archive

This run's ISO week (38) matches the already-published issue #38 (itself a fallback number, one ahead of its true ISO week, 37), so this issue is numbered #39 — the next integer after the highest number in the archive — rather than colliding again. Spotlight stays with openclaw/openclaw for a third straight issue, again on the strength of a distinct, substantial release: v2026.9.5 (Sep 19) ships Atomic Updates that check the next version before switching over, plugins you can install without restarting the Gateway, shareable conversations, GPT Live for meetings and phone calls, browser pages you can work on alongside your agent, revisitable conversation archives, and guided setup for a specialist agent team. NousResearch/hermes-agent and different-ai/openwork also shipped this week: hermes-agent's v0.21.3 (Sep 14) patches a remote-dashboard session-refresh bug and a duplicate state.db writer-handle leak; openwork's v0.18.48 (Sep 15) firms up live workflow apps and session recovery. Added deanpeters/Product-Manager-Skills (★ 7k) to the repo list — a focused, actively maintained PM skills library for Claude Code and other agents. Star counts across the full list re-verified against the GitHub API at publication.

This run's ISO week (37) matches the already-published issue #37 (itself a fallback number from a previous collision, whose true ISO week was 36), so this issue is numbered #38 — the next integer after the highest number in the archive — rather than colliding again. Spotlight stays with openclaw/openclaw for a second straight issue, but on the strength of a distinct, substantial release: v2026.9.3 (Sep 8) ships a persistent Skill Workshop, a searchable meeting library with full-transcript search and Markdown/JSONL export, and optional team activity reports, alongside safer staged updates and warm prompt-cache reuse across sessions. NousResearch/hermes-agent also shipped this week — v0.21.2 (Sep 11), a patch closing a state.db reliability regression that v0.21.0 introduced (second-writer lock conflicts that could corrupt the session store) — noted in its row. opf/openproject's v17.8.0 is now a week old and no longer carries a "this week" claim. Star counts across the full list re-verified against the GitHub API at publication.

This run's ISO week (36) matches the already-published issue #36 (itself a fallback number from a previous collision, whose true ISO week was 35), so this issue is numbered #37 — the next integer after the highest number in the archive — rather than colliding again. Spotlight moves to openclaw/openclaw for a genuine this-week reason: v2026.9.2 (Sep 5) adds GPT-6 Astra model support plus reliability work — replies now survive Gateway restarts without losing their place in the recovery queue, and long transcripts stop blocking chat and dashboard responsiveness. NousResearch/hermes-agent also shipped this week (v0.21.0, Aug 31 — Bot Mode, a built-in society of named agents with their own avatars in a shared group chat) but had already held the spotlight for two straight issues, so it moves to a regular repo row this time. opf/openproject's v17.8.0 (Sep 2) lets its MCP server create and update work packages directly — noted in its row, but openclaw's release was the bigger story. Star counts across the full list re-verified against the GitHub API at publication.

This run's ISO week (35) matches the already-published issue #35 (itself a fallback number from last week's run, whose true ISO week was 34), so this issue is numbered #36 — the next integer after the highest number in the archive — rather than colliding again. Spotlight stays with NousResearch/hermes-agent for a third straight week: v0.20.6 (Aug 27) is a real, substantial tag — ~525 merged PRs rolled up, including a 50+ server remote MCP catalog and opt-in OS-keychain encryption for stored secrets — not a rubber-stamp repeat. openclaw/openclaw shipped 2026.9.1-beta.1 this week too, but its changes are restart/config reliability work with no PM-facing story. Added opf/openproject (16k stars, verified via the GitHub API) — a mature, actively-maintained open-source Jira alternative with agile planning, Gantt charts, and time tracking, clearing the >1k bar by a wide margin. Considered unopim/unopim (10.9k stars) but it's a Product Information Management / DAM platform for e-commerce catalogs, not product management — different "product," skipped. Star counts across the full list re-verified against the GitHub API at publication.

This run's date-based idempotency check correctly found last week's data-updated (Aug 17) older than this run's target Sunday (Aug 23) and proceeded to publish — but the ISO week number of today's run date is 34, identical to the already-published issue #34. Per the fallback rule, this issue is numbered #35 (next integer after the highest number in this archive) instead of colliding with #34; "wk 34" in the date line above reflects the true ISO week, not the issue number. Spotlight moved to NousResearch/hermes-agent for a genuine this-week reason — three tagged releases (v0.20.3 through v0.20.5, Aug 17–21) shipped real feature work, including a keyless web-search tier and NVIDIA-backed skill install scanning — not for star count, which is a routine ~1% weekly gain. Two repos were added after turning up ahead of the existing list on a fresh GitHub topic search, both verified via the API this run: makeplane/plane (57.5k stars, an actively-shipping open-source Jira/Linear/Monday alternative) and phuryn/pm-skills (25.6k stars, a 100+ skill PM marketplace). phuryn/pm-skills replaces deanpeters/Product-Manager-Skills, which occupies the same niche at roughly a quarter of the stars. Star counts across the full list re-verified against the GitHub API at publication.

The site went a full week without updating. Issue #33 was originally published via a manual test run on Monday Aug 10 — and the issue-numbering scheme (ISO week of the run date) gave it the same number the real Sunday Aug 16 automated run would have computed six days later. The idempotency check correctly, per its own instructions, saw "#33 already published" and exited without publishing anything: not a crash, a numbering collision. Corrected by hand here; the underlying rule is being fixed in CLAUDE.md so a mid-week manual run can no longer block the following Sunday's real one. Star counts verified against the GitHub API at publication. The spotlight moved to openclaw/openclaw for a genuine this-week reason — its 2026.8.1 beta ships secret egress host binding, a real security hardening feature — not for its total star count, which has sat essentially unchanged for weeks.

Star counts verified against the GitHub API at publication. The openclaw/openclaw description was corrected this issue — it was previously described as "the most-starred project on GitHub," which does not hold up on verification (it currently ranks 6th overall); the claim was dropped rather than replaced with an unverified substitute. The spotlight moved to NousResearch/hermes-agent, which shipped v0.20.0 ("Herald Release") this week — real-time streaming voice, an A2A v1.0 agent-to-agent protocol, and signed outbound webhooks. One Recommended-shelf link (cutlefish.substack.com, John Cutler's newsletter) returned a bot-challenge response on every fetch attempt this run rather than a normal 200, so it was dropped from this issue's live page pending re-verification — the URL itself is very likely still live, just unreachable by this job's fetcher today.

Star counts verified against the GitHub API at publication. Several figures carried over from earlier issues were wrong and were corrected in this one. The Signal section below was also corrected same-day: the original version linked articles as old as March while labeling them "this week" — every item now carries a verified publish date within its stated window. The spotlight was re-pointed same-day from openclaw/openclaw (386k stars, but no this-week story) to different-ai/openwork, which shipped v0.18.18 this week; openclaw stays at the top of the repo list.

Recorded as published. Star counts are not retroactively corrected — several were wrong, and the corrections are noted in issue #32. Two links in this issue returned 404 when the archive was built and are marked below; they predate link verification.