fix(#12265): fallback-slop sweep (packages/agent) — media-store fast-fail + page-scoped-context reported degrades

[lalalune]

Part of #12182 (fallback-slop sweep). Batch: packages/agent (#12265). Base: develop @ c0c2ddf374.

This is a precision slice of the ~444-suspect-site batch, not a completeness pass: it converts the two highest-confidence data-loss slop sites the issue calls out by name (media-store I/O) and upgrades the page-scoped-context provider's silent degrades to reported degrades, then annotates every justified handler it keeps. The high-risk boot path (runtime/eliza.ts) and the long ?? <lit> tail are intentionally deferred (sitesLeft).

Re-derived suspect counts (packages/agent, this tree)

pattern	count
empty catch	4
promise swallow (.catch(()=>{}))	40
return-default catch	200
?? <lit>	760
|| <lit>	24

Site table — verdicts (2 files touched)

site	pattern	verdict
api/media-store.ts readStoredMediaBytes	try/catch → return null	CONVERT — absence→null kept; real read error → throw ElizaError("MEDIA_STORE_READ_FAILED", {cause}) (exemplar #2)
api/media-store.ts writeStoredMediaFile	try/catch → return false	CONVERT — traversal→false kept; real write error → throw ElizaError("MEDIA_STORE_WRITE_FAILED", {cause})
api/media-store.ts maybeEvict (×3)	empty / log-warn	KEEP J6 best-effort maintenance
api/media-store.ts gcUnreferencedMedia (×2)	empty / log-warn	KEEP J6 best-effort GC
api/media-store.ts listMediaFiles (×2)	empty / log-warn	KEEP J6 scan / J4 empty-list degrade
api/media-store.ts touchOnServe	empty	KEEP J6 LRU hint
api/media-store.ts persistDataUrl	catch → null	KEEP J3 untrusted data-URL sanitize
api/media-store.ts readBackgroundPins	catch → []	KEEP J3 absent/corrupt ledger → no pins
api/media-store.ts pinBackgroundMedia	log-warn	KEEP J6 best-effort pin write
api/media-store.ts resolveMediaFile decode	catch → 400	KEEP J3 malformed path → explicit 400
api/media-store.ts resolveMediaFile stat	catch → 404	KEEP J4 missing asset → explicit 404
api/media-store.ts persistImageThumbnail / ensureThumbnailForStoredFile	catch → null	KEEP J4 optional thumbnail degrade
api/media-store.ts persistAttachmentUrlIfInline	log-warn+return url	KEEP J4 falls back to functional inline URL
api/media-store.ts deleteMediaFile	catch → false	LEFT (boolean-delete judgment call) — sitesLeft
providers/page-scoped-context.ts fetchLocalJson	catch {}	KEEP J4 port failover (exhaustion → caller renders explicit "unavailable"); empty catch removed
providers/page-scoped-context.ts renderBrowserLiveState	catch → null	CONVERT — add runtime.reportError (J4 degrade, now agent-visible)
providers/page-scoped-context.ts renderAutomationsLiveState	catch → null	CONVERT — add runtime.reportError (J4 degrade, now agent-visible)
providers/page-scoped-context.ts provider get	log-only + return EMPTY_RESULT	CONVERT — logger.error → runtime.reportError (J1 boundary)

Per-category tally

• Converted (slop → fast-fail / reported): 5 — 2 media-store throws + 3 page-scoped reportError upgrades.
• Annotated (justified, kept): 17 — J1 ×1, J3 ×3, J4 ×6, J6 ×8 (2 of the J4 also live on converted sites).
• Left (sitesLeft): the rest of the batch's ~444 suspect + 760 ?? <lit> sites, incl. deleteMediaFile and the entire runtime/eliza.ts boot path (high-risk — deferred, not touched).

Exemplar before → after (readStoredMediaBytes)

// BEFORE — EACCES/EIO indistinguishable from "no such media"; a backup silently drops bytes
try {
  return fs.existsSync(filePath) ? fs.readFileSync(filePath) : null;
} catch {
  return null;
}

// AFTER — absence is null; a real read failure throws to the export boundary
if (!fs.existsSync(filePath)) return null;
try {
  return fs.readFileSync(filePath);
} catch (err) {
  // error-policy:J2 context-adding rethrow
  throw new ElizaError(`media read failed for ${fileName}`, {
    code: "MEDIA_STORE_READ_FAILED", cause: err, context: { fileName },
  });
}

Its test (media-store.test.ts, real fs — no mocks): a directory placed where the bytes should be makes readFileSync throw EISDIR; the test asserts an ElizaError with code MEDIA_STORE_READ_FAILED and a preserved cause, while absence still returns null. The write path is proven with a real read-only store dir (chmod 0o555, skipped as root) asserting MEDIA_STORE_WRITE_FAILED. The provider test drives pageScopedContextProvider.get with a throwing getTasks / getRoom and asserts runtime.reportError fires with the right scope while the provider still degrades (brief-only / empty context).

Verification

• bun run --cwd packages/agent test src/api/media-store.test.ts src/providers/page-scoped-context.test.ts src/services/agent-export.media.test.ts src/services/agent-export.roundtrip.test.ts → 70 passed (63 pre-existing green + 7 new error-path).
• bun run audit:error-policy-ratchet → passes: page-scoped-context.ts emptyCatch 1→0, media-store.ts 5→5 (all annotated J6), zero server-console added.
• bun run --cwd packages/agent typecheck → my four files clean. Two pre-existing, unrelated failures remain in untouched packages (plugins/plugin-discord missing @elizaos/plugin-meetings; plugins/plugin-local-inference logger typing) — not introduced here.
• Logger-only preserved; logger import dropped from page-scoped-context (top-level catch now uses runtime.reportError).

Evidence

• Real-fs induced failures are the tests themselves (EISDIR read, read-only-dir write) — no mock stands in for the failing dependency.
• Real-LLM trajectory: N/A — no prompt/model/action behavior changed; the provider's happy-path output is byte-identical, only the error branch now reports. The reported failure reaches the agent via the existing RECENT_ERRORS provider (foundation Fallback slop foundation (#12182): AGENTS.md error policy, ElizaError, runtime.reportError + ERROR_REPORTED event, recent-errors provider, escalation wiring, lint guards #12263), which is unit-tested in core.

Refs #12265

🤖 Generated with Claude Code

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: c61dfe23-48d1-4cf9-89cd-f4116d37853b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fallback/12265-sweep

---

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.}

[lalalune]

Local verification from my side:

• bun run --cwd packages/agent test src/api/media-store.test.ts src/providers/page-scoped-context.test.ts src/services/agent-export.media.test.ts src/services/agent-export.roundtrip.test.ts passed: 4 files, 70 tests.
• bun run audit:error-policy-ratchet passed.
• bunx @biomejs/biome check packages/agent/src/api/media-store.test.ts packages/agent/src/api/media-store.ts packages/agent/src/providers/page-scoped-context.test.ts packages/agent/src/providers/page-scoped-context.ts exited 0, with one fixable info in page-scoped-context.ts for an unnecessary continue.

Not merging from this pass because GitHub checks are still queued/unstable and the new packages/agent/src/providers/page-scoped-context.test.ts starts with // comments instead of the required top-file /** ... */ prose header from the repo comment-cleanup contract.

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.