Add Google Calendar, self-modify, brain memory, and reliability improvements#248
Open
jihwan97-hash wants to merge 37 commits intocloudflare:mainfrom
Open
Add Google Calendar, self-modify, brain memory, and reliability improvements#248jihwan97-hash wants to merge 37 commits intocloudflare:mainfrom
jihwan97-hash wants to merge 37 commits intocloudflare:mainfrom
Conversation
Configure explicit Anthropic provider with all required fields: - Add baseUrl (required for custom provider config validation) - Define available Claude models (Opus 4.5, Sonnet 4.5, Sonnet 4, Haiku 4.5) - Set Claude Sonnet 4.5 as the default primary model - Include API key in provider config when available This fixes "Unknown model" errors that occurred because moltbot's built-in catalog doesn't include newer Claude model IDs. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add CLAUDE_ACCESS_TOKEN and CLAUDE_REFRESH_TOKEN support for Claude Max subscription - Create OAuth auth profiles in start-moltbot.sh for OpenClaw authentication - Map OAuth token to ANTHROPIC_API_KEY for backward compatibility - Add Brave Search API integration for web search functionality - Upgrade to openclaw@2026.2.1 in Dockerfile - Update .gitignore to exclude sensitive runtime configs Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add plugins.entries config for Telegram/Discord/Slack - Update config paths to use native OpenClaw paths (.openclaw/openclaw.json) - Update sync.ts to support both legacy and new config paths - Add CLAUDE.md agent instructions template - Force container rebuild with updated start-moltbot.sh Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Support CLAUDE_ACCESS_TOKEN for Claude Max subscription OAuth tokens - Create auth-profiles.json with correct format for OpenClaw - Add automated health check in cron handler (every 5 minutes) - Auto-restart gateway if not responding - Update validation to accept OAuth token as API key alternative Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fixed multiple syntax errors in start-moltbot.sh - Changed bind mode from 'any' (invalid) to 'lan' - Simplified to minimal config that starts reliably - Telegram configuration crashes the gateway - needs manual setup via OpenClaw UI Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update startup script to use openclaw doctor --fix for auto-configuration - Telegram token is now read from TELEGRAM_BOT_TOKEN env var - Remove manual config keys that OpenClaw doesn't recognize - Bump version to v45 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Sync OpenClaw config/pairing data to R2 every 60 seconds - Restore from R2 backup on container startup - Pairing, sessions, and credentials now persist across restarts Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Replace rsync with `timeout 60 cp -rf` for syncing to R2 - Replace rsync with `timeout 30 cp -rf` for restoring from R2 - S3FS can hang indefinitely on file operations, timeout prevents this - Check for openclaw.json file existence instead of directory contents Tested: Pairing persists across container restarts Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add "model": "anthropic/claude-sonnet-4-5" to agent defaults - Update startup script to v49 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
## Performance - Parallel R2 restore and config write in startup script - Conditional doctor execution (only when channel tokens set) - Startup timing logs for debugging ## Reliability - Auto-recovery with exponential backoff (3 attempts, 2s/4s/8s) - Enhanced /api/liveness endpoint with detailed health checks - Error trap in startup script with line number logging ## Features - GET /api/admin/conversations - List conversation sessions - GET /api/admin/conversations/:id - View specific conversation - GET /api/admin/skills - List installed skills - Enhanced /api/admin/storage/sync with duration and size - GET /debug/disk - Show disk usage - POST /debug/gc - Garbage collection for old files Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
OpenClaw expects model config as an object:
"model": { "provider": "anthropic", "model": "claude-sonnet-4-5" }
Not as a string:
"model": "anthropic/claude-sonnet-4-5"
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
R2 restore was overwriting the new config with old backup. Now: restore first, then write config to ensure correct format. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Clone a configurable GitHub repo into the workspace on boot and symlink OpenClaw bootstrap files (IDENTITY.md, SOUL.md, etc.) so the agent loads its identity automatically on every new session. Configured via GITHUB_REPO_URL and GITHUB_TOKEN secrets. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Sets dmPolicy to "allowlist" and writes the owner's Telegram user ID (via TELEGRAM_OWNER_ID env var) to the allowFrom file on each boot, so the bot responds immediately without requiring manual pairing. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace `exec openclaw gateway` with a restart loop that automatically restarts the gateway on crash (backoff 5s-120s, max 10 retries, resets after 60s of successful running) - Add GITHUB_PAT as fallback when GITHUB_TOKEN is not set Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The stored git remote URL kept the old revoked PAT token, causing pulls to fail. Now updates the remote URL with current credentials before every pull. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Runs restore-crons.js from the cloned repo in the background after the gateway port is ready, restoring brain memory, kimchi premium monitor, and healthcheck cron jobs automatically on every boot. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Previously only bootstrap .md files were symlinked. Now symlinks all files and directories (memory/, scripts/, rules_*, etc.) so the bot's brain memory system, cron scripts, and other data are accessible in the workspace. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Doctor --fix was wiping the model config to empty {}. Now sets the
model via `openclaw models set` after doctor completes, ensuring
claude-sonnet-4-5 is always used instead of falling back to the
expensive opus default.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Create web-researcher skill with research.js (Serper Google Search) and study-session.js (autonomous round-robin topic study) - Wire SERPER_API_KEY through worker to container - Register auto-study cron job every 6 hours on gateway startup - Default topics: crypto, AI, tech trends, Korea tech Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Prevents "session file locked" errors after gateway restarts by deleting all .lock files in ~/.openclaw before launching the gateway. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
On startup, kill any other running start-moltbot.sh processes and stop any lingering gateway instances. Prevents duplicate processes from piling up across admin API restarts. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use --name, --every, --session, --message flags instead of positional args. Also skip registration if auto-study cron already exists. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When the gateway restarts, OpenClaw cron jobs are lost. The 5-minute worker cron now checks if crons exist after confirming gateway health and restores them if missing. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Trim prompt-guard/SKILL.md from 571 to 38 lines (~1,300 tokens/msg saved) - Trim cloudflare-browser/SKILL.md from 99 to 30 lines - Trim web-researcher/SKILL.md from 77 to 28 lines - Change auto-study cron from 6h to 12h (50% fewer sessions) - Shorten auto-study cron message - Add IDENTITY.md and update CLAUDE.md personality Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
openclaw doctor --fix (v2026.2.9) no longer auto-enables channel plugins. Explicitly run `openclaw plugins enable` and `openclaw channels add --use-env` for Telegram/Discord/Slack after doctor completes in start-moltbot.sh. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add runCommand() helper that kills processes after use, preventing zombie buildup - Add cleanupExitedProcesses() sweep at start of each cron run - Batch multi-process operations into single bash -c commands (crons, sync, r2) - Reduce cron frequency from */5 to */10 minutes - Trim AGENTS.md context (~50% smaller) to reduce per-conversation token cost - Consolidate study queries (2→1 per topic) and extend interval (12h→24h) - Use Haiku 3 + thinking off for auto-study cron - Pass --token flag for gateway-authenticated CLI commands (cron add/list) - Add gateway.remote.token to openclaw.json config - Skip redundant doctor runs on container restart via sentinel file - Reduce verbose request/WebSocket logging to essentials Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ogging - Add contextPruning (cache-ttl 1h), compaction (safeguard), heartbeat (30m) to openclaw.json — prevents settings loss on restart when doctor is skipped - Reduce subagent maxConcurrent from 8 to 4 - Merge CLAUDE.md + IDENTITY.md into single file (54→27 lines, ~500 tokens saved per conversation) - Remove --verbose flag from gateway for production - Fix outdated "12시간" reference to "24시간" Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds a data prep script that reads JSONL conversation logs, filters noise, and outputs structured text for the agent to summarize. Two crons handle AI processing: daily (Haiku) for conversation summaries and weekly (Sonnet) for cross-memory pattern analysis. Includes cron recovery in crons.ts. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Simplify crons.ts to health-check only (registration stays in start-moltbot.sh) - Unify process identification with shared GATEWAY_COMMANDS/isGatewayProcess() - Update cron models from claude-3-haiku to claude-3-5-haiku-20241022 - Remove redundant 60s R2 sync loop (scheduled rsync in sync.ts handles it) - Add shared constants to config.ts (OPENCLAW_CONFIG_DIR, CLAWD_DIR, CRON_MODELS) - Update CLAUDE.md capabilities (brain-memory, browser control) - Fix pre-existing sync.test.ts failures (align with batched command pattern) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Crons using Haiku were failing with "model not allowed" because only Sonnet was registered in the model allowlist. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- HOT-MEMORY.md: self-updating core memory (~400 tok, always loaded) - warm-memory system: topic-indexed on-demand knowledge retrieval - self-modify skill: safe file modification with allowlist, token limits, automatic backups, changelog, and protected content validation - Skill creation/deprecation: agent can build and archive its own skills - Cron modification: agent can adjust schedules (min 6h guardrail) - Weekly self-reflect cron (Sonnet): replaces brain-insights, adds memory pruning, HOT-MEMORY compression, and cross-memory insights - Compact output mode (--compact) for brain-memory and study-session crons to reduce context size - Compressed SKILL.md files (~60% smaller always-loaded context) - R2 sync for warm-memory and modification-history persistence Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add google-calendar skill with calendar.js (list, create, search, freebusy, update, delete) - Add sync-today.js to auto-sync calendar events to warm-memory/calendar.md every 6h - Add freebusy --emails support to check other people's availability - Pass Google Calendar OAuth env vars through gateway (types.ts, env.ts, env.test.ts) - Add calendar instructions to TOOLS.md injection in start-moltbot.sh - Symlink HOT-MEMORY.md and CLAUDE.md to workspace root for OpenClaw - Fix cron name grep patterns to prevent false matches (brain-memory vs brain-memory-system) - Add google-auth-setup.js helper script for OAuth token setup Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…itoring, R2 fallback - Prevent cron from killing gateway during startup (grace period + 30s timeout) - Add WebSocket reconnection with retry on unexpected container close - Accept WebSocket immediately with status messages instead of 180s block - Parallelize R2 restore and GitHub clone in start-moltbot.sh - Add cron status, uptime, and last sync time to /api/liveness - Add postdeploy script to verify gateway health after deploy - Fall back to R2 binding API when S3FS sync fails Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The ws module installed globally wasn't found by Node scripts because /usr/local/lib/node_modules isn't in the default require search paths. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Headless Chrome script that navigates to any URL, renders JavaScript, and extracts clean article text. Works on SPAs and dynamic sites that plain HTTP fetch can't read. Agent can combine with research.js for search + deep content extraction. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ove flawed gate
Two root causes prevented cron jobs from being registered on startup:
1. The outer conditional gate (if CRON_SCRIPT || SERPER_API_KEY) blocked
brain-memory and self-reflect from registering when they have independent
prerequisites. Removed the gate so each cron checks its own prereqs.
2. nc (netcat) is not installed in the container, and Debian bash doesn't
support /dev/tcp. Replaced nc -z with a Node.js-based port_open() function
using require('net').createConnection(). Added retry helper for robustness.
Also includes external changes: gateway token auth, WebSocket proxy improvements,
auto-approve pairing loop, container reset endpoint.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
--emails), and auto-sync to warm-memory every 6hnc(not installed) with Node.js port check, removed flawed conditional gate that blocked brain-memory and self-reflect registration, added retry logicTest plan
🤖 Generated with Claude Code