Autonomous PM assistant for Telegram → Notion task capture with optional Tavily research enrichment.

PM work often gets buried in chat. Ghost PM converts unstructured conversation into a reliable task system without manual copy/paste. Unlike a human, which isn't always available, Ghost PM collects inputs from each member 24/7, removing the need for a human PM. Additionally due to it being built on Openclaw, you can integrate this with any platform of your choice.

Key benefits:

• Less admin overhead: tasks and dates are captured automatically.
• Better execution: every task requires a due date, so work is time-bound.
• Cleaner tracking: dedupe prevents duplicate entries.
• Faster research: researchable tasks are enriched with summary + links.
• Safer autonomy: asks clarifying questions when required data is missing.

Create this agent in its own workspace (for example ~/.openclaw/workspace-pm) so PM logic, memory, and tools stay isolated from your personal assistant (main).

This separation gives you:

• independent persona + instructions
• separate files/state for PM automation
• safer routing (only PM group goes to PM agent)
• cleaner repo structure for sharing/deploying

• Interprets task-like messages from a PM Telegram group
• Requires a due date (asks follow-up if missing)
• Parses natural dates (tomorrow, in 3 days, next friday, etc.)
• Writes clean task titles + raw message to Notion
• Deduplicates using Dedup Key
• Enriches research tasks via Tavily
• Optionally rewrites research into one coherent paragraph via OpenAI

• skills/notion-pm/notion-client.py — Notion CRUD + schema-aware writes
• skills/notion-pm/pm_handler.py — decision logic, date parsing, pending clarification flow
• skills/notion-pm/run_ingest.py — wrapper entrypoint for inbound messages
• skills/notion-pm/research_client.py — Tavily (+ optional OpenAI summarize)

mkdir -p ~/.openclaw/workspace-pm
openclaw agents add pm --workspace ~/.openclaw/workspace-pm

Then route only your PM Telegram group to this agent (example in openclaw.json.example).

cd ~/.openclaw/workspace-pm
python3 -m venv .venv
source .venv/bin/activate
pip install notion-client

cp .env.example .env

Fill in Notion/Tavily/OpenAI keys in .env.

Open your Notion DB → ... → Connections → add your integration.

1. Go to https://www.notion.so/profile/integrations/internal
2. Create a new Internal integration
3. Copy the API token (ntn_... / secret_...)
4. Put it in .env:

   NOTION_API_KEY=your_notion_token

Create a database (for example: PM Inbox) and include this schema:

Required properties:

• Task Name (Title)
• Message (Rich text)
• Status (Select: New, In Progress, Done)
• Received At (Date)
• Due Date (Date)
• Dedup Key (Rich text)
• Priority (Select: High, Medium, Low)

Optional (recommended for research enrichment):

• Research Summary (Rich text)
• Research Links (Rich text)

In the Notion DB page:

• Click ... (top-right)
• Open Connections
• Add your integration

Without this connection, API calls will fail with access errors.

Open your database page URL and copy the database UUID, then set:

NOTION_DATABASE_ID=your_database_id

python skills/notion-pm/notion-client.py ping

If this succeeds, your Notion token + DB access are configured correctly.

Required properties:

• Task Name (Title)
• Message (Rich text)
• Status (Select: New, In Progress, Done)
• Received At (Date)
• Due Date (Date)
• Dedup Key (Rich text)
• Priority (Select: High, Medium, Low)

Optional properties for research enrichment:

• Research Summary (Rich text)
• Research Links (Rich text preferred)

cd ~/.openclaw/workspace-pm
source .venv/bin/activate

# 1) Notion auth
python skills/notion-pm/notion-client.py ping

# 2) Research client
python skills/notion-pm/research_client.py "best AI tools for PM roadmap planning" --max-results 3

# 3) Ingest task with due date
python skills/notion-pm/run_ingest.py --message "task: ship onboarding copy by tomorrow high" --chat=-100123 --message-id=1 --source=telegram

# 4) Missing due date flow
python skills/notion-pm/run_ingest.py --message "task: research competitor pricing" --chat=-100123 --message-id=2 --source=telegram
python skills/notion-pm/run_ingest.py --message "next friday" --chat=-100123 --message-id=3 --source=telegram

• Set RESEARCH_CONFIDENCE_THRESHOLD to control when research runs.
• If OPENAI_API_KEY is missing, summaries fall back to Tavily output.