Ask the same source questions in both nodes

[leggetter]

The problem

Source → Get or Create offered Source Type and a raw JSON config field. The trigger, which creates sources in the same project, had labelled fields for every verification scheme. So creating a verified Stripe source from the action node meant hand-writing:

{"auth_type":"STRIPE","auth":{"webhook_secret_key":"whsec_..."}}

Source Config (JSON) is documented as "advanced", but for any platform source it was the only route.

	Trigger	Action node (before)	Action node (after)
Source Type	✅	✅	✅
Verification mode	✅	❌	✅
Platform secret	✅	JSON only	✅
HMAC secret / header / algorithm / encoding	✅	JSON only	✅
API key header + value	✅	JSON only	✅
Basic auth user + password	✅	JSON only	✅

How

The fields move to descriptions/SourceProperties.ts and both nodes spread them in.

n8n has no way to add a displayOptions.show condition to a property from outside it, so the shared factory takes the extra conditions and merges them into each field. The trigger shows these unconditionally; the action node needs them gated on resource and operation.

buildSourceConfig now types its context structurally instead of as IHookFunctions. The trigger reads parameters without an item index and the action node reads them with one, so the action node passes a shim that closes over its index. Narrowing the type rather than changing the signature leaves the twelve existing call sites in the test suite untouched.

Config is omitted from the create body when nothing is configured — config: {} is not the same as absent, and would ask Hookdeck to blank it.

Testing

npm run lint	0 errors, 2 warnings (the accepted icon ones)
npm run build	pass
npm run scan	pass
verify-package-load.mjs	29 checks
node --test test/unit.test.mjs	110/110

The trigger's properties are byte-identical before and after. I built the tree both ways, loaded both node descriptions, and diffed them — the trigger half is unchanged, and the action node gains exactly ten fields and loses none. That was the main regression risk in moving a 157-line block between files.

Verified against the live API, not only the mocks, since the mocks would happily accept a config shape Hookdeck rejects:

Stripe, labelled field only  -> type=STRIPE  authenticated=true   auth.webhook_secret_key set
Generic + HMAC, labelled     -> type=WEBHOOK authenticated=true   all four HMAC fields present
Generic, no verification     -> type=WEBHOOK authenticated=false  no auth

Three new unit tests cover the field path, the JSON escape hatch still winning where they overlap, and config being omitted rather than sent empty.

Notes

No version bump, per CONTRIBUTING.md. This is additive — new optional fields, no renames — so 0.3.0.

This is the first of the two priorities discussed. The second, the clunkiness around Update Existing Source (#14), is deliberately not here. Worth noting the two interact: with this change, a properly typed source can be created up front, so the adopt-an-existing-source-then-reconfigure path gets hit far less often. That should make #14 easier to judge.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XtT85W8PPr2S8RW2WhCtpi