fix(ux): customer copy guides the next action without internal jargon

[seonghobae]

Audit rules applied

1. No internal implementation boundaries in customer-visible text — removed base64, data-URI internals, byte-limit plumbing, and sanitization jargon from rendered titles/errors.
2. Every explanation guides the next action — size errors now name the limit, unreadable/oversized content offers retry-or-alternative, paste overflow says "try less at once", sanitize rejection suggests plain-text paste.

Scope

9 user-visible strings across Toolbar, inlineImagePolicy, converter/base64, Base64Image, SafeClipboard. Host-contract exceptions (headless API errors, programmer invariants) deliberately unchanged per PRD boundary. 6 doc-contract/unit test files updated to shipped truth; targeted vitest + typecheck pass.

---

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[devin-ai-integration]

Devin Review found 1 potential issue.

[seonghobae]

Exact-head CI failure on current Draft head 5b8037d26a7527434b4aa11cbf8c2fdd92cbc059 is deterministic and Inkspan-local; no rerun should be used as evidence without a source/test change.

Evidence from CI run 32923213855, job 98040776141:

• checkout and the explicit exact-checkout guard both bound execution to 5b8037d26a7527434b4aa11cbf8c2fdd92cbc059;
• all 156 test files / 886 tests passed;
• global coverage still failed at 99.98% statements, 99.95% branches, 99.98% lines;
• src/extensions/Base64Image.ts is the only reported executable gap: uncovered line 127 in normalizeImageError, where the current branch uses /* v8 ignore next */ before the non-Error fallback.

Falsified hypothesis: the single-line v8-ignore commit restores the enforced 100% gate. It does not on the exact head above. The smallest causal remedy is not another ignore-shape-only change: add a realistic RED that makes the existing public image insertion/paste path receive a hostile non-Error rejection (for example a primitive rejection from the file/host callback boundary), assert the normalized user-facing Error, then retain the reachable fallback without suppression. GREEN acceptance is focused test + full pnpm coverage at 100% + exact-head CI, followed by refetch of this PR's current head/base and checks. If a realistic public path proves the branch unreachable instead, remove the dead fallback rather than suppressing it.

I am not editing this branch while its dedicated writer is active; this comment records the first causal boundary and executable acceptance criteria on the existing owner PR.

[seonghobae]

Ownership correction for the converter portion of this broad UX Draft: src/converter/** is already owned by canonical Draft #160 (fix/blob-size-preflight-20260811), so #390 should not carry a competing converter implementation/test writer.

The unresolved Devin finding here is valid: the guard accepts the inclusive boundary (bytes <= maxBytes) while #390 says “under”. I reproduced it test-first on the canonical owner instead of patching this branch:

• fix(reliability): preflight converter payload allocations #160 test-only RED: d9982368602d472f15ebd67628c1a080cd0a5d71
• CI RED: run 32929178203, build/test job 98057960241, exact checkout d998236...; the new owner-local contract failed against the predecessor copy.
• fix(reliability): preflight converter payload allocations #160 narrow source repair: 0727ecba1c8e2e53a3424dafb226ffd7af77c7c9, changing only the canonical converter copy to “at or below” and exact bytes/KB/MB guidance while retaining the existing bytes > maxBytes guard.
• The new fix(reliability): preflight converter payload allocations #160 exact-head CI generation is currently non-terminal, so this is not yet claimed as GREEN or merge evidence.

Please reconcile #390 by dropping its src/converter/base64.ts and converter-test overlap and inheriting the eventual canonical #160 result rather than duplicating the fix here. Leave this thread unresolved until #390’s actual diff no longer carries the contradictory converter text; do not treat #160 predecessor/current status as approval for #390.