Wire up launch catalog: scaffold CLI + remaining T1/T2/T3 adapters (per LAUNCH_CATALOG_SPEC)

[khaliqgant]

Spec: PR #39 (docs/LAUNCH_CATALOG_SPEC.md) — merge first, then this work cites it.

State

Landed on main (14 adapters): github, gitlab, linear, notion, slack, teams, hubspot, salesforce, pipedrive, jira (#38), asana, clickup, intercom, zendesk (#40).

Catalog target: 54. Gap = 40 adapters + tooling.

Uncommitted on the maintainer's local main (not yet a PR): packages/{mailgun,sendgrid,shopify,stripe}/ plus modified workflows/044-tier1-verify.ts and .github/workflows/publish.yml. Either land or discard before starting — don't step on it.

Blocker first: scaffold CLI (§8.1)

Without this, T2 + T3 waves won't fit. Extend @relayfile/adapter-core's existing cli.ts with a scaffold subcommand exposed as a bin:

• In packages/core/package.json, add "bin": { "relayfile-adapter": "./dist/cli.js" }.
• In root package.json, add "gen:adapter": "npm exec --workspace=@relayfile/adapter-core -- relayfile-adapter scaffold" and "catalog:audit": "npm exec --workspace=@relayfile/adapter-core -- relayfile-adapter audit".
• Add templates/adapter/ skeleton mirroring the github package layout (package.json, tsconfig, src/{index,adapter,types,path-mapper,writeback}.ts, src/webhook/verify.ts, mapping.yaml stub, **/tests).
• scaffold <name> --openapi <url> --auth <kind> --webhook-sig <algo:header> [--nango-template <name>] emits the package, generates types.ts from OpenAPI via types-generator.ts, lifts hints from NangoHQ/integration-templates/integrations/<name>/ if the flag is passed, and appends rows to docs/CATALOG.md + docs/MIRAGE_PARITY.md.
• Smoke test: npm run gen:adapter -- foo --openapi https://... produces a package whose npm run build passes.

T1 wave — remaining 7 adapters (spec §5)

Each needs path-mapper + writeback + signature verifier with passing+tampered fixtures.

• discord (§5.3) — Ed25519 over X-Signature-Ed25519 + X-Signature-Timestamp. Interaction webhooks only at launch; gateway sidecar deferred.
• pagerduty (§5.6) — X-PagerDuty-Signature HMAC-SHA256.
• sentry (§5.7) — Sentry-Hook-Signature HMAC-SHA256.
• stripe (§5.8) — Stripe-Signature v1 (timestamp + HMAC-SHA256, anti-replay window). Note: packages/stripe/ is already on the maintainer's working tree uncommitted; reconcile first.
• gmail (§5.9) — Pub/Sub users.watch ingest sidecar.
• google-calendar (§5.10) — events.watch push channels (≤30d expiry, refresher needed).
• google-drive (§5.11) — changes.watch push channels.
• s3 (§5.15) — SQS poller mode (S3 → EventBridge/SNS/SQS → relay).

T2 wave (spec §6)

Generate via scaffold from public OpenAPI; each needs only path-mapper + writeback fixtures.

• bitbucket, vercel, outlook-mail, onedrive
• dropbox, box, posthog, datadog
• gcs, azure-blob, r2, supabase
• confluence, trello, telegram, smtp-imap, ssh
• google-docs, google-sheets

T3 wave (spec §7)

Catalog-only; OpenAPI-driven, read-only, placeholder writeback returning 501.

• freshdesk, shortcut, coda, langfuse
• sharepoint, google-slides, netlify
• postgres, mongodb (synthetic VFS — see §7)
• semantic-scholar, arxiv

Tier reconciliation

Spec §4 vs what shipped:

• pipedrive shipped as T1 in feat(adapters): tier-1 batch 1 — hubspot, salesforce, pipedrive, jira #38 but spec lists it T3 — promote spec or demote adapter.
• clickup, zendesk shipped as T1 in feat(adapters): tier-1 batch 2 — asana, clickup, intercom, zendesk #40 but spec lists them T2 — same call.
• mailgun, sendgrid, shopify (uncommitted on maintainer's local) aren't in the spec at all — add rows to §4 (likely T2) or drop the work.

Catalog audit + parity gate (spec §9)

• Author docs/MIRAGE_PARITY.md from https://docs.mirage.strukto.ai/home/resource-matrix (32 rows) so the catalog audit has something to compare against.
• relayfile-adapter audit subcommand: count packages, cross-check against MIRAGE_PARITY.md, fail CI if relayfile count < Mirage count.
• Wire into the existing turbo CI flow.

Open questions to resolve before starting

These are spec §10 questions — pick before swinging.

1. Which Mirage rows do we deliberately not match? (Proposal: skip Paperclip, OPFS, OCI.)
2. Headline number for marketing splash: 50, 54, or 60 (with stretch rows)?
3. Default provider in quickstart docs: Nango or Pipedream?
4. postgres/mongodb: ship query.json writeback at launch or pure read-only with a follow-up?
5. Discord: launch with interaction-webhook-only and call it T1.5, or block on gateway sidecar?

Sequencing for one-day agent

Mon AM: PR #39 merge → tier reconciliation decisions (spec §10 Q1–5) → scaffold CLI lands.
Mon PM: T1 wave (7 adapters, parallel batches).
Tue AM: T2 wave via scaffold (~2h/adapter).
Tue PM: T3 wave (~30min/adapter once scaffold is solid).
Wed: hygiene — READMEs, parity matrix CI gate, version cut, npm publish --workspaces.

References

• PR docs(catalog): land launch catalog spec #39 — the spec
• PR feat(adapters): ship discoverable schemas for every write endpoint #41 — discoverable write schemas (in flight by codex)
• PR docs: writeback adapter contract spec (and Slack adapter outline) #35 — writeback contract + Slack outline
• Issue Add read-side helpers (GraphQL queries + filter builders + response types) to provider adapters #24 — read-side helpers (related but separate)
• NangoHQ/integration-templates for mapping hints