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

Agents log 2026-10-21

Author
Tom Rochette
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / 17 minutes read (~3253 words)
agents 99
Table of Contents

(maintenance, coordinator tooling)
#

  • Scripted the refresh coordinator mechanics in agents/worklists/: collect_reports.py (worker reports to /tmp/opencode/agent-reports, day-log assembly, last_checked bumps, flag collection), coordinator_check.py (one-call precommit bundle: verify_section, banned terms in the diff, updated-bump audit, tree by scope), make_dispatch.py (slim dispatch files), run_state.py (interrupted-run recovery doctor) [glm-5.3-flash]
  • generate_worklists.py: worklists now carry per-category digest_lines slices of standing-decisions.md so workers consult the digest on demand; fixed parse_old resetting the slug on every non-slug line, which dropped last_pushed and last_release on any no-signals regeneration [glm-5.3-flash]
  • worker-contract.md: report schema moved to the report file with a five-line chat summary; reads reduced to contract, work list (with digest slice), and dispatch file [glm-5.3-flash]

(daily refresh, wave 3 (resumed))
#

  • slop-grader: added from the skills-worker handoff, the category’s second Jev-based member, MIT npm CLI grading documents and diffs against Markdown rulesets, 35 stars and 2,727 npm downloads in its first month, the one-thread HN footprint stated as the caution.
  • better-harness: alpha count corrected, v0.7.0-alpha2 (2026-09-12) shipped on GitHub only and never reached npm, so the npm channel still ends at stable 0.6.6 plus 0.7.0-alpha1.
  • hunk: recorded the v0.24.0 stable release (2026-10-19), ending the v0.23.x stable line.
  • langfuse: recorded the v4.64.0 (2026-10-19) and v4.65.0 and v4.66.0 (both 2026-10-20) releases.
  • plannotator: recorded the v0.28.24 release (2026-10-20).
  • reviewbench: onboardings corrected to sixteen (opswiki, alma-reviewer, and OhMyBug merged 2026-10-19) and the –exclude-producer leave-producer-out judging flag recorded (2026-10-20).
  • workshop: activity line sharpened, no default-branch commit or release since 2026-08-22, the 2026-10-20 activity an outside pull request closed unmerged the day it opened.
  • verbatim-citation-gate: unreleased post-v0.3.0 fixes on main recorded (2026-10-19) and the PyPI absence re-verified as of 2026-10-21.
  • mlr-bench: rotation re-verification, every cited URL live, the zero-open-issues claim corrected to one, a survey-collection invitation filed 2026-10-15.
  • kitaru: rotation re-verification, the kitaru.ai fold into zenml.io recorded (pricing a workspace tab, one subscription across both products, price unchanged), two reference URLs moved to zenml.io.
  • kiln: rotation re-verification, every cited URL live, the evaluations docs reference moved to its new evals-and-specs path, v1.2.0 still newest on GitHub and PyPI.
  • phoenix: activity line refreshed, arize-phoenix 20.23.0 (2026-10-16) still the newest platform release.
  • evaluation-review: matrix extended to nineteen columns with slop-grader, maturity cells refreshed (Langfuse v4.66.0, Hunk v0.24.0 stable).
  • spec-driven-development: archcore - added, the Apache-2.0 typed-document context layer whose review names the side that is wrong (code-wrong or spec-wrong)
  • spec-driven-development: looptroop - admission retracted by the coordinator: looptroop-ai/LoopTroop is already the orchestration member (since 2026-09-27); the same-repository second note and its matrix column, index entry, and prose were removed
  • spec-driven-development: ai-dlc - recorded the stable v2.11.1 release (2026-10-20, worktree approval fix, plugin marketplace model, breaking for plugin scripts)
  • spec-driven-development: ai-unified-process - status re-verified, pushed 2026-10-20, no release change
  • spec-driven-development: arc-kit - status re-verified, pushed 2026-10-20 (docs and site commits), v6.18.2 still newest
  • spec-driven-development: bmad-method - status re-verified, pushed 2026-10-20, v6.12.4 still newest
  • spec-driven-development: cc-sdd - npm 25,174 downloads last month (as of 2026-10-21), otherwise unchanged
  • spec-driven-development: gsd - npm re-read (43,267 and 51,518 last month), pushed 2026-10-21 with unreleased Pi-native packaging work
  • spec-driven-development: intent-router - status re-verified, pushed 2026-10-20 (install-command test gating), no release
  • spec-driven-development: mycelium - recorded v0.328.1 and v0.328.2 (2026-10-20), the OpenCode 2.x plugin port now guards write as well as edit
  • spec-driven-development: openspec - corrected the v1.14.3 symlink claim (the changelog reframe says symlinked layouts are tolerated, not officially supported), npm 2,833,610 last month
  • spec-driven-development: spec-kit - recorded the v1.2.2 release (2026-10-19)
  • spec-driven-development: spec-kitty - PyPI 3,251 last month byte-identical on a third read, v4.0.0rc7 still newest, pushed 2026-10-21
  • spec-driven-development: superpowers - status re-verified, pushed 2026-10-20 (template-process commit), v7.0.0 still newest
  • spec-driven-development: spec-driven-develop - rotation re-verification of every cited URL, still quiet since 2026-07-26, HN still empty
  • spec-driven-development: frink - logged out-of-scope here, cross-category candidate routed to orchestration
  • task-management: agentrq - Pricing section and Status sentence corrected out of rotation to the v0.13.0 dual licence (AGPL-3.0-only repository plus a binary-only commercial licence from Contextual, Inc., agentrqd and agentrq-ws Apache-2.0), re-verified against the README.
  • task-management: agent-kanban - Rotation re-verification: every cited URL live, nothing material changed, forks corrected to 43, v1.13.4 still the newest tag.
  • task-management: kanban-md - Rotation re-verification: every cited URL live, unchanged, v0.39.0 still the newest tag.
  • task-management: veritas-kanban - Rotation re-verification: every cited URL live, v6.2.1 still the newest tag, main branch quiet since its release commit while side-branch pushes continue, backlog corrected to 48.
  • task-management: backlog-md - Recorded v1.55.0 (2026-10-20: JSON draft lists, launcher signal forwarding, frontmatter preservation, UTC timestamps, duplicate-ID repair) and refreshed counts.
  • task-management: beads - Counts refreshed (1,519 open issues, pushed 2026-10-21), v1.3.2-rc.2 still the newest tag.
  • task-management: epiq - Recorded v1.13.1 and v1.13.2 (2026-10-19, TUI memory-leak fix) and v1.14.0 (2026-10-20: board and lane descriptions holding agent conventions handed out over MCP, board flow stats, Epiq-Ref commit-body linking) and refreshed counts.
  • task-management: its-a-plan - Recorded v1.6.0 (2026-10-20: signed invocation-URL agent endpoint needing no polling runner, expiring invites, allowed email domains, Microsoft 365 OAuth mail, MCP client-visibility surface, breaking workspace and OAuth changes) and refreshed counts.
  • task-management: kata - Counts refreshed (pushed 2026-10-21), v0.19.0 still the newest tag.
  • task-management: operon - Recorded 4.1.0 (2026-10-20: desktop lightbox, nested previews, source chips, vault-local ICS calendars) and refreshed counts.
  • task-management: ordewell - Recorded the v0.10.4 to v0.10.10 line (2026-10-17 to 10-19: saved sessions load idle, planner restarts between turns, unified context-and-cost usage line, runner stopped at task_complete) and refreshed counts.
  • task-management: pad - Recorded the v0.18.1 patch (2026-10-18: Stripe billing portal, export deadline, refused MCP calls naming their cause) and refreshed counts.
  • task-management: matrix AgentRQ license and pricing cells corrected to the v0.13.0 dual licence and release and activity cells refreshed for ten members.
  • task-management: no credible dedicated awesome-list for task management surfaced; the broad sources still cover the category.
  • control-planes: tracekit - note created from the awesome-ai-governance scan (added to the source’s Agent Action Records section 2026-10-19): the MIT Certificate-Transparency-style signer that decides each tool call before it runs and exports witness-cosigned .tkb evidence.
  • control-planes: tracekit - admitted above the bar (138 stars, 9 forks as of 2026-10-21, five releases 2026-10-16 to 2026-10-20); matrix column added, twenty to twenty-one, category index updated.
  • control-planes: agent-control - full rotation re-verification, every cited URL live; train moved to v8.14.0 (2026-10-19) with PyPI 8.14.0 and npm 3.4.0, ending the 3.3.0 hold.
  • control-planes: agentfield - train moved to v0.1.143-rc.4 (2026-10-20); PyPI published 0.1.142 (2026-10-19), its first non-rc.
  • control-planes: arifos - self-measurement moved to 7/13 floors measured pass with the kernel self-reporting degraded (2026-10-20 stamp), against the unchanged 2026-10-11 independent audit reading of 2 pass, 4 fail, 7 unmeasured; note now keeps the two witness objects separate.
  • control-planes: cordum - train moved to v1.2.0 (2026-10-17, scheduler lock-fence and audit-exporter fixes), ending the four-month v1.1.0 hold; train-lags caution rewritten.
  • control-planes: hol-guard - train moved to v3.51.0 (2026-10-20, PyPI in step); the coverage-contract page refused fetchers, so its renewal state stands as last verified 2026-10-19.
  • control-planes: ibm-contextforge - full rotation re-verification, every cited URL live; train moved to v1.1.0 (2026-10-20, modern MCP negotiation default, redesigned web UI, federated completions, catalog pagination, Origin/Host transport security), PyPI in step.
  • control-planes: microsoft-agent-governance-toolkit - full rotation re-verification, every cited URL live, no release since v4.1.0.
  • control-planes: okto-pulse - PyPI moved to 0.4.1 (2026-10-20), a patch above the breaking 0.4.0.
  • control-planes: openappa - train moved to v0.37.0 (2026-10-20, two stable releases that day).
  • control-planes: paperclip - scale refreshed in place (6,296 open issues, pushed 2026-10-21, release pause now nine days); backlog caution extended with the 2026-10-21 regrowth.
  • control-planes: vestige - scale refreshed in place (38 open issues and pull requests, pushed 2026-10-20, v5.0.0 still latest).
  • control-planes: links, front matter, as-of dates, matrix membership and sort, and index list verified across the category; no broken links, no missing tags, no stray type fields, all matrix rows at header width.
  • hermes: rotation re-verification, every cited URL re-fetched live, v0.22.2 still latest, Portal tiers and the seven-backend list unchanged
  • eigent: rotation re-verification, every cited URL re-fetched live, v1.0.6 still latest, pricing tiers and the Claude-just-Claude baseline unchanged
  • openwork: v0.18.59 (2026-10-20) recorded; pricing page sixteenth churn, the two add-ons back and the grandfather clause off, three main prices unchanged
  • nanoclaw: v2026.10.1-rc.1 (2026-10-20) opened on the calendar-versioning release line
  • assistant-runtimes-feature-matrix: re-verification moved eight status cells and completed the references with the four missing column repos (AnythingLLM, Octop, Open WebUI, PrivateGPT)
  • assistant-runtimes: volatile refreshes on the pushed members (anything-llm, astrbot, nanobot, octop, open-webui, openclaw, openworker, private-gpt, qwenpaw, zeroclaw), star figures untouched everywhere
  • assistant-runtimes: Orkas-AI/Orkas-VideoStudio resolved out of category, the absorbed-family rejection stands (skills-kind extraction of the absorbed Orkas desktop), routing flagged to skills
  • assistant-runtimes: openhuman stays held pending q13, scale evidence gathered (42.2k stars, above the member floor, self-promoted HN footprint)

(daily refresh, wave 4)
#

  • software-factory: uzi (vtmocanu) reopen trigger still unfired: no stable release (v0.86.0-rc.19 prerelease, 2026-10-19), no HN coverage under “uzi dark factory” or “vtmocanu” (checked 2026-10-21)
  • software-factory: antfarm (snarktank/antfarm) rejected: 2,495 stars, MIT, dormant since 2026-02-26, installs a fixed specialist-agent team into OpenClaw rather than owning a loop; CodeMachine-CLI dormancy precedent
  • software-factory: fabro - pushed date refreshed to 2026-10-20; nightly-only train unchanged, v0.389.0-nightly.0 still the newest tag as of 2026-10-21
  • software-factory: fluent - rotation re-verification: every cited URL re-fetched (repository, five release-notes documents, fluent.computer redirect, Discord invite API, agentskills.io), all claims confirmed, v0.6.0 still newest, no push since 2026-10-11
  • software-factory: loki-mode - train reached v11.18.0 (2026-10-20, no-engine-change checkpoint release, npm latest and next in sync); CI reduced at v11.17.99 to a single release workflow with no test gate in front (Tests, nightly, security-audit, SBOM workflows removed)
  • software-factory: machinist - 132 commits past v0.4.0 as of 2026-10-21, still no release past v0.4.0, pushed 2026-10-20, zero independent coverage re-verified
  • software-factory: ouroboros - train to v0.59.3 (2026-10-20, GitHub and PyPI): v0.58.0 binds only confirmed content to the Seed with a signal-gated verify, v0.59.0 self-answer binds only checked evidence
  • software-factory: superplane - RC train advanced to v1.0.0-rc2 (2026-10-19) and v1.0.0-rc3 (2026-10-20), first CLI component cli/v1.0.0-rc1 (2026-10-19); v0.30.0 remains the newest stable platform tag
  • software-factory: zeroshot - train to v10.15.0 (2026-10-19, npm in sync): Pi fourth harness (v10.14.0), renewable delivery credentials outside agent environments (v10.13.0), README leads with custom workflows and stops advertising Zeroshot Cloud (v10.15.0)
  • software-factory: software-factory-feature-matrix - Zeroshot coding-agent and pricing cells moved for the Pi harness and the Cloud de-advertising; 2026-10-19 awesome-software-factories sweep resolutions recorded; Changes ordering defect fixed
  • software-factory: category fit re-checked, all ten members remain loop-owning factories, no moves or pivots; links, front matter, section order, and matrix table width verified clean
  • agent-analytics: rotation re-verification, every cited URL live, repository (pushed 2026-10-09, LICENSE still absent), npm 0.5.37, docs, pricing shell, and zero-hit HN footprint unchanged.
  • agentmeasure: pushed date to 2026-10-20 on a billing-narrative docs clarification, the $10 and $20 services re-verified unchanged.
  • agentsview: token-usage docs schema caution back to version 5, the sixth recorded flip, with an unreleased gptme parser landing on main.
  • agenttrace: pushed date to 2026-10-19 on side branches, release line still v0.10.1.
  • ccusage: recorded v20.0.37 (2026-10-19, requestless resp_ content blocks counted once), downloads 870,995, npm latest 20.0.37.
  • claude-code-agent-monitor: recorded v2.4.0 (2026-10-20, remote transcript-ingest endpoint, two first-time outside contributors).
  • codeburn: recorded unreleased Claude Code plugin and MCP Registry listing merged 2026-10-20 after v0.9.27.
  • coucou: recorded Mac v0.5.2 (in-notch updates, Homebrew cask, Claude Code plan-approval cards, 14 languages) and Windows/Linux v0.5.1, installer still unsigned.
  • git-ai: rotation re-verification, how-git-ai-works docs repointed to get-started, background-agent installs (Claude Web, Codex Cloud, Cursor Agent, Devin) added, 14-agent table unchanged.
  • memex: recorded v0.30.0 (2026-10-20, no release notes, index refresh decoupled from scan TTL).
  • token-monitor: recorded v0.75.0 and v0.76.0 (2026-10-19 and 10-20, Antigravity subagent grouping and Gemini 4 Argon quota, notch gestures).
  • tokscale: recorded unreleased devin-cli duplicate-heal and submit-integrity fixes of 2026-10-20, downloads 180,721.
  • session-analytics-feature-matrix: schema reading line moved to version 5, no table cells changed.
  • session-analytics: entrant scan of anhermon/awesome-agent-observability found one addition (Aident Loadout), resolved as a cross-category candidate, no membership change.
  • addy-osmani: newest essay is now “Agentic Verification” (2026-10-14), carrying the review-shift argument into agent-run verification loops
  • ai-jason: refreshed as-of figures, latest upload still 2026-09-23 (uploads tab), RSS 404 again, third-party count 103 videos and 234K subscribers
  • aris: rotation re-verification counted the main skill set at 85 (from 82), recorded the October additions (lean-formalize, Grok and Gemini/Antigravity reviewer bridges, selectable reviewers, citation-crosscheck, bounded fan-out), moved the last push to October 20, noted the report’s HF #1 Paper of the Day badge
  • karpathy-autoresearch: rotation re-verification found the README’s own MIT declaration beside the missing LICENSE file; the unclear-legal-ground caution and the pricing clause were rewritten
  • autocontext: volatile refresh moved the push date to October 21, still off main, default branch newest commit remains September 29
  • kapso: volatile refresh moved the push date to October 20 (unreleased evaluation-run durability fix on main), release train still v0.7.10
  • local-deep-research: volatile refresh marked the October 20 push off main, newest main commit still October 19, release train still v1.10.7 on GitHub and PyPI
  • openresearch: volatile refresh moved the push date to October 21 (chat-activity unification with goal, plan, and experiment state), v0.2.19 still the latest release
  • automated-research: matrix updated the ARIS skill count and the Karpathy license cells in the same run as the notes
  • automated-research: rotation re-verified autoresearchclaw and dexter with no changes needed, every cited URL fetched live
  • llamafile: caution corrected, the x64 requirement is AVX (Intel Core 2006+/AMD K8 2003+) not AVX2, and the README compatibility note is now the 4GB Windows single-file limit; supported-systems docs cited.
  • plano: acquisition citation moved from the Business Wire release (now 403 to fetches) to DigitalOcean’s announcement post.
  • litellm: v1.106.1 stable 2026-10-20 with v1.105.2, v1.104.4, and v1.103.6 patched the same day, v1.107.0-rc.2 on the train, PyPI latest 1.106.1.
  • bifrost: five component lines patched 2026-10-20 (transports v2.3.2, telemetry v1.9.2, semanticcache v1.7.2, routing v1.2.2, prompts v1.1.12).
  • llm-gateway: v1.22.0 released 2026-10-19, repo pushed 2026-10-20.
  • magnitude: @magnitudedev/cli 0.2.10 published 2026-10-20, a memory-planning patch for unified-memory GPUs.
  • model-access: rafska/awesome-local-llm’s 2026-10-20 commit added Muse Spark Open and Hy-Memory, both rejected as HF model collections (model publishers family-route out).

(daily refresh, resolution wave)
#

  • orchestration-resolution: orchestration: frink admitted on the spec-driven routing handoff, the MIT desktop app pairing four-pane parallel agents with schedule-, webhook-, and event-triggered Flows (Linear, Shortcut, ClickUp, Sentry, PostHog, Vercel), 58 stars as of 2026-10-21, no independent coverage found and stated in the note [glm-5.3-flash]
  • orchestration-resolution: orchestration: gwq admitted as the second worktree-lifecycle CLI beside Worktrunk, d-kuro’s Apache-2.0 Go worktree manager with a global cross-repository view, status dashboard, and tmux sessions, 475 stars, v0.1.2 (2026-10-02) [glm-5.3-flash]
  • orchestration-resolution: orchestration: apowerb admitted into the framework wing, thaink²’ Apache-2.0 open-core agent platform on Google ADK and LiteLLM, 149 stars, PyPI 0.2.93 with 135 releases since July; note records the README-vs-docs token-quota discrepancy [glm-5.3-flash]
  • orchestration-resolution: orchestration: yylo admitted, JUNO AI’s MIT bounded-task orchestrator with per-task worktrees, evidence receipts, protected merges, and the Git-native Ledger, 65 stars, npm 0.2.12 ahead of a GitHub train stopped at 0.2.1-rc.14; note records the badge-wall and zero-coverage cautions [glm-5.3-flash]
  • orchestration-resolution: orchestration: matrix extended from sixty-nine to seventy-three columns with Frink, gwq, YYLO, and apowerb, re-sorted case-insensitively, all cells traced to the new notes; category index and its Changes updated to match [glm-5.3-flash]
  • harnesses-resolution: deepseek-harness - recorded the dsh-v0.2.1-rc.2 prerelease (October 17, three hours after rc.1), which added DeepSeek-V4.1-Flash-1017 support and DeepSeek Search and Fetch APIs in web search settings; the project remains prerelease-only.
  • harnesses-resolution: warp-agent-cli - folded the WarpBench field report (September 2, 2026) into the Factories status line as ecosystem evidence: a 30-task replay benchmark on the Warp Agent harness cut Warp’s internal factory cost per PR from about $80 to $30, an unreplicated vendor claim third-party coverage flags as still needing outside proof.
  • harnesses-resolution: copilot-cli - repointed the two GitHub docs references to their post-reorganization paths, both fetched at the new locations.
  • harnesses-resolution: harness-feature-matrix - repointed the Copilot CLI docs reference to its post-reorganization path.
  • harnesses-resolution: harnesses: resolved the CUGA cross-category candidate as a rejection (see Entrant resolutions).
  • sandboxing-resolution: sandboxing: overalls - added from the software-factory handoff (cross-category routing confirmed), ghuntley’s loom renamed to Overalls, the self-hosted persistent Firecracker microVM dev-machine platform, research preview launching 2026-11-03, 1,557 stars and MIT as of 2026-10-21; matrix to twenty-three columns and category index updated
  • sandboxing-resolution: sandboxing: overalls - loom history recorded in the note (repository created 2025-12-18, announced 2026-01-17 as The Weaving Loom do-not-use infrastructure, renamed by mid-October 2026 per the 301 redirect and the 2026-10-16 website repo); cautions: no release or tag, solo author, no audit, zero independent coverage pre-launch
  • model-access-resolution: kimi-code: recorded the two kimi-k3 API moves its price history lacked, the sixteenth move on 2026-10-19 (listing input leg $0.304 to $0.239, output $11.936 and cache $0.21 held, input floor to Wafer $0.238/$16.00, output floor collapsing to Relace $0.80/$3.00) and the 2026-10-21 recomposition (Relace re-priced to $0.30/$12.00, output floor to Wafer’s second endpoint $0.36/$11.90, Sail Research output leg $5.50, roster to 28 with Parasail and Modal joining) [glm-5.3-flash]
  • model-access-resolution: kimi-code: recorded the China ladder’s fifth tier, Ultra at ¥1,399/month (¥1,119 with annual), present by the help page’s 2026-10-19 capture, international ladder unchanged; both ladder as-of dates moved to 2026-10-21 [glm-5.3-flash]
  • cross-category: stars scan found no new stars since the 2026-10-17T05:59:29Z checkpoint (newest starred_at is the checkpoint itself, BuilderIO/agent-native, already a skills member); checkpoint unchanged
  • cross-category: broad sources carried over unchanged since the 2026-10-16 full scan for ai-for-developers/awesome-ai-coding-tools, e2b-dev/awesome-ai-agents, and slavakurilyak/awesome-ai-agents (no commits since)
  • cross-category: andyrewlee list’s four 2026-10-17 additions (ignite, solenta, agent-console, upcomputer) held on the standing mission-control family rejection
  • cross-category: AMO list changes run through 2026-10-19, covered by the 10-19 AMO wave; the two riders (apowerb, yylo) resolved here
  • cross-category: apowerb: reported to orchestration on the Dify/Flowise self-hosted-platform precedent (149 stars as of 2026-10-21, Apache-2.0, pushed 2026-10-20)
  • cross-category: yylo: reported to orchestration as a CLI task-lifecycle orchestrator (65 stars as of 2026-10-21, MIT, pushed 2026-10-17)
  • cross-category: geoai-skills: rejected on the standing skills exclusion (18-skill pack; scope is formats, registries, marketplaces, optimizer and discovery tooling)
  • cross-category: i18next-cli: rejected on the standing skills exclusion (single vendor in-product skill inside the i18next CLI repo)
  • cross-category: cuga-agent: pending routing confirmed to harnesses over control-planes (self-described enterprise agent harness, 1,045 stars as of 2026-10-21, Apache-2.0, pushed 2026-10-20, IBM-backed)
  • cross-category: gwq: routing corrected to orchestration on the worktrunk precedent (git worktree manager for parallel agents, 475 stars as of 2026-10-21), not sandboxing or software-factory
  • cross-category: overalls: loom-to-Overalls rename confirmed (ghuntley/loom 301-redirects); sandboxing candidate confirmed (persistent Firecracker microVM agent machines, 1,557 stars as of 2026-10-21, MIT, pushed 2026-10-21, research preview, launches 2026-11-03)
  • cross-category: warpbench: the field report (warp.dev/factories/benchmarks/warpbench, Zach Lloyd, 2026-09-02) folds into the warp note as vendor ecosystem evidence, not people-and-publications
  • cross-category: orkas-videostudio: skills routing confirmed as the nearest category, with the pack-exclusion caution left to the skills worker (331 stars as of 2026-10-21)
  • cross-category: frink: orchestration routing corroborated by a sighting in the andyrewlee list’s Autonomous Task Runners section; already resolved in the digest the same day
  • kimi-code (model access): recorded the 2026-10-19 and 2026-10-21 price-history rows plus the China Ultra tier row found during the re-fetch [glm-5.3-flash]
  • orchestration: frink, gwq, apowerb, and YYLO admitted with notes, closing the 2026-10-19 pending-candidate defect; matrix and category index extended [glm-5.3-flash]
  • harnesses: CUGA rejected as not coding-harness kind (web/API automation surfaces, AppWorld/WebArena benchmarks, extending the deepagents ruling); DeepSeek Harness rc.2 recorded; WarpBench folded into the warp note with the skeptical field-report framing; copilot-cli and the harness matrix’s Copilot docs references repointed to the reorganized paths [glm-5.3-flash]
  • sandboxing: overalls (ghuntley/overalls, the renamed Weaving Loom) admitted, matrix to twenty-three columns, index updated [glm-5.3-flash]
  • kimi-code (harnesses): daily-churn refresh, main listing $0.239/$11.936 with cache reads listed at $0.00, Wafer the input floor at $0.238/$16.00, roster 28 (coordinator) [glm-5.3-flash]
  • geoffrey-huntley: The Weaving Loom’s rename to Overalls recorded with the link to the new sandboxing note (coordinator) [glm-5.3-flash]
  • vscode-copilot: the cloud-agent docs reference repointed to the copilot-surfaces path (coordinator) [glm-5.3-flash]
  • cross-category scan: no new stars since the 2026-10-17 checkpoint (unchanged) [glm-5.3-flash]