kquwpmUh5zmU9S4MHgPu

Co-authored-by: Claude <noreply@anthropic.com>

* feat: event system for meridia

* fix: archive buffered moments after session summary to prevent inflation

Buffered moments were never transitioned after being consumed by
session-capture, causing them to be re-counted on every subsequent
/new command. Now transitions them to 'archived' after saving the summary.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

* feat: add configurable subsystem debug log suppression

* docs: add changelog entry and test script for suppressSubsystemDebugLogs

* docs: add implementation summary

* feat(meridia): multi-factor memory relevance scoring system

Implement a comprehensive scoring system that evaluates whether experiences
should become long-term memories based on five weighted factors:

- **Novelty**: Is this new information or repetition?
- **Impact**: Does this change understanding, behavior, or system state?
- **Relational**: Does this connect to known entities/people/projects?
- **Temporal**: Is this time-sensitive or evergreen?
- **User Intent**: Was this explicitly marked as important?

Features:
- Configurable weights for each factor (normalized to sum to 1.0)
- Named threshold profiles: balanced, aggressive, conservative, minimal
- Override rules for specific tools (glob-pattern matching)
- evaluateMemoryRelevance() as primary entry point
- shouldCapture(), shouldPersistToGraph(), shouldUseLlmEval() helpers
- Backward-compatible evaluateHeuristic() now delegates to scoring system
- New evaluateHeuristicDetailed() returns full scoring breakdown
- 37 comprehensive tests covering all factors and edge cases

Files:
- extensions/meridia/src/meridia/scoring/types.ts
- extensions/meridia/src/meridia/scoring/config.ts
- extensions/meridia/src/meridia/scoring/factors.ts
- extensions/meridia/src/meridia/scoring/index.ts
- extensions/meridia/src/meridia/scoring/scoring.test.ts
- extensions/meridia/src/meridia/evaluate.ts (updated)

* SDK runner: close message processing parity gaps with Pi embedded agent

- Gap 1 (HIGH): Add streaming block replies during the event loop via
  blockReplyBreak/blockReplyChunking params. Supports "message_end"
  (flush at message boundaries) and "text_end" (per-chunk) modes,
  matching Pi embedded's streaming delivery. Final onBlockReply is
  skipped when streaming was active to avoid duplicate delivery.

- Gap 2 (MEDIUM): Add onReasoningStream callback support. Extract
  thinking/reasoning text from SDK events classified as "system" and
  emit via the new callback, enabling reasoning display in UIs.

- Gap 3 (MEDIUM): Reset chunks/assistantSoFar at message boundaries
  so onPartialReply reflects only the current message, not accumulated
  text from all prior turns.

- Gap 5 (LOW-MED): Add enforceFinalTag support. When enabled, only
  content inside <final>...</final> tags is returned, preventing
  intermediate narration leakage from non-Claude models. Adds shared
  extractFinalTagContent() utility.

- Gap 6 (LOW-MED): Map all payload fields (mediaUrl, mediaUrls,
  replyToId, replyToTag, replyToCurrent, audioAsVoice) in
  adaptSdkResultToPiResult to prevent silent field loss.

Wire all new params through the full chain:
  executor.ts → AgentRuntimeRunParams → sdk-agent-runtime →
  sdk-runner-adapter → sdk-runner

https://claude.ai/code/session_013FRzxnpzCwncubghL7GGCF

* additional merge conflict resolution

---------

Co-authored-by: Claude <noreply@anthropic.com>

* feat: meridian memory clasiffication

* fix: minor safety check around config missing

…ross 13 files) (#207)

Cover all new V2 modules: event normalizer, gates/budget, sanitize/redact,
phenomenology (taxonomy, prompt, heuristic), artifacts collector, kit builder
and resolver, retrieve (intent, ranker, vector-adapter), reconstitution
pack-builder, and fanout dispatcher. Export renderKit from resolver for
testability.

https://claude.ai/code/session_01Ko4xrCYytJBNW5VAxzfemE

Co-authored-by: Claude <noreply@anthropic.com>

* feat: phase 0 to 3 for meridia enhancements/refactoring

* feat(meridia): phase 2 refactoring — sanitize, migrations, fanout, adapter tests

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: clean up stray brace and unused imports from merge resolution

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix(meridia): use readPositiveNumber for compaction/capture thresholds

readNumber accepts 0/negative values, but scheduleIntervalHours,
minExperiencesForCompaction, maxPerHour, minIntervalMs etc. must be
positive. Previously, the old local helpers enforced > 0; this was
lost when readNumber was extracted to event.ts. Switch to
readPositiveNumber which falls back to defaults for non-positive values.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

* feat: add configurable subsystem debug log suppression

* docs: add changelog entry and test script for suppressSubsystemDebugLogs

* docs: add implementation summary

* feat(meridia): multi-factor memory relevance scoring system

Implement a comprehensive scoring system that evaluates whether experiences
should become long-term memories based on five weighted factors:

- **Novelty**: Is this new information or repetition?
- **Impact**: Does this change understanding, behavior, or system state?
- **Relational**: Does this connect to known entities/people/projects?
- **Temporal**: Is this time-sensitive or evergreen?
- **User Intent**: Was this explicitly marked as important?

Features:
- Configurable weights for each factor (normalized to sum to 1.0)
- Named threshold profiles: balanced, aggressive, conservative, minimal
- Override rules for specific tools (glob-pattern matching)
- evaluateMemoryRelevance() as primary entry point
- shouldCapture(), shouldPersistToGraph(), shouldUseLlmEval() helpers
- Backward-compatible evaluateHeuristic() now delegates to scoring system
- New evaluateHeuristicDetailed() returns full scoring breakdown
- 37 comprehensive tests covering all factors and edge cases

Files:
- extensions/meridia/src/meridia/scoring/types.ts
- extensions/meridia/src/meridia/scoring/config.ts
- extensions/meridia/src/meridia/scoring/factors.ts
- extensions/meridia/src/meridia/scoring/index.ts
- extensions/meridia/src/meridia/scoring/scoring.test.ts
- extensions/meridia/src/meridia/evaluate.ts (updated)

* SDK runner: close message processing parity gaps with Pi embedded agent

- Gap 1 (HIGH): Add streaming block replies during the event loop via
  blockReplyBreak/blockReplyChunking params. Supports "message_end"
  (flush at message boundaries) and "text_end" (per-chunk) modes,
  matching Pi embedded's streaming delivery. Final onBlockReply is
  skipped when streaming was active to avoid duplicate delivery.

- Gap 2 (MEDIUM): Add onReasoningStream callback support. Extract
  thinking/reasoning text from SDK events classified as "system" and
  emit via the new callback, enabling reasoning display in UIs.

- Gap 3 (MEDIUM): Reset chunks/assistantSoFar at message boundaries
  so onPartialReply reflects only the current message, not accumulated
  text from all prior turns.

- Gap 5 (LOW-MED): Add enforceFinalTag support. When enabled, only
  content inside <final>...</final> tags is returned, preventing
  intermediate narration leakage from non-Claude models. Adds shared
  extractFinalTagContent() utility.

- Gap 6 (LOW-MED): Map all payload fields (mediaUrl, mediaUrls,
  replyToId, replyToTag, replyToCurrent, audioAsVoice) in
  adaptSdkResultToPiResult to prevent silent field loss.

Wire all new params through the full chain:
  executor.ts → AgentRuntimeRunParams → sdk-agent-runtime →
  sdk-runner-adapter → sdk-runner

https://claude.ai/code/session_013FRzxnpzCwncubghL7GGCF

* additional merge conflict resolution

* Merge origin/main into claude/review-message-processing-djLHt

Resolve merge conflicts in meridia scoring module by taking
origin/main's refactored multi-factor scoring system (scorer.ts,
defaults.ts, updated types/factors/index). Remove HEAD-only
scoring/config.ts and scoring.test.ts that used the old API
(origin/main has proper tests in factors.test.ts and scorer.test.ts).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix(meridia): add missing awaits in CLI + fix sanitizeForPersistence JSON corruption

---------

Co-authored-by: Claude <noreply@anthropic.com>

Audit the current apps/web/ codebase against the original 15-item
UX improvement report. Removes 2 items already implemented (ErrorState
component, home panel hierarchy), identifies 5 bugs (fake AI response
in production, missing error boundary, null safety gap, demo delays,
loading state inconsistency), and adds 3 new improvement items
(ErrorState adoption, skeleton loading gaps, consistent back nav).

https://claude.ai/code/session_01NsfuXFnsTT5iq9ENWEb95u