Regression: `costUsageCache` in `src/gateway/server-methods/usage.ts` had no
delete/prune/evict path. The TTL check at L310 only gates stale reads — on a
miss after expiry, `set()` overwrites the same key but never removes stale
keys. `parseDateRange` derives cacheKey from `getTodayStartMs`, so cacheKey
rolls at every UTC 00:00, and additional axes (days / startDate / endDate /
utcOffset) multiply cardinality. The macOS menu polls `usage.cost` every ~45s
with no params, exercising `parseDateRange`'s default branch every day. Over
gateway uptime the map grows monotonically.
Three sibling caches in the same subsystem already implement MAX + FIFO
eviction (resolvedSessionKeyByRunId, TRANSCRIPT_SESSION_KEY_CACHE,
sessionTitleFieldsCache). This change mirrors their pattern:
- `COST_USAGE_CACHE_MAX = 256` (matches RUN_LOOKUP_CACHE_LIMIT and
TRANSCRIPT_SESSION_KEY_CACHE_MAX).
- New `setCostUsageCache(cacheKey, entry)` helper checks size + evicts
`keys().next().value` when adding a new key would exceed the cap.
- The three existing `costUsageCache.set(...)` call sites now route through
the helper. TTL-on-read, in-flight dedup, and overwrite-on-same-key
semantics are preserved.
Adds `src/gateway/server-methods/usage.cost-usage-cache.test.ts` which drives
growth through `__test.loadCostUsageSummaryCached` with 600 distinct
(startMs, endMs) pairs (mirrors day rollover + range switches). Pre-fix the
Map grows to 600; post-fix it plateaus, the last key is retained, and the
first key is evicted (FIFO).
AI-assisted (fully tested). 432 server-methods tests pass, pnpm check +
pnpm build clean.
eb10803691 tightened the reply-run empty-turn gate to only count
baseBodyFinal (strict user body) and to always append the '[User sent
media without caption]' placeholder to any prefix. That broke the Control
UI webchat path: images arrive via opts.images and do not stamp
sessionCtx.MediaPath (by design — see chat.directive-tags.test.ts
assertion that ctx.MediaPath stays undefined on dispatch). For pure-image
webchat turns the gate therefore returned 'I didn't receive any text in
your message', and when a caption was present the placeholder text leaked
into the Control UI user bubble on top of the inbound-context prefix.
Revert the three get-reply-run.ts hunks from eb10803691 back to the stable
2026.4.5 behavior: check baseBodyForPrompt.trim() (which includes the
inbound-context prefix) for the empty-turn gate, and fall back to the
plain '[User sent media without caption]' placeholder only when the whole
prompt body is empty.
Drop the media-only test the same commit added for metadata-only-prefix
bail-out; it encoded the exact behavior this reverts.
Fixes#69358.
Refs #69427.
@clawdbot/lobster/core returns both resumeToken and approvalId when a
workflow step needs approval, but the lobster plugin was dropping
approvalId in three places: normalizeEnvelope, the tool schema, and the
embedded-runner resume branch.
Agents forced to round-trip the ~155-byte base64url resumeToken across
tool calls are one stray truncation away from "Invalid token". The
8-hex approvalId is a disk-indexed alias (~/.lobster/state/approval_*
.json) — stable and escape-safe.
Changes are additive: token-based resume keeps working unchanged,
callers just gain an approvalId path.
