Weekly digest -- 2026-08-17 (attention layer + Memory & Skills real + 2 new runtimes)

[vivekchand]

Theme of the week

The big shift this week: ClawMetry stopped guessing and started knowing. The needs-you attention layer tells you exactly which agent is waiting on you across all runtimes; Memory & Skills went from a stub to a real GitHub-style file viewer with per-runtime sync; and the trial paywall was armed — the monetization rail is finally closed. Two new runtimes (DeepSeek Harness, Exo) pushed the catalogue to 21, and the approval-delivery seam moved fully behind the cloud license.

---

Shipped (~52 feature/fix/test PRs in OSS; cloud data unavailable — see Bot meta)

OSS (vivekchand/clawmetry)

Features

• feat: needs-you layer — which agent is waiting on you, across all runtimes #4916 + [RELEASE] Needs-you attention layer: which agent is waiting on you (carries #4916) #4922 + feat: one-line runtime wiring for needs-you (clawmetry hook attention) #4924 Needs-you attention layer — which agent is waiting on you, wired across all runtimes with a single clawmetry hook attention one-liner
• Add Exo harness (exo) as the 21st runtime — paid, adapter in clawmetry-pro #4942 Exo harness (exo) — 21st runtime — paid, adapter in clawmetry-pro; catalogue now at 21 runtimes
• Add DeepSeek Harness (dsh) as a paid runtime — OSS wiring #4841 DeepSeek Harness (dsh) — new paid runtime — OSS wiring, adapter in clawmetry-pro
• [RELEASE] Make Memory + Skills real: one runtime picker, files that open, per-runtime cloud sync #4851 + [RELEASE] Skills/Memory: GitHub-style file viewer — raw source, markdown Preview toggle, full screen, copy #4886 Memory & Skills made real — one runtime picker, files that open, per-runtime cloud sync; GitHub-style viewer with raw source, markdown Preview toggle, full-screen, copy
• Session replay: canonical schema + replay_events table (part 1 of #4813) #4828 + feat(transcript): wire _loadReplayTree into viewTranscript (#4814) #4857 Session replay foundation (parts 1+2) — canonical schema, replay_events table, /api/replay-tree endpoint, _loadReplayTree wired into viewTranscript
• Phase 1/4: seam between approvals and approval delivery #4904 + Phase 3+4/4: delete the OSS delivery impl, leave a 402 stub and a boundary guard #4905 Approval delivery → cloud-only — OSS keeps a 402 stub + boundary guard; delivery impl deleted from OSS
• feat(quality): rename Evals to Quality, report card + inline builder #4824 Quality tab (ex-Evals) — renamed Evals → Quality, report-card layout + inline builder
• feat(evals): score-first tab reorder + per-session drill-down drawer #4806 Evals: score-first reorder + per-session drill-down drawer
• feat(sdk): add clawmetry.trace — dependency-free OTLP tracing SDK (#4783) #4804 clawmetry.trace SDK — dependency-free OTLP tracing SDK for AI applications (sdk: ship clawmetry.trace, a dependency-free Python tracing SDK for AI applications #4783)
• feat(tracing): show OTLP traces in the Tracing tab, and put Tracing back in the nav #4789 + feat(otel): OTLP receiver that works out of the box (JSON without protobuf, port 4318, hex ids) #4785 OTLP traces back in the Tracing tab — receiver works out of the box (JSON, port 4318, hex IDs, no protobuf dep)
• Show which Claude surface a session came from + org-wide coverage card #4895 Show which Claude surface a session came from + org-wide coverage card
• Give every alert a destination, and stop shipping rules nothing evaluates #4903 + Re-mirror alert rules stranded by an older daemon #4913 Every alert has a destination — no more rules that nothing evaluates; stranded rules re-mirrored from older daemons
• feat(cli): --turn-off-cloud-sync purges cloud data + kicks daemon #4799 --turn-off-cloud-sync purges cloud data — kicks daemon, nukes traces from the cloud
• Polished Grafana-style date/time range picker on Activity & Sessions #4771 Grafana-style date/time-range picker on Activity & Sessions
• Cost tab: show "$0 out-of-pocket — covered by <plan>" instead of raw API costs #4774 Cost tab: shows "$0 out-of-pocket — covered by <plan>" instead of raw API costs
• feat(entitlement): has_all_bundle + has_all_bundle_at + /api/entitlement/has-all-bundle{,-at} #4769 + feat(entitlement): has_all_bundle_at_path + missing_all_bundle_at_path + /api/entitlement/{has,missing}-all-bundle-at-path #4847 + feat(entitlement): has_all_at_path_batch + missing_all_at_path_batch + /api/entitlement/{has,missing}-all-at-path-batch #4858 + feat(entitlement): has_all_from_path_batch + missing_all_from_path_batch + /api/entitlement/{has,missing}-all-from-path-batch #4878 + feat(entitlement): has_all_bundle_at_path_batch + missing_all_bundle_at_path_batch + /api/entitlement/{has,missing}-all-bundle-at-path-batch #4902 + feat(entitlement): has_all_bundle_batch_at_path + missing_all_bundle_batch_at_path + /api/entitlement/{has,missing}-all-bundle-batch-at-path #4909 + feat(entitlement): has_all_bundle_from_path_batch + missing_all_bundle_from_path_batch #4914 Entitlement API blitz — has_all_bundle*, missing_all_bundle*, at-path and batch variants; full /api/entitlement/* surface expanded
• feat: hook lifecycle install manifest + atomic install/uninstall API (#4817) #4846 Hook lifecycle — install manifest + atomic install/uninstall API (Hook lifecycle contract — install manifest + clean uninstall (runtimes must not error out when clawmetry is removed) #4817 shipped)
• obs(nemoclaw): surface dcodeProxyEnv* signals for dcode proxy-env activation (#4810) #4843 NemoClaw: surface dcodeProxyEnv* signals for dcode sandbox activation
• feat(adapter/openclaw): surface Canvas dashboard pin state in events (#4864) #4876 OpenClaw: Canvas dashboard pin state captured in events ([obs-gap:openclaw] OpenClaw: failed Canvas dashboard pin state not captured #4864)
• feat(interceptor): capture Fish Audio TTS calls as tts_call events (#4749) #4768 Fish Audio TTS intercepted as tts_call events ([obs-gap:openclaw] openclaw: Fish Audio speech synthesis usage/cost untracked #4749)
• feat(sessions): surface keptLocalPaths in workspace.conflict events (#4907) #4932 Sessions: keptLocalPaths surfaced in workspace.conflict events ([obs-gap:openclaw] OpenClaw: cloud workspace conflict transcript events unmapped #4907)
• Derive the advertised runtime count from the entitlement catalogue #4853 Advertised runtime count derived from entitlement catalogue (not hardcoded)

Fixes

• fix: drop ghost sessions from the Sessions list, and stop the false heartbeat SILENT banner #4943 Drop ghost sessions from the Sessions list + stop false heartbeat SILENT banner
• fix(trial): arm the paywall, stop paid-runtime ingest on lapse, remove the pro package #4870 Arm the trial paywall — stops paid-runtime ingest on lapse, removes clawmetry-pro package
• Memory & Skills: fix catalog roots for all 20 runtimes, real project resolution, drop n8n key leak #4938 Memory & Skills: fix catalog roots for all 20 runtimes, real project resolution, drop n8n key leak
• Fix Alerts tab wedged on 'Loading alerts…' once a rule is enabled #4936 Alerts tab wedged on "Loading alerts…" once a rule was enabled — fixed
• fix: ask if the daemon is alive, not when it last heartbeated #4927 Heartbeat banner: ask if daemon is ALIVE, not when it last heartbeated
• fix: only claim "waiting on you" where there is evidence #4933 "Waiting on you": only claim it where there is evidence
• fix: c6-health streak reset after STOP CONDITION MET comment #4935 c6-health: streak resets correctly after STOP CONDITION MET comment
• Home: name the sessions that are alive instead of guessing "idle" #4891 Home: names the sessions that are actually alive, stops guessing "idle"
• fix(paywall): overlay fits the window and never traps the free-runtimes escape #4873 Paywall overlay: fits the window + never traps the free-runtimes escape
• fix: count workspace.conflict events in message_count and JSONL fallback (#4865) #4869 Count workspace.conflict events in message_count and JSONL fallback
• Desktop shell checks PyPI every 60s, and honours that interval on failure #4879 Desktop shell: checks PyPI every 60s and honours failure interval
• [RELEASE] fix: Linux AppImage opened a black window (missing PyGObject overrides) + launch-gate all three desktop builds #4880 Linux AppImage: fixed black window (missing PyGObject overrides); launch-gate on all three desktop builds
• fix(trial): plug 2 paywall bypasses + add "continue free" escape #4832 Trial paywall: plug 2 bypasses + "continue free" escape hatch
• fix(connect): non-interactive callers no longer hang on encryption-key input() (root cause of trial-less cloud signups) #4792 clawmetry connect: non-interactive callers no longer hang on encryption-key input()
• fix(onboarding): cloud sign-in activates 7-day Pro trial (parity with self-host) #4788 Cloud sign-in now activates the 7-day Pro trial (parity with self-host)
• Memory tree: disambiguate generic filenames, don't replace them #4862 Memory tree: disambiguate generic filenames, don't replace them
• fix(sync): add missing typing.Any import and narrow bare except #4908 fix(sync): missing typing.Any import + narrow bare except
• Remove the Developer > Gateway screen #4885 Remove the Developer > Gateway screen (dead UI surface)
• [RELEASE] Paywall overlay fits the window and quotes the price Stripe charges (carries #4873) #4875 Paywall overlay quotes the price Stripe actually charges
• fix(onboarding): dashboard gate honours the desktop shell's onboarding stamp #4763 / fix(onboarding): close both paths (dashboard modal + desktop pane) that re-prompt after successful OTP #4775 / fix(desktop-onboarding): pane records signed_in from identity, not apply_cm_key exit code #4776 Onboarding gate: honours desktop shell stamp, closes both re-prompt paths
• fix(desktop-artifacts): install libgirepository-2.0-dev so Linux job stops blocking every release #4766 Desktop artifacts: libgirepository-2.0-dev installed so Linux job no longer blocks every release

Tests

• test(e2e): golden-path coverage for /api/attention and /api/hooks/attention #4928 e2e golden-path coverage for /api/attention and /api/hooks/attention
• test(e2e): OTLP ingest smoke -- POST /v1/traces to /api/spans round-trip #4796 OTLP ingest smoke — POST /v1/traces to /api/spans round-trip

Docs / chore

• README: link each runtime name to its clawmetry.com runtime guide #4945 README: every runtime name links to its clawmetry.com runtime guide
• docs: weekly roadmap synthesis (2026-08-14) #4820 Weekly roadmap synthesis (2026-08-14)
• docs(windows-signing): activation guide + VERSIONINFO on the tray .exe #4798 Windows signing: activation guide + VERSIONINFO on tray .exe
• i18n sync passes (chore(i18n): sync translations #4825, docs(i18n): sync translated READMEs #4859, chore(i18n): sync translations #4887, chore(i18n): sync translations #4919)

Cloud (vivekchand/clawmetry-cloud)

⚠️ Cloud repo data unavailable this run. GitHub MCP access for this session is scoped to vivekchand/clawmetry only — vivekchand/clawmetry-cloud calls were denied. Cloud PR/issue/commit data could not be fetched.
Fix: add vivekchand/clawmetry-cloud to the GitHub MCP server's allowed-repo list for the weekly-digest session, or grant this scheduled session access to the private repo.
Cloud-side context inferred from OSS: approval delivery seam, entitlement API blitz, paywall rail, and per-runtime cloud sync of Memory & Skills all have corresponding cloud-side counterparts — but those PRs aren't counted below.

---

Biggest user-visible win

#4916 + #4922 + #4924 — Needs-you attention layer

Before this week, if an agent was waiting for your input you had no way to know without opening its transcript. Now ClawMetry shows a dedicated attention surface across all runtimes: which session is paused, why it's waiting, and what runtime it belongs to. Wire it with one command — clawmetry hook attention — and the dashboard pings the moment an agent needs you. For anyone running multiple agents in parallel, this is the first time the tool actually tells you where to look.

---

What users asked for (intel)

No intel-feedback or intel-pain issues were filed this week in the OSS repo. The obs-gap scanner is generating harness-gap issues (see Concerning signal) but there are no user-sourced intel hits to surface.

⚠️ If users are filing feedback in the cloud repo or via other channels, it's invisible to this run — cloud repo is inaccessible (see above).

---

Backlog health (OSS only — cloud unavailable)

• Open issues: 69 OSS (cloud unknown)
• Open PRs: 4 OSS — 1 draft (#4957 feat entitlement batch, opened today), 0 stale (>14 days); oldest non-draft: #4707 (fix desktop uninstall, opened 2026-08-09, 8 days)
• Oldest open issue: feat(v0.13): Native Event Stream Consumer — reference consumer for openclaw/openclaw#6467 #34 feat(v0.13): Native Event Stream Consumer (opened 2026-02-28, ~170 days)
• Median open-issue age: ~45 days (many large backlog epics from May–July dragging the median up)
• Top label counts (OSS):

  Label	Count (est.)
  triaged	~60
  bot-plan-only	~30
  enhancement	~30
  harness-gap + observability	~20
  bot-pr-opened	~8
  weekly-digest	6 (incl. this one)
  bug	1 (#4782 OTLP Tracing tab — open 5 days)

---

Bot productivity (OSS; cloud unavailable)

• Obs-gap / harness-gap bot: 10 new obs-gap issues spawned PRs (bot-pr-opened) and were closed as fixed this week. However, 9 obs-gap issues were closed as duplicates — the dedup logic is misfiring (same gaps for OpenClaw secret egress, Canvas pin state, workspace conflicts, Fish Audio TTS were refiled on Aug 12, Aug 14-15, and again Aug 16-17). This is the biggest bot hygiene problem this week.
• Triage bot: operating normally; all new issues arrive pre-tagged triaged.
• Intel scout: 0 intel issues filed this week (no intel-feedback / intel-pain labels seen).
• Smoke watchdog: no regression issues observed in OSS backlog.
• PR babysit: no <!-- bot-babysit --> marker activity visible.

---

Concerning signal

The obs-gap scanner is re-filing duplicates every 2–3 days. This week, at least 9 issues were filed as duplicates of existing open obs-gap issues — the same signal for OpenClaw secret egress (#4807 → #4906 → #4953), workspace conflicts (#4747 → #4809 → redundant), and Fish Audio TTS (#4724 → #4749 → #4956) keeps appearing. The scanner is not checking existing open issues before filing, which is flooding the backlog with noise and making it harder to see what's genuinely new.

Next step: add a dedup check in the obs-gap scanner — query open issues for the same runtime:* + gap title before filing. Close the oldest duplicates in bulk.

Also worth watching: #4782 (Tracing tab: OTLP apps have no trace to open) has been open 5 days while the OTLP SDK and receiver shipped this week. Now that users can send traces, they'll hit this immediately. It should be the first fix next week.

---

Suggested focus next week

• Fix #4782 — OTLP Tracing tab shows events but no trace tree; SDK + receiver shipped this week so users will hit this on day 1 [OSS]
• Session replay epic — parts 1+2 landed (Session replay: canonical schema + replay_events table (part 1 of #4813) #4828, feat(transcript): wire _loadReplayTree into viewTranscript (#4814) #4857); issues Session replay: canonical schema + replay_events table + /api/replay-tree endpoint + renderer skeleton #4813–Session replay: OpenClaw mapper — ACP replay stream + subagent_runs + flow_runs + approvals audit #4816 (Claude Code mapper, OpenClaw mapper, capture mode + approvals, replay tree renderer) are the next layer [OSS]
• Fix obs-gap scanner dedup — 9 duplicate issues filed this week; add an existence check before filing a new gap ([obs-gap:openclaw] OpenClaw: secret egress host-binding audit events not captured #4807-family, [obs-gap:openclaw] openclaw: cloud workspace conflict transcript events not ingested #4747-family, [obs-gap:openclaw] openclaw: Fish Audio TTS usage/cost untracked #4724-family) [OSS bot]
• Cloud repo access for the weekly digest — add vivekchand/clawmetry-cloud to MCP session scope so next week's digest can report cloud PRs/commits [infra]

---

Code activity

OSS (vivekchand/clawmetry)

• Commits: 49 (past 7 days)
• Contributors: Vivek Chand, github-actions[bot]
• Releases: v0.12.685 → v0.12.726 (+41 patch versions; ~6 named releases per day on busy days)
• Releases published this week: v0.12.721 through v0.12.726 visible in releases API (plus many intermediate patch tags)

Cloud (vivekchand/clawmetry-cloud)

• Commits: unavailable (repo not accessible — see above)
• Contributors: unavailable
• Releases: N/A (proxy: cloud-side approval delivery seam + entitlement API blitz inferred from OSS counterparts)

---

Bot meta

Auto-generated 2026-08-17 07:25 UTC by the weekly digest bot. Tagged triaged so other bots leave it alone.
Cross-repo design as of 2026-05 — but cloud repo (vivekchand/clawmetry-cloud) was inaccessible this run due to MCP session scope. OSS-only data used throughout.