## (daily refresh, 1)

- hermes: recorded v0.22.3 and v0.22.4 (both 2026-10-21) and v0.22.5 (2026-10-23, about 81 pull requests since v0.22.4) on the semver release line
- openclaw: recorded v2026.10.6 (2026-10-22, patched-Node requirement responding to the compromised-certificate October Node.js security releases), the v2026.10.10 line at beta.3, and the mass close of 541 issues on 2026-10-22 taking open issues to 8,856
- openwork: recorded v0.18.60 (2026-10-22)
- openwork: pricing churned a seventeenth time, the two add-ons leaving and the grandfather clause returning
- openworker: rotation re-verification, the Windows download marked coming soon on the official site, v0.3.1 still latest
- qwenpaw: rotation re-verification, recorded v2.2.2.post1 (2026-10-22, emSign CA trust remediation, Node 22.23.4) and the QwenPaw-Flash lineup at five sizes (2B, 4B, 9B, 27B, 35B-A3B)
- nanoclaw: v2026.10.1-rc.2 (2026-10-22) on the open release-candidate line
- assistant-runtimes: routine volatile refresh moved drifted issue counts and push dates on anything-llm, astrbot, eigent, nanobot, octop, open-webui, picoclaw, private-gpt, and zeroclaw
- automated-research: autokernel - Added: RightNow AI's MIT autoresearch loop for GPU kernels, keep-or-revert behind a five-stage correctness gate, 1,603 stars as of 2026-10-23.
- automated-research: openresearch - Status refreshed: v0.2.21 (October 23) the latest release, Trackio widened to Spaces, Hugging Face runs, and orx metrics, SSH and Slurm storage root now configurable.
- automated-research: local-deep-research - Status refreshed: main advanced to October 23 on the daily cadence, release train still v1.10.7, open issues 1,228 as of 2026-10-23.
- automated-research: kapso - Status refreshed: October 19-22 repository-memory whole-file fixes and the Weaviate 1.39.13 bump, v0.7.10 still latest.
- automated-research: aide-ml - Harmonized the cautions line's star figure (1,568 versus the status line's 1,576) and recorded the October 13 and 16 pushes in Changes.
- automated-research: aris - October 21-22 pushes checked: docs-only (README numbering, September news archive), no note change.
- automated-research: autocontext - October 21 push checked: feature-branch activity only, main unchanged since 2026-09-29, no note change.
- automated-research: autosci - October 21 push checked: WeChat QR code docs commit, no note change.
- automated-research: deepanalyze - October 13 push checked: README news update already in the status, no note change.
- automated-research: autoresearchclaw - Rotation re-verification: all seven cited URLs live, v0.5.0 still latest, dormant since 2026-08-19, no change.
- automated-research: dexter - Rotation re-verification: all six cited URLs re-fetched (the LICENSE 404 is the recorded evidence of the missing license), unchanged since 2026-09-23, no change.
- automated-research: agent-laboratory - Rotation re-verification: all six cited URLs live, dormant since 2025-08-20 confirmed, no change.
- automated-research: ai-researcher - Rotation re-verification: all six cited URLs live, no license file confirmed, dormant since 2025-10-16, no change.
- automated-research: aiscientist - Rotation re-verification: all five cited URLs live, benchmark tracks confirmed, dormant since 2026-07-21, no change.
- automated-research: matrix extended to thirty-two columns with AutoKernel, OpenResearch surface cell moved to v0.2.21.
- automated-research: index entry list re-sorted with AutoKernel, five star as-of dates aligned to the notes' baselines (figures untouched).
- paperclip: rotation re-verification, release pause at eleven days with the open-issue backlog regrowing again to 6,425
- crowdstrike-falcon-guardian: rotation re-verification, every reference re-fetched, no drift
- databricks-unity-gateway: rotation re-verification, all seven references re-fetched live, no drift
- arifos: README self-measurement moved to 10 of 13 floors pass, failing F1 and F8 (2026-10-21 stamp), the 2026-10-11 independent audit reading unchanged
- cc-safety-net: npm train moved to 2.6.18 (published 2026-10-23), after 2.6.16 (2026-10-19) and 2.6.17 (2026-10-22)
- hol-guard: release train moved to v3.55.1 (2026-10-23, PyPI in step); coverage-contract page fetchable again and still unrenewed fourteen days past expiry, still pinned to 3.0.1
- control-planes: matrix volatile cells refreshed; routine push-date and issue-count refreshes in Agent Control, AgentField, IBM ContextForge, the Microsoft toolkit, Okto Pulse, OpenAPPA, Tracekit, and Vestige
- evaluation-review: Jevals - replaced dead jevbench.xyz source with the moved canonical board at benchmarkheaven.com/jev-models, restated the accuracy caution on live sources, added the Red Hat AI Safety bench as critical source (rotation).
- evaluation-review: HermesGate - hermes-labs-ai org repository now redirects to roli-lpci/hermes-gate, collapsing the twin-repository arrangement; provenance record updated (rotation).
- evaluation-review: FrontierHarness Eval - rotation re-verification clean: all seven cited URLs live, repository unchanged since 2026-09-08.
- evaluation-review: Langfuse - recorded the v4.67.0 and v4.68.0 releases (2026-10-21 and 2026-10-22).
- evaluation-review: Phoenix - recorded the arize-phoenix 20.24.0 and 20.25.0 releases (2026-10-21 and 2026-10-23).
- evaluation-review: Plannotator - recorded the v0.28.25, v0.28.26, and v0.28.27 releases (2026-10-21 to 2026-10-23).
- evaluation-review: ReviewBench - recorded the TraceHound onboarding (merged 2026-10-22, seventeen third-party reviewers).
- evaluation-review: Crit, Hunk, Kiln, Kitaru, PR Lens, Workshop - volatile checks clean: no new releases, pushes are patch-level fixes or non-branch activity, no note changes.
- evaluation-review: evaluation-review - links, front matter, section order, and matrix widths verified across all nineteen notes; membership and sort order unchanged; matrix maturity cells refreshed (Langfuse v4.68.0, Phoenix v20.25.0).
- openrouter: -pro model ids (gpt-5.6-terra-pro, gpt-6-sol-pro, gpt-6.1-sol-pro, gpt-5.6-luna-pro and siblings) judged provider-side serving tiers, OpenAI's own ids with dated snapshots on OpenAI and Azure endpoints at base-identical token prices but pro-mode reasoning that spends far more tokens; recorded and distinguished from the colon routing variants
- openrouter: pricing page re-fetched 2026-10-23, fee state unchanged (Standard 5.5% with $0.80 minimum, Business 8%, BYOK free to $25,000 then 5%)
- litellm: release train, v1.106.2 stable 2026-10-21 with v1.105.3/v1.104.5/v1.103.7 patched the same day, v1.107.0-rc.2 still the train head, PyPI latest 1.106.2, repo pushed 2026-10-23
- localai: v4.12.0 released 2026-10-22 (web UI redesign, streaming voice, native decision routing, image upscaling, Qdrant stores), release notes count 1,896 top-level gallery entries against the site's 1,255 counter, contributors 252 to 256
- ollama: rotation re-verification with every cited source unchanged, plus v0.40.7 (2026-10-18), v0.40.8 (2026-10-21), and v0.40.9 (2026-10-22) recorded, the last adding Phonon speech behind /v1/audio/speech and the d3 decision-model family
- bifrost: pushed 2026-10-23, no releases past the 2026-10-20 component patches
- llm-gateway: pushed 2026-10-22, v1.22.0 (2026-10-19) still latest
- magnitude: pushed 2026-10-23, @magnitudedev/cli@0.2.10 (2026-10-20) still latest
- kimi-code: kimi-k3 aggregator floors unchanged (input Wafer $0.238, output Sail Research $5.50, listing $0.239/$11.936), roster 28 to 29 endpoints
- vercel-ai-gateway: rotation re-verification, all cited sources unchanged (29-agent table, surcharges, BYOK fallback, issue 11280 still closed)
- claude-plans: rotation re-verification, all cited sources unchanged (Pro $17/$20, Max $100/$200, Max API-credits row $100/$200, Fable rules)
- glm-coding-plan: rotation re-verification, all cited sources unchanged (tiers $18/$80/$168, allowances, multipliers, off-peak rule, no promotion listed)
- model-access: rotation quarter covered vercel-ai-gateway, ollama, claude-plans, and glm-coding-plan; no membership change, matrix and category index untouched

## (daily refresh, 2)

- agent-sandbox: v1.0.8 recorded (2026-10-22: MCP command output bounded at a 30,000-character default with new required truncation fields, breaking for strict schema validators, opt-in failed-Pod recovery, GKE Pod Snapshot primitives, Go SDK provisioning parity, sandbox-router now a Kustomization); release count to twenty-eight.
- microsandbox: v0.7.9 recorded (2026-10-22: live port and label modifications on running sandboxes, SDK-visible exit outcomes, JSON network-policy denials by default, exec backpressure, HTTP CONNECT proxy basic auth).
- e2b: SDK train to 2.59.0 (2.57.0 and 2.58.0 on 2026-10-20: tls egress-proxy option, streaming stdin/PTY handles; 2.59.0 on 2026-10-22: per-sandbox DNS allowlists in the egress firewall, apiUrl/sandboxUrl deprecated for next-major removal); PyPI at 2.59.0.
- opensandbox: release-1.2.0-rc.2 recorded (2026-10-21: OSEP-0022 Credential Vault landed end to end for fast sandboxes, egress-held with durable per-subject persistence and sealed pause/resume transfer, user-reachable vault API, pool-pinned MITM CA, port-scoped NetworkRules in the classic profile).
- openshell: v0.1.5 recorded (2026-10-20: push-mode config delivery at about 100 ms, OCSF authentication events, operator-opt-in virtiofs host mounts on MicroVMs, MicroVM shutdown write-flush fix, Kubernetes pod-annotation allowlists, versioned Sandbox Protocol published for independent backends).
- nvx: daily prerelease train current through 2026-10-23; nothing material.
- artifact-fs: 2026-10-21 push is branch activity; main quiet since 2026-08-12, tags end at 1.0.0-rc.10, no releases.
- brood-box: rotation re-verified, every cited URL alive; 2026-10-21/22 activity is dependency bumps, v0.0.25 latest, HN still zero stories.
- cubesandbox: rotation re-verified, every cited URL alive; post-v0.7.3 commit train (E2B-compatible Sandbox.fork across SDKs, node-local restart policy, cursor pagination), v0.7.3 latest, HN threads hold at 7/5/3.
- drop: rotation re-verified, every cited URL alive; commit train through 2026-10-23 with the supervisor process moved into its own user namespace, v0.3.1 latest.
- sandboxing: matrix maturity cells refreshed (agent-sandbox, microsandbox, openshell, e2b, opensandbox), no membership change, stars untouched.
- ax: note created, the local agent-experience graph ingesting six harnesses plus skills, git, and GitHub PRs into DuckDB, with recall, cost routing, and the retro improve loop, $20/seat founding cohort recorded, 121 stars baseline.
- codexmonitor: note created, the dormant MIT Tauri command center for Codex, last commit 2026-03-26 and v0.7.64 the last release, the official-app comparison recorded.
- session-analytics: matrix extended to nineteen columns with ax and CodexMonitor, intro and reading and choosing prose extended, and the agentsview, Memex, Token Monitor, and AgentMeasure cells refreshed.
- agentmeasure: pricing pivot recorded, the $10 and $20 paid diagnostics retired 2026-10-21 for free measurement runs, an open-sourced review method, and the AMSBench benchmark work; rotation re-verified with every cited URL live.
- agents-observe: rotation re-verified, every cited URL live, repository state and release line unchanged.
- agentsview: v0.46.0 and v0.46.1 recorded, opt-in web-chat imports (Claude.ai through a Chrome extension, ChatGPT, Codex web and cloud tasks) joined the sources.
- ccusage: release line moved to v20.0.41 (Antigravity model ID mapping and a --start-of-week weekly option).
- claude-code-agent-monitor: release line moved to v2.4.4 (patch fixes for the events-today day boundary and release tooling).
- coucou: Mac v0.5.3 (Mochi update notices) and Windows 0.5.2 MSI installers, still unsigned.
- memex: v0.31.0 added the Goose session database as the nineteenth engine.
- token-monitor: v0.77.0 recorded and README coverage moved to 46-plus graded tools, the newest rows account-level quota sources.
- tokscale: pushed date refreshed, no new release after v4.20.0.
- fabro: corrected the star-history light source and refreshed volatile lines (pushed 2026-10-22, nightly train v0.393.0-nightly.1); no material change
- fusion: recorded the beta train resuming at v0.78.0-beta.9 (2026-10-23, Pi runtime pinning) with npm latest still 0.77.0 and 2,559 downloads in the month ending 2026-10-18
- har: rotation re-verification found no material change (v1.16.0 still newest, HAR HQ Team $400 unchanged, marketplace at seven plugins)
- loki-mode: recorded the v11.18.0-to-v11.60.17 burst (about forty minors in three days, minutes-apart single-fix releases quoting the review verdict), npm latest 11.60.15 ahead of next 11.60.10, Docker pulls 225,945
- machinist: rotation re-verification recorded 176 commits past v0.4.0, the native-agents default mode (advisory child reviews), the evals directory's issue-lifecycle evaluation, and the opt-in docs-only auto-merge; zero independent coverage holds
- ouroboros: recorded the v0.59.4 through v0.59.7 patches (2026-10-21 to 2026-10-23, sourcing-gated Seed content and context-pack reporting)
- superplane: recorded the RC train at v1.0.0-rc5 (2026-10-23) and component tags runner v0.2.1 and fleet-manager v0.1.5; v0.30.0 still the newest stable platform tag
- zeroshot: recorded v10.16.0 (machine delegations with expiring API keys kept out of run inputs, remote-base preflight) with npm in sync at 10.16.0
- software-factory-feature-matrix: extended Machinist's loop-owner and acceptance cells and recorded the awesome-software-factories 2026-10-22 sweep resolutions in the long tail
- spec-driven-development: Added PDD Kit (itai-gendler/pdd-kit, the standing product-model column) with matrix column in sorted position and index entry [glm-5.3-flash]
- pdd-kit: created from the awesome-spec-driven-development scan, listed there as BusinessLens until the 2026-10-21 rename (a 2026-10-13 sweep had rejected it below the bar under the old name), admitted at 80 stars with the zero-coverage record as the critical angle [glm-5.3-flash]
- spec-kit: recorded the v1.2.3 release of 2026-10-22 (workflow MCP inventory, .presetignore support, Linear Weave v1.0.2) [glm-5.3-flash]
- mycelium: recorded the v0.328.3 through v0.328.5 releases (2026-10-21 to 23, Mistral Vibe page, judgment-stopping diamond skills, canvas-health next-item hook) [glm-5.3-flash]
- superpowers: recorded the v7.0.1 release (2026-10-21 brainstorming retune after the formulaic-questions feedback, design-doc-before-building fix, 20k-character compaction trims, human-written-summary contribution rule) [glm-5.3-flash]
- ccpm: added to task-management after the coordinator handoff resolved, the MIT agent skill whose board is GitHub Issues with parallel agents in one epic worktree (8,414 stars as of 2026-10-23)
- agentrq: recorded v0.13.1 and v0.13.2 (2026-10-21 and 2026-10-23, MCP attachments shipped and the operator's per-user connections view) and refreshed counts
- backlog-md: recorded the v1.55.1 and v1.55.2 security hardening releases (native Git HTTPS proxies temporarily unsupported) and refreshed counts
- beads: recorded the v1.3.2-rc.3 prerelease (2026-10-23) and refreshed counts
- its-a-plan: recorded v1.7.0 (2026-10-22, agent run history, runner queue with transcripts, 60-minute run limits, Traditional Chinese, four high and two medium security advisories fixed) and refreshed counts
- ordewell: recorded v0.10.11 and v0.10.12 (2026-10-22, one settings surface, task_complete held to turn close, completions surviving forced exits) and refreshed counts
- veritas-kanban: corrected the main-branch claim, main active again 2026-10-23 with MCP-priority and hardening fixes, backlog corrected to 23 open issues and pull requests
- kata: refreshed counts (pushed 2026-10-23)
- pad: refreshed counts (pushed 2026-10-23)
- task-master: rotation re-verification with every cited URL re-fetched, repo still quiet since April, Hamster tiers and the MIT claim unchanged, npm eased to 69,470 last month
- plandb: rotation re-verification with every cited URL re-fetched, release gap near six months, agentfield.ai restructured around an unpriced enterprise program, HN no-story claim re-confirmed
- task-management: feature matrix extended from fifteen to sixteen columns with CCPM, release and activity cells refreshed, PRD-provenance reading now splits five ways