fix: Kanban helper syncs only the card's leading link, not trailing date/reference links

[chhoumann]

Summary

Fixes #80 ("Kanban board faulty behavior"). When a Kanban card moves lanes, the
helper updates the linked note's tracked property (e.g. status) to the new lane.

The bug: the helper iterated over every wikilink in the board file and wrote
the card's lane into each linked note. A card line like

- [ ] [[Project A]] @[[2026-06-01]] and see [[Reference Note]]

therefore clobbered the date note (@[[2026-06-01]]) and the reference note with
the lane name - or raised a spurious 'status' not found in <board> notice when
those notes lacked the property (the root of the long-running #45 noise).

A Kanban card is a top-level task whose content begins with its link; only
that leading link identifies the note to sync. The fix scopes updates to that
leading link.

Root cause (restated)

The card's identity was never modeled. updateFilesInBoard looped over
cache.links (the whole file's links) and getTaskHeading reverse-mapped each
link to a lane via a fragile substring search (taskContent.includes(link.original)).
Trailing links and in-text references were indistinguishable from the card itself.

What changed (src/automators/onFileModifyAutomators/kanbanHelper.ts)

• Leading-link-only card detection. A link is a card link only when it is the
  leading link of a top-level - [ ] task line (only the checkbox + whitespace
  precede it). This excludes @[[date]] links, mid-text references, indented
  sub-checklist items, timestamp-prefixed archive entries, and links inside lane
  headings.
• Lanes from the metadata cache's headings, not a substring search - so
  ATX-closed (## Done ## → Done) and setext lane names resolve correctly
  (verified live).
• Cache/content drift guard. Card detection is validated against the
  freshly-read board content (the cached link must still sit at its recorded
  position with its recorded text); on a mismatch the link is skipped rather than
  acted on with stale data.
• Per-card isolation (issue Kanban board faulty behavior #80 fault 1). Each card is processed in its own
  try/catch; an unresolvable link or a throwing property read (e.g. malformed
  YAML in a linked note) no longer aborts syncing of the remaining cards.
• Clearer notice. The missing-property notice now names the linked file and
  board, and only fires for genuine card notes.

Contract note

A card syncs a note only when the card starts with the link (- [ ] [[Note]] …).
This matches the reporter's explicit intent in #80 ("…the [[Thing]] link is
obviously not meant to be updated" for prose cards) and is the only contract that
fixes #80 without silently clobbering incidental links. A purely descriptive card
like - [ ] Read [[Book]] is intentionally not synced.

Validation

Reproduced-then-fixed live in an isolated Obsidian instance (the 80-kanban
worktree vault), driving the real vault modify-event pipeline.

Before the fix (on today's master, post-#119): moving Project A clobbered
2026-06-01 and Reference Note to In Progress.
After the fix: Project A → In Progress; 2026-06-01 and Reference Note
unchanged.

Generalization proven live (not just the reporter's setup):

Scenario	Result
Card under ATX-closed ## Done ##	lane Done (not Done ##)
Card after a card whose note lacks the property	still updated; accurate notice for the bad one
Rapid moves Shipping → Review → Live (within debounce)	final lane Live
Second board tracking a different property (phase)	isolated from the first board
A card whose property read throws	logged; later cards still sync

Commands run:

pnpm run lint    # 0 errors (16 pre-existing warnings in untouched files)
pnpm run build
pnpm run test    # 71 unit tests pass
pnpm run test:e2e  # 9 live e2e tests pass (2 new Kanban + 7 existing)

Tests

• __tests__/kanbanHelper.test.ts: regression coverage for the date/reference
  clobbering, prose-then-link and indented sub-items, heading links,
  ATX-closed/setext lanes, stale-cache drift, the throwing-card loop, and
  multi-lane boards - with a fixture builder asserted against the live cache shape.
• tests/e2e/kanban-helper.test.ts (new): live modify-pipeline regression for the
  no-clobber behavior and the unresolvable-card-doesn't-abort behavior.

Review

A 3-lens adversarial design review ran before coding (folded into the
cache-headings + drift-guard design) and a 3-lens adversarial review ran on the
committed diff. The diff review's one blocking finding - fault 1 still
reachable via a throwing property read - is fixed in commit fae4d5a with live
proof and a regression test.

Out of scope / follow-ups

• The drift guard validates the card link's identity, not a lane heading renamed
  in place while the cache lags (a transient, self-healing state the 5s modify
  debounce makes negligible). Deriving lanes fully from freshly-read content is a
  reasonable follow-up but would drop the cache's free ATX-closed/setext handling.
• Kanban board faulty behavior #80 also asks why Kanban doesn't interoperate with Auto Progress - a separate
  automator concern, not addressed here.

Closes #80.

[github-actions]

🎉 This PR is included in version 1.9.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀