↓ Skip to main content
tomrochette.com
  1. Agents/
  2. Agents log/

Agents log 2026-10-22

Author
Tom Rochette
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / 13 minutes read (~2471 words)
agents 101
Table of Contents

(daily refresh, 1)
#

  • code-review: multi - Rotation re-verification: v1.18.1 and v1.19.0 recorded (OpenCode 2.0 support, agy Antigravity backend, 26 releases); OpenAgentSkill removed its August review post and its audit page now blocks auto-install; dead citation annotated, every cited URL re-fetched.
  • code-review: openqodex - Rotation re-verification: weekly npm downloads fell to 1,918 (week of October 12 to 18) from the 4,533 launch week; v0.13.0 still latest with only a benchmark-tooling PR merged; every cited URL re-fetched.
  • code-review: kodus - Corrected the release record: 2.2.7, web-1.2.8, and mcp-manager-2.1.3 published October 20 after that day’s check had recorded none; push date October 22.
  • code-review: open-code-review - Recorded v1.13.3 (empty comment batches reported, MCP tools/list pagination, CI checkout fix, NixPkgs provider directory).
  • code-review: code-review-feature-matrix - Recorded OpenCodeReview v1.13.3 and refreshed OpenQodex weekly downloads to 1,918; star figures untouched.
  • code-review-graph: rotation re-verify, every cited URL re-fetched, no claim changed, chart markup corrected, forks and open issues refreshed
  • vanna: rotation re-verify, every cited URL re-fetched, README slimmed by the vendor (Tool Memory and agent architecture now docs-only), pricing tiers unchanged, chart markup corrected
  • wrenai: rotation re-verify, every cited URL re-fetched, docs-versus-README source-count discrepancy closed (docs requote stars at 17.8k, drop the count), pricing tiers unchanged, chart markup corrected
  • context7: adoption figure moved to the trailing month (2,640,895 npm downloads to 2026-10-18) after npm’s counter reported zero-day gaps, pricing unchanged
  • graft: language count to 24 with Gleam, npm @trailhq/trail-code at v0.23.1, Trail Brain pricing unchanged
  • graphify: release v0.9.93 recorded (extraction fixes, credential redaction on rationale labels), hosted plans unchanged
  • repowise: PyPI at 0.58.0 (2026-10-20), no claim change
  • entroly: volatile refresh, versions unchanged at 1.0.87
  • headroom: volatile refresh, package unchanged at 0.41.0
  • jevgrep: volatile refresh, no release change
  • repomix: volatile refresh, no release change
  • rtk: volatile refresh, stable release unchanged at v1.0.1 (dev 1.0.2 RCs only)
  • sol-pi: volatile refresh, no release change
  • context-engines: entrant scan carried over, mapped source unchanged since 2026-05-28
  • aeon: rotation re-verify, all cited URLs re-fetched; proof-of-work counters 141 findings and 105 repositories to 143 and 107; repo counters refreshed (293 forks, 1,486 commits, pushed 2026-10-22); skill count 86, single release, token links, and empty HN footprint all hold
  • rivet-agents: rotation re-verify, all cited URLs re-fetched; docs now list Freestyle beside E2B and Daytona in the sandbox adapter (prose updated); pricing tiers unchanged; v0.5.3 still newest; counters refreshed (134 forks, 85 open issues and PRs)
  • trigger-dev: recorded the v4.7.5 and v4.7.6 releases (2026-10-20), version claim moved from v4.7.4; counters refreshed
  • claude-code-action: release train refreshed to v1.0.258 (2026-10-22), Latest badge still the 2025-08-26 announcement; counters refreshed (834 open issues and PRs)
  • claude-code-hooks: open-issue count refreshed to roughly 15.6k; 33-event surface untouched
  • n8n: push and [email protected] release re-verified routine; forks hold at 61.2k
  • airut: pushed date refreshed to 2026-10-22; counters and 0.27.0 release hold
  • github-agentic-workflows: added the github/gh-aw star history chart to the Status section, the last repo-subject note in the category without one
  • hybrid-execution: the Decision Index jumped to edition 0.4 (label 0.4.1, generated 2026-10-22, 226 configurations, vision 55): the Full score is now 60 percent private same-skills plus 40 percent new-domains, the public index shown but uncounted.
  • decision-index: rewrote the definition, methodology, status, cautions, and bottom line around the fully private 0.4.1 score; GLiDE V1.2 first at 71.52, Jev 1.14 forty-first at 60.24, the 1.13 reference forty-eighth at 59.22.
  • jev: recorded the 0.4.1 rebuild (vendor rows to forty-first and forty-eighth behind a shipped GLiDE V1.2 and 52 new entrants) and Ollaya’s twenty-four families.
  • decider: 0.4.1 rows recorded (31B NVFP4 nineteenth at 66.03, pplx-decider v1.2 fourth, a new Flash v1.2 4B row); pushed 2026-10-20.
  • anyjev: the method’s family scored at three 0.4.1 rows for the first time (lm6q seventh 67.57, Tacitus 31B eighteenth 66.47, L0 twenty-seventh 63.65); pushed 2026-10-21.
  • kev: Kev 27B fell to forty-sixth of 226 at 59.56 on the 0.4.1 rebuild; forks 640, pushed 2026-10-21.
  • clm: 0.4.1 rescored CLM-v0.1-8B to 6.49, rank 218 of 226, against Jev 1.14’s 60.24.
  • von: 0.4.1 rescored archived von-1.3 to 11.65, rank 208 of 226.
  • winnow: 0.4.1 rescored Winnow-12B to 46.57 (93rd) and Winnow-E4B to 32.72 (144th).
  • nimble: a Bespoke Nimble v4 Max row entered the 0.4.1 board at seventeenth (66.56); v3 moved to 51.06 (79th) and v2 to 42.06 (112th).
  • strands-decider-2b: the adverse reading moved to 20.72, rank 178 of 226, on the 0.4.1 rebuild; pushed 2026-10-21.
  • h2o-lightning-4b: the 0.4.1 board entered an H2O Lightning 31B sibling row (sixteenth, 66.58); the 4B remains unmeasured there.
  • atomic-agents: recorded v2.14.0 (opentelemetry-instrument launcher) and v2.15.0 (OpenTelemetry GenAI spans on opentelemetry-api alone, Phoenix no longer required), PyPI 2.15.0, pushed 2026-10-21.
  • ollaya: recorded v0.15.0 (Largha family, typed-decisions scoreboard additions, Libraries section) and v0.16.0 (AMD ROCm/HIP on Linux), the library now twenty-four families, pushed 2026-10-22.
  • laya: recorded 0.4.10 through 0.4.13 (a Rust SDK laya-onnx, an OptionStabilityHook, language-detection diagnostics), PyPI 0.4.13.
  • valen: recorded the validated vLLM serving plugin and deployment guides from the 2026-10-22 pushes.
  • baml: the 0.22.3 nightly train continues through the 20261020 build, pushed 2026-10-22, still no stable 0.22.3.
  • guardrails: pushed 2026-10-21 (branch activity, no release, default branch unchanged).
  • cua-s1: host pushes current to 2026-10-22; the cua-s1 component untouched since the 2026-10-20 client security fix.
  • jeeves: rotation re-verification, every reference holds; weights 371 downloads and 7 likes.
  • jevlike: rotation re-verification, every reference holds, repository still frozen at launch day.
  • nanojev: rotation re-verification, every reference holds; Hugging Face rolling counters re-read (weights about 3,600 downloads), weights still license-less.
  • marvin: rotation re-verification, every reference holds, release train still quiet since v3.2.7 (2026-03-04).
  • openai-structured-outputs: rotation re-verification, all six references live, the gpt-6-luna pricing example unchanged at $0.10/$0.50.
  • hybrid-execution-feature-matrix: Decision Index column and board paragraph moved to the 0.4.1 fully private score; maintenance cells moved for eleven push-signal members; Ollaya to 24 families; Laya API surface to 0.4.13.
  • hybrid-execution: instructor and tool-eval-bench volatile-checked with no material change, left untouched.
  • people-and-publications: steve-yegge - fully re-verified in rotation, all cited URLs live, no material change (catalog still 165 essays, homepage still presents Gas Town as current work).
  • people-and-publications: dax-raad - GitHub bio moved from “building SST and @withbumi” to “building opencode”, What-it-is updated.
  • people-and-publications: birgitta-bockeler - memo series count to 33 with “Are Harnesses Worth the Cost?” (Naresh Jain, 2026-10-13); sensors article date corrected to 2026-05-27 per the article’s revision log.
  • people-and-publications: ai-jason - fully re-verified in rotation, RSS feed serves again and still lists the 2026-09-23 video as newest, no change.
  • people-and-publications: ray-amjad - three new uploads through 2026-10-20 (“The S TIER Claude Code Mods”, “Your Prompts Are Too SMALL for Claude Code”, “Karpathy Is Right: Your Attention Is the New Bottleneck”), Jev video retitled “How to Use Jev + Claude Code”, cohort renamed “The Software Factory Engineer” (190 lessons, 2026-10-26 to 2026-11-06).
  • people-and-publications: people-and-publications-feature-matrix - Ray Amjad cadence cell moved to 18 uploads April 9 to October 20, 2026.
  • people-and-publications: people-and-publications - links, front matter, and as-of dates checked across all 33 notes, no type fields, membership and sort order intact, no broken internal links.
  • steve-yegge: reverted the out-of-policy 2026-10-12 star-figure refresh (18.3k, no Changes bullet); prose restored to the 18.1k as of 2026-09-18 baseline, correction bullet added (coordinator)
  • queue: added a ? for tom: bullet on the TypeChat npm-trigger interpretation (pre-revival npm line vs revival-era publication)

(daily refresh, 2)
#

  • a2ui: rotation re-verification, every cited URL live, spec table unchanged (v0.9.1 current, v1.0 candidate), no releases newer than the 2026-10-17 Python wave.
  • pic-standard: rotation re-verification, every cited URL live, v0.9.1 train and 91-vector corpus unchanged, repository pushed 2026-10-21.
  • uhp: implementing product’s release train reached v0.35.15 (2026-10-22) with v0.36.0 release candidates, conformance suite 2026.10.19.post3 at the latest tag, spec unchanged at its 2026-10-19 fifth version.
  • langchain-agent-protocol: streaming bindings released as langchain-protocol 0.0.20 (2026-10-20), OpenAPI document still 0.1.6.
  • webmcp: draft report revised 2026-10-20 (agent-ecosystem-risks discussion and a new editor, no API surface change).
  • ag-ui: release train moved to release/2026-10-21; SDKs stay at 1.0.3 and the npm download window is unchanged.
  • protocols-feature-matrix: WebMCP maturity cell moved to the 2026-10-20 draft and the lagging AG-UI maturity cell caught up to packages 1.0.3 (2026-10-16).
  • graphrag: rotation re-verification re-fetched every cited URL live (maintenance-mode warning, paper, docs, launch thread unchanged); pushed 2026-10-20, 56 open issues, PyPI 3.3.0 still latest at 52 releases
  • langchain: rotation re-verification re-fetched every cited URL live (dcode docs, retrieval architectures page, Octomind thread, LangSmith pricing unchanged); 17,048 commits, pushed 2026-10-22, splitters catalog still without an AST chunker
  • chonkie: pushed 2026-10-20; PyPI 1.7.0 and 1,255,276 downloads/month unchanged
  • code-graph-rag: pushed 2026-10-22, open issues fell 590 to 543; PyPI 0.2.10 still latest
  • docling: PyPI 2.141.0 (2026-10-20, 229 releases); pushed 2026-10-22, 921 open issues; monthly downloads unchanged
  • lightrag: pushed 2026-10-20, open issues 393 to 397 aligned across sections; PyPI 1.5.7 still latest
  • llamaindex: pushed 2026-10-21, 1,020 open issues counting pull requests
  • open-codebase-index: npm train at 0.37.1 (2026-10-20) and 0.37.2 (2026-10-21), 35 versions; pushed 2026-10-22; monthly downloads unchanged
  • pageindex: pushed 2026-10-22, 150 open issues; SDK train paused at 0.4.4
  • ragflow: advisory record grew six to eight (third 2026 critical with no patched version, CVE-2026-45312 prompt-generator SSTI, plus the GHSA-c6q8-23rp-32cv private-agent access bypass); pushed 2026-10-22, v1.0.0-rc2 still latest, the ZeroPath advisory still absent from the global database
  • unstructured: PyPI 0.27.33 (2026-10-22, via 0.27.32 of 2026-10-20, 0.27.31 skipped); pushed 2026-10-22
  • retrieval-feature-matrix: maintenance row refreshed to 2026-10-22 and the RAGFlow displacement cell and reading line aligned to three 2026 critical RCEs; no membership change
  • retrieval: link, front-matter, and banned-term checks clean across all 16 notes, the matrix, and the index; no membership change, no category-index change
  • cate: rotation re-verification, all cited URLs re-fetched live and standing, repository pushed 2026-10-21, no release after v2.0.5
  • jetbrains: rotation re-verification, $8.33/$25.00 ladder with 10 and 35 AI Credits and Junie Lite re-observed unchanged, the AI family entry page now presents itself as JetBrains Air
  • windsurf: rotation re-verification, all cited URLs standing, devin.ai/pricing 404ed to the fetcher again with the 2026-10-12 ladder standing dated
  • codeg: recorded v0.39.0 (October 21, agent-written commit messages, run-on-this-computer for remote-folder commands, codeg-server proxy sign-in) as the new latest release
  • px0: recorded v0.1.19 and v0.1.20 (October 20 to 21, notes carry only changelog links) as the new latest releases
  • whiteboard: recorded v0.3.4 (October 20, survey-repetition and Windows fixes, one contributed) as the new latest release, repository pushed 2026-10-22
  • codenomad: dev train re-tagged v1.0.0-dev from October 20 (stable still v0.20.1), server package at 1,112 downloads for the month ending 2026-10-18
  • agent-native: npm stable crossed to 0.3.45, catalog re-counted at 24 skills unchanged, framework pushes remain app development.
  • gepa: paper accepted to ICLR 2026 as an oral; docs-site adoption roster (Shopify, Google, Microsoft, OpenAI, Dropbox, Pydantic, Comet ML) and the DSPy integration recorded.
  • skillmd: index inched to 1,132,721 skills from 30,620 authors, 1,948 reviewed unchanged; hero label back to “safety-reviewed”.
  • skills-sh: leaderboard top ten reordered (hyperframes, media-use, reddit-automation entered; frontend-design, handoff, setup-matt-pocock-skills dropped out); skillregistry.io dark again, its second outage.
  • snyk-agent-scan: v0.6.9 added VS Code extension MCP-server discovery; the MCP-execution caution corrected for the README’s consent flow.
  • cross-category: - cross-category: rejected perch (asafofer/perch, 22 stars as of 2026-10-22, zero independent coverage) as a notch supervision tail, mission-control dashboards-and-tails family; it surfaces and forwards approvals but runs, isolates, and governs nothing [glm-5.3-flash]
  • cross-category: - cross-category: stars scan found no starred_at after 2026-10-21T06:31:39Z; checkpoint value unchanged, overwrite with 2026-10-21T06:31:39Z or keep as-is [glm-5.3-flash]
  • cross-category: rakazo (elie222/rakazo, 5,152) and openmausbot (milind-soni/openmausbot, 5,398) rejected under the absorbed runtime family, Grok-bot-alternative runtimes, claw-successor kind (coordinator)
  • cross-category: clawbench (tiger-ai-lab/clawbench, 1,072) stays out on the final browser-automation ruling, trackers-kind only if that family ever opens (coordinator)
  • cross-category: automaker (3,228), agent-of-empires (3,373), and the orchestrator-directory’s unmatched tail batch-rejected on the mission-control and AMO thin-tail family lines (coordinator)
  • cross-category: tweakcc (piebald-ai/tweakcc, 2,556, edits Claude Code system prompts and toolsets) held pending q#8 as harness-modification kind (coordinator)
  • cross-category: stars scan found no starred_at after the 2026-10-21T06:31:39Z checkpoint; checkpoint unchanged
  • cross-category: github/gh-aw confirmed as the executions member github-agentic-workflows (worklist subject_repo null, generator backfills from the chart next run)

(recovery of the interrupted 2026-10-22 run, appended 2026-10-23, glm-5.3-flash)
#

The orchestration, memory, harnesses, and section-root workers finished their disk edits but the run died before their reports were collected and committed. Edits verified coherent against each file’s diff and committed; workers’ own reports were never written, so lines below are reconstructed from the diffs.

  • ag2: Recorded v1.2.1 (2026-10-21) and corrected the PyPI claim: ag2 caught back up at 1.2.1, so the PyPI line tracks the GitHub release line again.
  • agent-orchestrator: Recorded v0.14.4 (2026-10-21) as the newest stable from the nightly train; push 2026-10-22.
  • agent-swarm: Recorded the v2.4.3 through v2.4.5 releases (2026-10-20 to 21).
  • agno: Recorded v3.1.7 and v3.1.8 (2026-10-20 to 21), PyPI in sync at 3.1.8; push 2026-10-21.
  • apowerb: Pushed 2026-10-22, PyPI 0.2.95 across 140 releases.
  • gastown: Corrected the default-branch activity line: a 2026-10-20 Go toolchain security commit (distrusting compromised eMudhra CAs) landed after the shutdown; the release line still ends at v1.2.1 (2026-06-06).
  • hcom: Recorded v0.7.31 (2026-10-21), PyPI in sync; push 2026-10-21.
  • mastra: Recorded @mastra/core 1.79.0 (2026-10-21); push 2026-10-22.
  • multica: Recorded v0.6.4 and v0.6.5 (2026-10-20 to 21); push 2026-10-22.
  • oh-my-codex: Recorded the train to v0.21.29 (2026-10-22), npm in sync across 158 published versions.
  • omnara: Recorded v0.2.9 on both the omnarad and cluster lines (2026-10-21).
  • openrig: Recorded v0.6.14 (2026-10-21).
  • orca: Recorded v1.4.237 (2026-10-21) as the latest desktop release.
  • raven: Recorded v0.2.5 (2026-10-20, remote agent execution on registered machines, conditional DAG branches, Windows improvements); push 2026-10-22.
  • ruflo: Recorded the v3.59.x burst of five releases (2026-10-21 to 22).
  • sim: Recorded v0.9.29 through v0.9.31 (2026-10-20 to 22); push 2026-10-22.
  • squad: Recorded v1.1.0 (2026-10-20); push 2026-10-22.
  • t3code: Recorded v0.1.1 (2026-10-21) after v0.1.0 opened Orchestrator V2’s stable line; push 2026-10-22.
  • worktrunk: Recorded v0.81.0 (2026-10-20, interactive foreground hooks, wt merge --branch, OpenCode 2 worktrees routed through the plugin with breaking hook-context changes); push 2026-10-22.
  • mimo-code: Repository counters refreshed (1,431 commits, push 2026-10-22); star baseline untouched.
  • claude-mem: Recorded the v15.0.0 release (Smart memory search now the default recall path, three-step flow, the old one-call lookup demoted to dumb_search) and the v15.0.1 project-scope consistency fix; npm in sync at 15.0.1.
  • deja: Recorded v0.23.0 (incremental index updates, sub-agents as first-class sessions, closed the seam where recalled history could run a command in opencode or end a dsh turn).
  • emulo: The placebo page now advertises a placebo-testing service for other AI features (free for the first three submissions, none bought as of 2026-10-22), beside the unpriced hand-edit offer.
  • mem0: Added the first independent academic comparison naming Mem0 (SWE-ContextBench, arXiv 2602.08316: weakest of four memory-retrieval frameworks on fail-to-pass coding-task tests at the highest per-task cost).
  • memos: Corrected the PyPI reference: MemTensor’s package is MemoryOS (2.0.33); the bare MemOS name belongs to arkohut’s unrelated Pensieve screen-recorder project, so the previously cited 0.37.0 was the wrong package.
  • supermemory: Recorded SWE-ContextBench naming Supermemory best overall of four frameworks on coding-task context reuse (the caution now reads one independent datapoint each way), plus the README’s SMFS product line and expanded plugin list.
  • model-selection-for-coding-tasks: Daily-churn price pass held on 2026-10-22 (kimi-k3 listing and host floors unchanged, gpt-5.6 terra/luna still listed); as-of clauses moved to 2026-10-22.
  • model-provider-feature-matrix: Synced to the held 2026-10-22 price pass, no vendor cell moved.