MIT-licensed reference implementations for the most common AI agent integrations B2B SaaS founders ask Forge to build. Each template ships:
- A working agent skeleton (Python or Node)
- Integration adapter for the named service
- Example prompts (versioned, hashable for AttestProto attestation)
- Test harness with sample data
- README with environment variable list + run instructions
Triage incoming Stripe refund requests against your refund policy. Routes high-confidence approvals straight through, flags policy edge cases for human review. ~150 LOC. Fits Standard tier ($30k) or DIY in 1-2 weeks.
Score inbound HubSpot leads against ICP criteria, enrich with company data, route to the right account executive based on territory + product fit. ~200 LOC. Fits Standard tier ($30k).
Watch a #incidents Slack channel, summarize ongoing incidents in plain English every hour, post executive update + auto-draft incident retrospective at resolution. ~120 LOC. Fits Starter tier ($15k).
Forge's premium proposition is delivery (4 weeks, code-yours, async). The patterns themselves are public — sharing them lowers integration risk for every customer and signals that nothing about Forge's value is locked behind «trade secret» templates.
If you fork these and DIY the build, that's a win. If you fork them and realize you'd rather have someone else handle the 4-week ship — that's why Forge exists.
MIT for all code. CC0 for prompt templates. No warranty, no liability — these are reference impls, not production-grade for your specific stack without adaptation.
Issues + PRs welcome at github.com/Aispotlightlab/forge-templates. Before adding a new template, check existing patterns to avoid duplication. Keep each template under 250 LOC + single integration; complex multi-integration agents are bespoke (i.e. Forge's actual product).