felipe/vibeflow-n8n
1
0
Fork 0
mirror of https://github.com/domfelipe/vibeflow-n8n.git synced 2026-08-07 13:16:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add outcome contracts for v0.9

Browse source
This commit is contained in:
Felipe Domingues 2026-07-22 13:56:34 -03:00
parent b312f18251
commit bb581ab7f8
23 changed files with 1107 additions and 95 deletions
Download patch file Download diff file Expand all files Collapse all files

12
docs/product-brief.md
View file

@ -6,7 +6,7 @@ Vibeflow is an open-source safety and contract gate for exported n8n workflows,
## Problem
Natural-language workflow generation is now common. The remaining failure is operational: a structurally valid workflow can still leak a credential, answer while disabled, duplicate a side effect, lack a human fallback, retry unsafely, or run forever.
Natural-language workflow generation is now common. The remaining failure is operational: a structurally valid workflow can still leak a credential, answer while disabled, duplicate a side effect, lack a human fallback, retry unsafely, run forever, issue a refund, notify a customer, or destroy data without the required controls.
Existing builders and MCP servers should keep building. Vibeflow checks the result before production.
@ -21,12 +21,14 @@ Existing builders and MCP servers should keep building. Vibeflow checks the resu
Given an exported workflow, produce a reproducible pass/fail report with concrete remediation and no network access.
## Version 0.8 scope
## Version 0.9 scope
- dependency-free Node.js CLI;
- text, JSON, and SARIF output;
- configurable VF000-VF009 policies;
- safe and unsafe fixtures;
- configurable VF000-VF013 policies;
- outcome contracts for money, customer, privileged, and destructive-data actions;
- structural checks for approval, durable audit, idempotency, limits, failure notification, and recovery;
- safe and unsafe support and refund fixtures;
- GitHub Action;
- Codex skill and plugin package.
@ -40,7 +42,7 @@ Given an exported workflow, produce a reproducible pass/fail report with concret
## Differentiation
Vibeflow starts with policies learned from real conversational-agent operations: an off switch before inference, human handoff after low-confidence output, duplicate-event protection, explicit error paths, bounded retries, and trust-boundary hygiene.
Vibeflow starts with policies learned from real operations: an off switch before inference, human handoff after low-confidence output, duplicate-event protection, explicit error paths, bounded retries, trust-boundary hygiene, and declared controls around real-world outcomes. It asks not only whether the workflow is valid JSON or uses a suspicious node, but what it can do when input is messy or a model is wrong.
## Evidence gate