chore(lint): clear the nightly maintenance report's warning backlog (#247)

[mavaali]

Summary

Clears the full lint backlog flagged by the nightly maintenance sweep's baseline report. Fixes #247.

The 2026-07-17 report listed 8 warnings + 9 infos, all mechanical:

• useOptionalChain guard chains (src/curation/lint.ts, src/eval/llm.ts, src/search/coverage.ts, src/tools/staged-actions.ts) — each rewrite verified semantics-preserving (a null/undefined guard followed by a property comparison collapses to ?. with identical truth tables)
• useTemplate string concatenations (src/import/langgraph-store.ts, src/utils/vault-gitignore.ts, two test files)
• useLiteralKeys index accesses (test/curation/coverage.test.ts)
• One unused import (src/tools/edges.ts)
• One non-null assertion (test/utils/config.test.ts — now ?.)
• Two noExplicitAny in tests, hand-fixed: vi.mocked(runBackfill) replaces the as any mock access, and the partial fixture cast goes through unknown

npm run lint now reports zero findings.

Closing #247 with this: the report issue is a rolling accumulator by design — the nightly workflow (.github/workflows/nightly-maintenance.yml) find-or-creates it by title, so after this closes, the next weekday run simply opens a fresh report. The report's other observations (aging PRs #168/#225/#229, pending-release changelog state) are human-triage items the sweep will keep restating.

Test plan

• npm run lint — zero errors, zero warnings, zero infos
• npm run build clean
• Full suites for every touched module pass (write/curation/import/eval/staged-actions/storage/config/gitignore — 350+ tests)
• No behavior change intended; no new tests required (mechanical cleanups under existing coverage)

Invariants

• No behavior change — every optional-chain rewrite preserves the original guard's truth table
• Error handling stays Result; no classes; no throws from tool handlers

🤖 Generated with Claude Code

https://claude.ai/code/session_01CMm7tS4iZ62ofdJB4rFpKN

---

Generated by Claude Code

[claude]

Review: correctness + simplification/efficiency

Went through this line by line against CLAUDE.md's invariants (Result-based error handling, no classes, advisory-only curation, existence-disclosure rules, frontmatter-only metadata, ephemeral SQLite index, git-as-versioning) — nothing here touches any of those; it's a pure mechanical lint cleanup.

Correctness: verified each useOptionalChain rewrite is truth-table-equivalent to its guard clause, specifically checking that no property access happens on a value that could be null/undefined in the new short-circuit path:

• src/curation/lint.ts (tierChange?.before) — safe, tierChange.after is only reached once tierChange is confirmed defined.
• src/eval/llm.ts (r.error?.kind) — safe, same short-circuit shape.
• src/search/coverage.ts (d?.tags.includes(...)) — safe, d.created only reached once d is truthy.
• src/tools/staged-actions.ts (mergeFrom?.length !== 2) — safe, mergeFrom[0]/[1] only reached once length is confirmed 2 (implying non-null).

Other changes checked:

• src/tools/edges.ts: confirmed relative/resolve from node:path are genuinely unused elsewhere in the file before removing the import.
• test/import/index.test.ts: vi.mocked(runBackfill) is valid — runBackfill is mocked via vi.mock(...) at the top of the file, so this isn't a false-cast.
• test/curation/coverage.test.ts / test/storage/index-db.test.ts / src/import/langgraph-store.ts / src/utils/vault-gitignore.ts: template-literal and literal-key rewrites are behavior-neutral.
• test/utils/config.test.ts: gitDir!.startsWith(...) → gitDir?.startsWith(...) changes the failure mode if gitDir were ever undefined (assertion mismatch vs. thrown TypeError), but both still fail the test — not a functional bug.

No test-file-per-tool gap: no tool is added or removed here, so CLAUDE.md's "every tool gets a test file" doesn't come into play.

No correctness bugs found. Nothing to flag as simplification/efficiency either — the diff is already minimal and each hunk is a direct, non-overlapping fix for its lint rule.

No inline comments posted since no confident findings survived review.