n8n-workflow-designlisted
Install: claude install-skill xsefirosus/sefi-agents
# n8n Workflow Design
Signature skill. This body is the Rule block the solutions-architect's output contract
checks; deep node patterns live in `references/node-patterns.md`, read on demand.
User instructions always override this skill.
All factual output follows the anti-hallucination skill: cite or mark UNKNOWN, never guess.
## Rule block (every workflow spec must cover)
1. Trigger inventory: every entry point (webhook, schedule, manual, event) named.
2. Idempotency: a re-delivered trigger must not double-act (dedupe key or upsert).
3. Retry and error branches: each external call has a retry policy and an error path.
4. Secrets handling: credentials via the platform's secret store, never inline.
5. Webhook security: signature or token verification on every inbound webhook.
6. Observability node: one node logs run outcome for diagnosis.
7. Cost-per-run estimate: state it, so the ROI review has a number.
## Standard spec output format
Trigger(s) -> nodes (in order) -> external calls -> error paths -> observability ->
cost-per-run. The solutions-architect fills this shape into `state/automation-<slug>.md`.
## Two predecessor-earned rules
- Model-authored notification text is sent plain, never through a Markdown or HTML parse
mode. Arbitrary model text containing `_` or `*` silently kills a parsed send and drops
the message entirely (observed live on a completion alert).
- Scope note: n8n is for client and deliverable workflows, never a hop inside sefi-agents'
own co