Schedule
Every 4 hours
Model
OpenAI GPT-4o-mini
openrouter/openai/gpt-4o-mini
Approval
Advisory only — no approval required
Pipeline Curator never moves deals or notifies prospects. It writes reports for you and your reps to read.
What it does
Every 4 hours, reads pipeline/deals.yaml. For each open deal computes days_in_stage, days_since_activity, age, and benchmarks against learned stage_norms. Flags real at-risk (over-flagging kills signal). Where useful, runs WHOIS or schema-extract on the company domain to catch red flags. Writes a daily digest of top actions; a Monday forecast splits Commit / Best-case / Pipe with probabilities derived from stage + age + activity, not from CRM fields you can't see.
Tools it can use
filesystem
file_read Read pipeline/deals.yaml file_write Daily digest + Monday forecast reports ollagraph
mcp_ollagraph_intel_whois Stale WHOIS check on company domain mcp_ollagraph_extract_structured Enrich sparse account notes from homepage memory
memory_store Track per-deal assessments + learn stage_norms Workspace
Paths under /var/lib/openfang/workspace/ this Agent reads or writes:
pipeline/reports/pipeline/shared.pipeline.* Starter prompts
Today's digest
At-risk + recommended actions
Run today's digest from pipeline/deals.yaml. Lead with: deals open, total $, at-risk count and $. Then top 3 actions. Then 1 win.
Monday forecast
Commit / Best / Pipe
Run the Monday weekly forecast. Buckets by your close-probability assessment — not by CRM stage field. Justify each bucket placement in 1 line.
Why is this stuck?
Per-deal coaching
Deal_id=42 has been in 'proposal' for 47 days. Run your stage health check, identify the real blocker, and recommend the next-action (specific — not 'follow up').
Coach me on slip risk
End-of-month gut check
It's the 24th of the month. Look at all deals targeted to close this month. Which are real-real, which are real-on-paper, which should I stop counting? Be brutal.