daniel/agent-claw
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
97 Commits 1 Branch 0 Tags
42dbdcde9e3f2150ee17a15207a917fe1616d150
Commit Graph
96 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
daniel 42dbdcde9e feat: factbase-cloud integration — knowledge_agent subagent with M2M auth 2026-05-15 23:32:23 -05:00
daniel ed6577ccf9 feat: billing tags on CDK stack + inference profile creation script 2026-05-15 20:35:02 -05:00
daniel 4f17bbd2c3 fix: intercept [HEARTBEAT] prompt, suppress chatty non-urgent responses 2026-05-15 18:34:14 -05:00
daniel e00702164d refactor: slim system prompt — SOUL.md+STATUS.md only, fix duplicate time injection 2026-05-15 16:42:27 -05:00
daniel 05fee423f2 feat: dynamic subagent loading from SSM 2026-05-15 15:19:08 -05:00
daniel 85efb082f7 fix: unconditional system prompt for call_aws availability 2026-05-15 11:49:03 -05:00
daniel 40f9712c54 fix: remove explicit MCPClient.start() - Strands calls it internally 2026-05-15 11:26:01 -05:00
daniel ebd5a57ece fix: pass aws_service=aws-mcp to aws_iam_streamablehttp_client 2026-05-15 10:32:32 -05:00
daniel 9c09dce519 deps: add mcp-proxy-for-aws to runtime dependencies 2026-05-15 10:28:51 -05:00
daniel 0eff46126f Wire AWS MCP Server via mcp-proxy-for-aws 2026-05-15 10:19:44 -05:00
daniel 266231d070 Add native boto3 AWS tools, remove broken AWS MCP client 2026-05-15 10:03:56 -05:00
daniel 17b1536dae fix: move MCPClient imports inside try block, add TOOLS.md placeholder 2026-05-15 09:29:07 -05:00
daniel add8c6c988 fix: add missing MCPClient/streamablehttp_client imports; fix EXECUTION_ROLE_ARN to actual AgentCore role 2026-05-15 09:14:33 -05:00
daniel 88ed337938 Add AWS MCP Server integration + IAM self-modify with approval gate 
- CDK: add compute/build, broad read-only, IAM self-modify (scoped to own role),
  IAM policy management, and SSM read permissions to runtime1Role
- config.py: load /agent-claw/aws-mcp-url from SSM at cold start
- main.py: connect to AWS MCP Server with SigV4 auth (_AwsMcpSigV4Auth);
  add request_iam_permission and apply_iam_permission tools
- agentcore.json: add EXECUTION_ROLE_ARN env var
 2026-05-15 08:56:06 -05:00
daniel 68aad4fb71 Read model-id from /agent-claw/model-id SSM param and pass to BedrockModel 2026-05-15 07:00:23 -05:00
daniel f31d732cb9 Read model-id from SSM and pass to BedrockModel in main.py 2026-05-15 06:58:54 -05:00
daniel 62862f00f0 Make agent and compaction model IDs configurable via SSM 2026-05-14 18:27:35 -05:00
daniel bdd334b6fb feat: add user-configurable MCP connections 
- manage_mcp_connection tool: add/remove/enable/disable/list MCP servers
- mcp_loader: dynamic connection with OAuth/bearer/none auth, token caching
- Secrets stored in SSM, never in DynamoDB
- MCP clients loaded per-session and cleaned up in finally block
 2026-05-13 21:55:01 -05:00
daniel 74f74ef877 refactor: migrate Secrets Manager secrets to SSM Parameter Store (free tier) 2026-05-13 12:55:16 -05:00
daniel 3a34e61479 feat: add windowed session history + LTM extraction/retrieval 
- New memory_manager.py with:
  - check_and_compact: runs compaction on flagged sessions (extracts LTM via
    Claude Haiku, stores as AgentCore Memory event, deletes old events)
  - check_window_and_flag: sets DynamoDB flag when session > 100 events
  - load_ltm: retrieves LTM extractions and formats as system prompt block
- Wired into main.py:
  - Compaction runs before session_manager creation (trims old events)
  - LTM block injected into system prompt
  - Window check runs after session close
- SESSION_WINDOW_SIZE = 100 (named constant)
- Compaction is idempotent (uses event timestamps as cursor)
- LTM retrieval failure is non-fatal (logs and continues)
 2026-05-13 11:57:50 -05:00
daniel d217842917 refactor: remove MEMORY.md from prompt, add AgentCore memory instructions 2026-05-13 11:48:53 -05:00
daniel 3cc90550b5 feat: add Telegram file attachment support (inbound + outbound) 
Inbound:
- tg-ingest detects document/photo/audio/video/voice attachments
- Downloads files via Telegram Bot API (getFile + download)
- Inlines small text files (<50KB) directly in the prompt
- Stores binary/large files to S3 (attachments/{chat_id}/{update_id}/{filename})
- agent-runner appends file context to the AgentCore prompt

Outbound:
- New send_file tool for the agent to send documents back to users
- TelegramAdapter.send_document uses multipart/form-data POST
- CDK grants tg-ingest S3 write access and passes bucket name env var
 2026-05-13 05:34:33 -05:00
daniel eba4f7db25 fix: align run_code with AWS docs pattern (invoke+executeCode, not execute_code wrapper) 2026-05-12 15:26:01 -05:00
daniel 9253d5046f feat: re-enable code interpreter tool (lazy code_session, no module-level init) 2026-05-12 15:05:26 -05:00
daniel 138f9224c3 fix: use sender from_id as actor_id in groups, not group chat_id 2026-05-12 14:17:02 -05:00
daniel 9d3a93a998 feat: capture message_thread_id for Telegram topic routing 2026-05-12 14:05:00 -05:00
daniel 3a49dadb69 Inject live datetime into system prompt per invocation with relative-time instruction 2026-05-09 14:56:25 -05:00
daniel c317d948b1 Migrate primary to labeled path, remove all flat-path fallback logic 2026-05-09 14:20:24 -05:00
daniel aaecbcfa02 Fix: list_google_accounts also shows flat primary secret 2026-05-09 14:17:53 -05:00
daniel bf89f7255a Fix: always load flat secret as primary regardless of labeled secrets 2026-05-09 14:14:38 -05:00
daniel ac260e4314 Add remove_google_account tool 2026-05-09 13:49:01 -05:00
daniel 6e04d8511c fix: two-step DynamoDB update for google_accounts; live SM lookup in list_google_accounts 2026-05-09 13:32:36 -05:00
daniel 38d828ef74 Multi-account Google support with user labels 2026-05-09 11:21:37 -05:00
daniel 01b258579b Phase 3: proactive heartbeat — EventBridge 30min rule, heartbeat-runner Lambda, HEARTBEAT_OK suppression 2026-05-08 20:14:16 -05:00
daniel eddbd98153 Fix: use build(credentials=creds) instead of creds.authorize() for google-auth compatibility; add traceback logging 2026-05-08 19:57:35 -05:00
daniel 9b56aa83df Fix Google OAuth: explicit IAM policy + strip OIDC scopes from credentials 2026-05-08 16:57:40 -05:00
daniel d68ddab8a2 OAuth callback: send Telegram confirmation message after Google auth 2026-05-08 16:29:05 -05:00
daniel 633ad03db0 Fix naive/aware datetime comparison: strip tz from expiry for google-auth 2026-05-08 16:05:45 -05:00
daniel 8a25eb2d5a Fix: pass expiry to Credentials so auto-refresh fires when token expired 2026-05-08 16:03:09 -05:00
daniel 9d21d5d2e5 Fix: import main in _actor_id() causes app.run() hang — use module-level var instead 2026-05-08 11:32:21 -05:00
daniel 54902cca8d Remove AgentCoreCodeInterpreter import+init: port 8080 conflict blocks event loop 2026-05-08 11:27:32 -05:00
daniel 2f15dd2af3 Remove code_interpreter from base_tools: port 8080 conflict hangs warm containers 2026-05-08 11:23:02 -05:00
daniel f4444cbd22 Fix: pass only authorized http to build(), not credentials, so timeout applies to API calls 2026-05-08 11:17:38 -05:00
daniel 350ce231a4 embed workspace-mcp as direct dependency, simplify google credential loading 
- Add workspace-mcp >= 1.20.0 to pyproject.toml (pulls google-api-python-client etc. transitively)
- Remove redundant google-api-python-client/google-auth/google-auth-httplib2 direct deps
- Rewrite google_workspace.py: single Secrets Manager call per tool (client_id/client_secret
  are already in the credentials secret stored by oauth-handler, no separate oauth-client secret needed)
- Mirror workspace-mcp output format for list_calendars and get_calendar_events
- Add body_format param to get_gmail_message (text/html/raw) matching workspace-mcp API
- Update uv.lock
 2026-05-08 11:12:06 -05:00
daniel 245c2d64f5 Add debug logging to google_workspace tools 2026-05-08 10:56:29 -05:00
daniel 6d0464ea07 Add httplib2 15s timeout + cache_discovery=False to prevent hangs 2026-05-08 10:53:19 -05:00
daniel 25cba295b0 Update uv.lock to include google-api-python-client and deps 2026-05-08 10:49:11 -05:00
daniel ad594f6797 Add direct Google Calendar/Gmail tools, remove workspace_mcp 2026-05-08 10:37:31 -05:00
daniel 943cf26d77 workspace-mcp: strip /workspace prefix for API GW proxy route 2026-05-08 10:27:46 -05:00
daniel 647cb516db Route workspace-mcp through API Gateway to bypass SCP Lambda URL block 2026-05-08 10:24:37 -05:00
daniel eaf19fa9c5 Add debug logging for google_email and workspace_mcp URL 2026-05-08 10:17:08 -05:00
daniel 700e9af2b8 Fix OAUTH_START_URL: use 'or' fallback in case env var is empty string 2026-05-08 09:52:01 -05:00
daniel 9bf6461e1b Disable extended thinking: causes blank responses via streaming retry 2026-05-08 09:44:42 -05:00
daniel f90171cb43 test-bot: use env vars for credentials when available 2026-05-08 09:38:20 -05:00
daniel c3432649c0 Add deploy-agentcore.sh: SSO creds + staging sync before agentcore deploy 2026-05-08 09:31:02 -05:00
daniel b728356fe4 Hardcode OAUTH_START_URL fallback (env var not propagating to runtime) 2026-05-08 09:23:23 -05:00
daniel 4e90440011 Hardcode scheduler Lambda ARN fallback (env var not propagating) 2026-05-07 23:31:11 -05:00
daniel 58ed60f7b7 Add EventBridge scheduling: schedule_reminder, list_reminders, cancel_reminder 2026-05-07 23:24:48 -05:00
daniel 825294d433 Inject current datetime into system prompt on every request 2026-05-07 23:21:05 -05:00
daniel 0a0e26ccd2 Enable extended thinking: budget_tokens=2000 2026-05-07 23:18:48 -05:00
daniel b919a13c76 Fix enrolled_services key mapping in agent-runner payload 2026-05-07 19:29:50 -05:00
daniel ce95cf4c12 Remove send_message @tool def: was causing session-history duplicates 2026-05-07 19:26:23 -05:00
daniel 08ad66a732 Log Telegram API response message_id to find duplicate source 2026-05-07 19:22:40 -05:00
daniel fa74ea784f Remove mid-stream flush on newlines: prevents split multi-turn responses 2026-05-07 19:13:02 -05:00
daniel fd479b8c00 Suppress exceptions in generator to prevent AgentCore retry duplicates 2026-05-07 19:09:46 -05:00
daniel 60573c360f Switch to sync entrypoint + callback delivery: eliminates AgentCore retry duplicates 2026-05-07 19:07:30 -05:00
daniel bbd9a99645 Fix duplicate: remove event.data fallback, only use contentBlockDelta.delta.text 2026-05-07 19:01:36 -05:00
daniel d44fd788f9 Fix broken send_telegram_direct: restore missing data= line 2026-05-07 18:47:16 -05:00
daniel e35599b522 Add stack trace logging to track duplicate send source 2026-05-07 18:43:56 -05:00
daniel b0b641b4c8 Add in-process dedup to prevent AgentCore retry duplicates 2026-05-07 18:38:55 -05:00
daniel 6098f4766a Fix Bedrock read timeout causing retry → duplicate messages 2026-05-07 18:31:44 -05:00
daniel 83b937c20e Remove fallback adapter.send() — streaming consumer handles delivery 2026-05-07 18:17:48 -05:00
daniel 89d0819189 Add logging to streaming path in agent-runner 2026-05-07 18:12:31 -05:00
daniel ae5e0df884 Remove send_message tool: let harness stream text deltas to Telegram 2026-05-07 17:03:34 -05:00
daniel 04c0aeeb8a test-bot: capture send_message tool calls in output 2026-05-07 16:53:15 -05:00
daniel d773985191 Restore send_message in base_tools (tool-based delivery works, streaming is fallback) 2026-05-07 16:52:12 -05:00
daniel 7b7ad578c0 Guard isinstance(event, dict) in SSE parser 2026-05-07 16:47:24 -05:00
daniel beb8dfc969 Fix SSE parsing: read data: prefix + contentBlockDelta.delta.text 2026-05-07 16:45:58 -05:00
daniel cc3b448291 Fix agent-runner: 600s read timeout on bedrock-agentcore streaming 2026-05-07 16:42:49 -05:00
daniel 6adec991da Wire streaming: agent-runner processes chunks, remove send_message tool 2026-05-07 16:32:02 -05:00
daniel 40a942b506 streaming: switch to stream_async + iter_chunks response drain 2026-05-07 16:27:26 -05:00
daniel 7f7f555983 Fix send_message docstring: remove unicode dashes that broke tool spec 2026-05-07 15:15:41 -05:00
daniel b69fdd479a Prompt send_message to fire incrementally instead of buffering 2026-05-07 14:54:21 -05:00
daniel 0951d2be31 Fix workspace bucket fallback + typing error logging 2026-05-07 09:35:09 -05:00
daniel 116d79ead5 Add WORKSPACE_BUCKET_NAME, TELEGRAM_BOT_TOKEN_SECRET_ARN, BRAVE_API_KEY_SECRET_ARN to agentcore env 2026-05-07 09:27:28 -05:00
daniel 92c87222e8 multi-tenant phase 3: per-user Home Assistant + enrolled services 
- tools/home_assistant.py: remove hardcoded URL/token; read from per-user
  config injected via set_ha_config() at invocation time; return helpful
  enrollment prompt when HA not configured
- main.py: inject HA config from user_profile.services at startup; add
  manage_service tool (enroll/remove/list) that persists to DynamoDB;
  show enrolled services in user context; add USERS_TABLE_NAME env var
- agent-runner/handler.py: pass services dict from DDB user record in
  user_profile payload; initialize services={} for new users
- cdk/lib/agent-claw-stack.ts: grant usersTable read/write to runtime1Role
  so manage_service tool can update user records
- agentclaw/agentcore/agentcore.json: add USERS_TABLE_NAME env var
 2026-05-07 09:10:39 -05:00
daniel 4f551ce069 Fix kiro mcp.json: remove invalid factbase server ref 2026-05-07 09:01:43 -05:00
daniel c54e9b1b22 Add kiro config (.kiro steering + settings) 2026-05-07 03:33:02 -05:00
daniel b1056beaa9 Phase 2: wire X-Actor-Id credential loading into workspace-mcp handler.py 
Replace cold-start single-user credential loading with per-request
multi-tenant loading via ASGI middleware:
- _setup_shared_environment(): loads OAuth client creds once at cold start
- _ActorCredentialsMiddleware: reads x-actor-id header per request,
  fetches per-user Google credentials from Secrets Manager
  (agent-claw/google-credentials/{actor_id}), writes to /tmp,
  sets USER_GOOGLE_EMAIL env var
- 5-minute in-memory cache to avoid redundant Secrets Manager calls
 2026-05-06 21:48:05 -05:00
daniel ac5bd78d5a multi-tenant Phase 2: per-user Google OAuth 
- workspace-mcp: add proxy.py (port 8080) that reads X-Actor-Id header,
  fetches per-user Google credentials from Secrets Manager, writes creds
  file, sets USER_GOOGLE_EMAIL, proxies to workspace-mcp on port 8081
- workspace-mcp: update bootstrap to start workspace-mcp on 8081 + proxy on 8080
- workspace-mcp: update Dockerfile to include proxy.py
- oauth-handler Lambda: new Lambda with /oauth/start + /oauth/callback
  routes; exchanges Google auth code, stores tokens in Secrets Manager
  at agent-claw/google-credentials/{actor_id_safe}, updates DynamoDB
- CDK: add OAuthHandler Lambda + GET /oauth/start + /oauth/callback routes
- CDK: remove shared google-workspace-credentials secret; add per-user
  secret IAM grants (agent-claw/google-credentials/*) for workspace-mcp
  role, runtime1 role, and oauth-handler role
- CDK: output OAuthStartUrl + OAuthRedirectUri
- agent-runner: pass google_email in user_profile payload
- main.py: pass actor_id as X-Actor-Id header in workspace-mcp MCP calls;
  skip workspace-mcp if user has no google_email; add connect_google_account
  tool that generates OAuth URL for the current user
- main.py: include google_email in user_context for system prompt
- agentcore.json: add OAUTH_START_URL env var for agent runtime
 2026-05-06 21:42:33 -05:00
daniel 841e729b18 Phase 1 cleanup: onboarding flow, per-user S3 MEMORY.md, seed script 2026-05-06 21:11:07 -05:00
daniel 893c110729 multi-tenant Phase 1: user registry + per-user memory 
- CDK: add agent-claw-users DynamoDB table (actor_id PK, RETAIN policy)
- CDK: grant agent-runner read/write on users table; add USERS_TABLE_NAME env
- CDK: fix cdk.json app field (was object, must be command string)
- CDK: add UsersTableName output
- agent-runner: get_or_create_user() auto-registers users on first contact
  (stores display_name, telegram_username, created_at, allowed)
- agent-runner: pass user_profile in AgentCore payload
- prompt_builder: split base prompt (cached) from per-user context (injected per-call)
  removes USER.md/MEMORY.md from shared load; user name/username injected dynamically
- main.py: extract user_profile from payload, build user_context string for prompt
 2026-05-06 20:36:22 -05:00
daniel 732b00fb66 agent-claw: automated task changes 2026-05-06 18:55:16 -05:00
daniel 38905bb1e9 Phase 0: CDK stack + Lambdas + AgentCore Runtime 1 scaffold 
- CDK TypeScript stack (AgentClawStack):
  - S3 workspace bucket with BucketDeployment seed
  - DynamoDB session-store (actor_id → session_id, TTL)
  - SQS FIFO message queue (serialized per actor)
  - Lambda: tg-ingest (webhook validation, typing action, SQS enqueue)
  - Lambda: agent-runner (SQS → InvokeAgentRuntime, session management)
  - API Gateway HTTP: POST /telegram → tg-ingest
  - AgentCore Runtime 1 IAM execution role
  - CDK outputs: WebhookUrl, WorkspaceBucketName, Runtime1RoleArn

- Runtime 1 (Python + Strands + BedrockAgentCoreApp):
  - main.py: entrypoint, Strands agent, tool wiring
  - channels/: ChannelAdapter Protocol + TelegramAdapter (decoupled)
  - tools/: web_search (Brave), web_fetch, read/write_workspace_file, send_message
  - prompt_builder.py: loads SOUL.md/AGENTS.md/USER.md from S3 (cached)

- Lambdas:
  - tg-ingest: validate X-Telegram-Bot-Api-Secret-Token, send typing, enqueue FIFO
  - agent-runner: session lookup/create in DDB, bundle batched messages, InvokeAgentRuntime

- workspace/: seed files (SOUL.md, AGENTS.md, USER.md, IDENTITY.md, HEARTBEAT.md)

NOTE: AgentCore Runtime 1 creation via CfnResource deferred — deploy CDK first,
create runtime manually with the output Role ARN, then redeploy with runtime1Arn context param.
 2026-05-04 09:00:23 -05:00
daniel 6ee2890831 Add final architecture design doc 2026-05-04 08:43:08 -05:00
daniel 0369a74ac1 Initial research: OpenClaw on AgentCore architecture 
- Architecture comparison (OpenClaw daemon vs AgentCore serverless)
- Component compatibility analysis
- Fargate analysis
- AgentCore rebuild plan (Telegram, zero always-on compute)
- Memory strategy: AgentCore Memory + factbase as structured KB
- Serverless relay patterns per channel
- All open questions resolved
- OpenClaw feature delta March→May 2026
- Build phases and cost estimates
 2026-05-04 08:28:52 -05:00