test: add behavior tests for connectors

[Chemaclass]

TL;DR — how much to trust the test data

These are characterization tests: they lock in the connectors' current observable behaviour rather than verify against live APIs. The fixtures are hand-built inputs typed against each connector's client interfaces, so the tests run with no real connection to Notion, Grain, Granola, Pylon, etc.

Confidence is two-layered:

• High (~95%) that fixtures match the declared types. Every builder is typed (Partial<Recording>, Partial<PylonIssue>, Partial<Note>, NotionNode), so tsc rejects any fixture that drifts from its interface. Values are realistic (ISO timestamps, UUID-shaped ids, Slack message_ts format, Pylon custom_fields value/values split, Granola source enum).
• Moderate that the types themselves match live API responses. The interfaces are hand-authored from docs/observation, not generated from a schema, and nothing here validates a fixture against a real payload. So the suite proves: given a correctly-shaped input, the serialize/parse logic is correct. It cannot catch a shape we mis-modeled, since fixture and type share the same assumption.

Soft spots to know about: notion/properties.test.ts casts property fixtures as never, bypassing type checking (plausible but unprotected); the HTTP/JSON-parse client layer is still ~0% covered.

To close the type-vs-reality gap later (when a real connection is available): capture one sanitized real response per source as a recorded JSON fixture and run it through the client's .json() parse + serializer once.

---

What

Adds happy-path, behavior-focused unit tests for the connectors that previously had no coverage. One commit per connector.

Connector	Coverage
grain	serializeRecording/idOf, keepRecording (GRAIN_AFTER cutoff + shouldSync)
granola	serializeNote (both transcript disclaimer branches), idOf/updatedAtOf
notion	serializePage/sanitizeBody/computePaths + property normalization across all Notion types
pylon	serializeTicket/serializeAccount ordering & links, filename slugs, parse* round-trips
repo	repoName/remoteUrl URL helpers

Approach

• Tests assert on observable output (markdown strings, decisions, round-trips) — no coupling to private helpers or internals.
• Network- and filesystem-free: minimal typed fixtures built from the real client types, explicit timestamps (no Date.now).
• Matches the existing zendesk/web test style (bun:test, inline snapshots, toContain, round-trips).
• repo/sync.ts: exported repoName/remoteUrl and guarded the auto-run with import.meta.main so helpers import cleanly under test (CI invocation unaffected).

Verification

bun test → 103 pass / 0 fail. Lint, typecheck, and prettier all clean.

---

Note

Low Risk
Changes are overwhelmingly new tests; the small sync.ts export/guard is a standard testability pattern with minimal runtime impact when run as the main script.

Overview
Adds behavior-focused unit tests (bun:test, fixtures, inline snapshots, round-trips) for connectors that previously had little or no coverage: Grain (keepRecording cutoff rules; recording markdown/idOf), Granola (note serialization, transcript disclaimers, idOf/updatedAtOf), Notion (paths, body sanitization, page frontmatter, property normalization), Pylon (tickets/accounts, filenames, parsers), and repo URL helpers.

The only production tweak is in repo/sync.ts: repoName and remoteUrl are exported for testing, and syncRepo auto-run is wrapped in import.meta.main so importing the module in tests does not trigger a sync (CLI/CI entry unchanged).

^{Reviewed by Cursor Bugbot for commit 6bf84b7. Bugbot is set up for automated code reviews on this repo. Configure here.}