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

Standing decisions digest

Author
Tom Rochette
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / 43 minutes read (~8532 words)
agents 113
Table of Contents

One line per standing decision, maintained by the daily refresh. The log under agents/logs/ stays the audit trail; this digest never contradicts it. Line format: - YYYY-MM-DD | category | decision | one-line reason or trigger (date = last confirmation).

Scope rulings
#

  • 2026-10-31 | assistant-runtimes | openhuman reads 42,441 stars on 2026-10-31, about +17 in a day (growth still slowing) against OpenClaw’s 393,510 | the q13 OpenClaw-class scale trigger stays unfired, openhuman held

  • 2026-10-31 | code-review | desktop human-review clients for agent PRs judge nothing and are evaluation-review kind on the judges-nothing family | areibman/bottleneck (157 stars, MIT Electron client, zero machine judgment, dormant since 2025-12-04, 2-point Show HN 2025-10-18) held on it; the coordinator handoff had routed it here as code-review-kind

  • 2026-10-31 | evaluation-review | areibman/bottleneck admitted as the queue-level judges-nothing review client beside Hunk, PR Lens, and Crit, resolving the code-review handoff | 157 stars, MIT, twelve releases, live installers; dormancy since 2025-12-04 and the 2-point zero-comment Show HN ride the note, reopen the dormancy reading on any push or release

  • 2026-10-31 | control-planes | the awesome-ai-governance thin tail grows by two: Agenthof (0 stars) and solvi (4 stars) join the sub-bar open-source tier with the governs-something test already passed, so only traction or coverage reopens them | both gate real actions but have zero independent coverage; keel-verifier resolved separately on the below-the-plane evidence-verifier family and the test checklist on the 2026-10-22 content ruling

  • 2026-10-31 | executions | a matrix cell edit re-pads its row back to header width in the same run | the 2026-10-26 n8n and 2026-10-30 Trigger.dev cell edits left the Trigger types and Guardrails rows 17 and 22 characters over header width until the 2026-10-31 pass re-padded them, padding only

  • 2026-10-31 | harnesses | a provider API fee attached to a member’s optional integrations lands in that member’s Pricing and Price history, not the matrix | the DeepSeek Search and Fetch fee ($0.75 per 1,000 successful requests, from 2026-11-01) landed on deepseek-harness; the matrix carries no DeepSeek-priced cell

  • 2026-10-31 | hybrid-execution | JevBench composite citations carry the method version alongside the board state, since the v1.8 cost re-basing moved self-hosted composites without any member-side signal | Torchcast 4B v1.3 80.5, H2O v1.1 77.9, Winnow Q8 72.21, Kev 27B 71.26, decider-12b v2 73.73 all read differently from the v1.7 basis

  • 2026-10-31 | memory | the TeleAI list’s 2026-10-30 to 10-31 additions resolve on standing families | Mnemosyne-28naem-del (154), daimon (21), Okto Neuron (9), and rsh1k/mnemosyne (1) below the 294-star floor, TokenRip closed-hosted with an MIT CLI and no open engine, MAQS an audit checklist (content), MAQS-Way a conformance spec (evaluation-review kind, flagged), World-T3/GESTO/VaM paper-section entries, and akephalos archived upstream with its repository removed

  • 2026-10-13 (ruled 2026-10-06) | retrieval | founding scope carries exactly two framework columns (LangChain, LlamaIndex); a third framework is an owner scope call, not an entrant signal | Haystack rejected under it, queue q#4 open

  • 2026-10-13 (ruled 2026-10-07) | evaluation-review | founding structure stands: Opik, promptfoo, and AgentOps are third-column or second-framework scope calls, not entrant signals | adoption between existing columns does not override structure, queue q#6 open

  • 2026-09-29 (ruled 2026-09-27) | orchestration | category scope includes multi-agent frameworks | owner approved the AMO directory’s in-scope gaps; the framework tier (CrewAI, AutoGPT, AutoGen, Dify, and peers) admitted 2026-09-29

  • 2026-10-10 | orchestration | self-evolving workflow frameworks slot in orchestration’s framework family, not automated research | EvoAgentX routed and admitted here

  • 2026-09-27 | assistant-runtimes | category scope includes local-first platforms | Open WebUI, AnythingLLM, and PrivateGPT admitted in the stars-scan resolution

  • 2026-09-26 | model-access | vendor consumer subscriptions that ship an agent are access products; editor-bundled access (Cursor, Copilot) stays in surfaces pending owner direction | claude-plans and chatgpt-plans admitted on the owner’s reading

  • 2026-09-26 | model-access | rejected by kind: consumer assistants without a coding-agent story, pay-per-token inference hosts, enterprise clouds, gray-market discount marketplaces | Poe, Perplexity, Together, Fireworks, Groq, DeepInfra, Bedrock, Vertex, Azure, codingplan.io-style resellers

  • 2026-09-27 | model-access | scope includes the self-hosted gateway software layer | litellm admitted; experiential followed the same day

  • 2026-10-21 | model-access | serving stacks, chat UIs, engines, and model publishers family-route out of model access | lemonade and llama-swap held as same-slot duplicates; NVIDIA dynamo (datacenter-scale inference serving, starred 2026-10-21) held on the family route; Decision 3.0 (HF model collection) held 2026-10-14

  • 2026-10-13 (ruled 2026-10-10) | surfaces | end-user app-building platforms and general-purpose agent workspaces stay out; coding support is the category spine | tambo rejected on the app-building kind call; OpenBitFun rejected as a Cowork-style workspace, queue q#14 open

  • 2026-09-12 (ruled 2026-08-24) | orchestration | terminal-family parallel-agent tools and standalone ADEs place in orchestration over surfaces | dmux and claude-squad set it, Conductor and Superset carried it, JetBrains Air applied it

  • 2026-09-16 (ruled 2026-08-22) | cross-category | vendor sibling products fold into the sibling note by default | gemini-cli/Antigravity set it; applied to oh-my-pi, Codex App, OpenCode Desktop, Warp Factories, LiteParse, cursor CLI and routines; oh-my-pi left the fold on its own bar clearance 2026-10-10

  • 2026-09-21 (ruled 2026-09-18) | hybrid-execution | model-layer members are admissible | Jev entered first; laya superseded the 2026-09-20 wave rejection once the model-layer objection no longer held

  • 2026-09-16 | hybrid-execution | engine internals stay cited inside member notes, below the application-layer scope | xgrammar and vLLM structured outputs live inside outlines

  • 2026-09-16 | automated-research | closed capability-research instruments qualify where no self-hostable runtime exists | Pion placed here over assistant runtimes

  • 2026-10-04 (ruled 2026-09-24) | model-benchmark-matrix | the matrix is model-only: harness benchmarks and harness-variance studies stay out as pointers | FrontierHarness Eval, HarnessTax, JevBench, the Scaffold Effect study

  • 2026-10-10 | code-review | review surfaces that judge nothing are evaluation-review members, not code-review | PR Lens shelved beside Hunk; ReviewBench confirmed evaluation-review territory 2026-10-07

  • 2026-10-10 | spec-driven-development | entrants route by primary artifact: a spec contract is spec-driven even when tracker-adjacent | Intent-Router shelved here over task management; WrenAI likewise context engines over hybrid execution

  • 2026-10-13 (ruled 2026-10-11) | context-engines | archived-by-owner tools are admissible on founding role and scale, not by default | Vanna admitted archived; ByteRover rejected on the distinguished precedent

  • 2026-10-07 | protocols | servers and utility layers are not protocols | IBM mcp-context-forge routed to control planes instead (admitted there 2026-10-11); agnix and rulesync rejected

  • 2026-10-10 (ruled 2026-10-07) | control-planes | a member must govern something: provenance recorders stay below the plane | git-ai and Invariant rejected on it

  • 2026-10-07 | sandboxing | hosted-only services and environment images without an isolation boundary stay out | Deno Sandbox under the Modal precedent; AIO Sandbox an environment image, not a boundary

  • 2026-10-10 | executions | data-engineering-first platforms stay out | kestra (29.5k) rejected and re-confirmed as data-engineering-first

  • 2026-10-13 (named 2026-10-05) | people-and-publications | a voice is admissible only when their breakout first surfaces inside the section’s coverage window; long-established and interview-only voices fail | applied in every weekly voice scan

  • 2026-10-21 | spec-driven-development | the thin-tail precedent (0-10 stars, no independent coverage, reopen on traction or coverage) applies to spec-standard and blueprint kinds | ANSS (6 stars), internal-tool-blueprints (0 stars), one-man-startup (9 stars), and aide (1 star, zero coverage, spec-toolkit kind) rejected on it

  • 2026-10-06 | trackers-and-leaderboards | single trend series are not trackers and evals nonprofits are people-category candidates | METR’s time-horizon series rejected as a tracker, queue q#5 open

  • 2026-10-17 | memory | the OKF-implementation family folds into the okf-agent-memory note as ecosystem evidence | mcp-memory (218 stars, dormant since 2026-08-30) and Kage held on it

  • 2026-10-18 | memory | positional star ranks (“Nth of M members”) are no longer maintained in note prose; the claim churned five correction rounds in two notes and went stale at every admission | memU and screenpipe rank lines dropped

  • 2026-10-17 | memory | Vestige pivoted from queryless memory to a fail-closed action firewall (Strata log v4.0.0 landing 2026-09-30, the Operator gate and $149 tier v4.2.0 2026-10-06) and routes out of memory on the primary-artifact reading | admitted to control-planes the same run, 4-point Show HN July 2026 pre-pivot

  • 2026-10-17 | skills | the OKF delivery-toolkit kind routes to the memory category’s okf-agent-memory ecosystem family, not to skills (the skills are the delivery mechanism, and OKF packages knowledge about systems, not agent capabilities) | scaccogatto/okf-skills (417 stars) folded there 2026-10-17

  • 2026-10-21 | evaluation-review | lukstei/slop-grader admitted as the document-and-diff grader beside Jevals, the category’s second Jev-based member | skills-worker handoff resolved on the citation bar; the thin one-thread HN footprint is stated in the note, on the HermesGate precedent

  • 2026-10-21 | spec-driven-development | context-engineering layers route here when the artifact is a typed spec set plus a convergence check that names the guilty side | archcore admitted on the Intent-Router primary-artifact precedent; the looptroop-ai/LoopTroop duplicate admission was retracted the same run, the orchestration member stands as the primary frame

  • 2026-10-21 | control-planes | Tracekit admitted as the category’s eleventh gate at 138 stars | governs calls before execution with a CT-style witness-cosigned evidence log, unlike any member; the experimental no-public-witness gap is stated in the note

  • 2026-10-22 | hybrid-execution | the Decision Index 0.4 Full score is fully private (60 percent same-skills, 40 percent new-domains, the public index shown but uncounted), so member rows cite the 0.4.1 reading | board regenerated 2026-10-22 with 226 configurations, GLiDE V1.2 first at 71.52, Jev 1.14 forty-first

  • 2026-10-22 | hybrid-execution | TypeChat’s npm-publication reopen trigger counts a revival-era publication only, not the pre-revival npm line | the npm typechat package (latest 0.1.4, 2026-08-21) predates the 2026-10-13 revival push, asked in queue

  • 2026-10-22 | cross-category | npm weekly download figures carry zero-day gaps, so adoption claims prefer the trailing-month window | context7’s npm week to 2026-10-18 reported 482,795 with three zero days inside the window while the trailing month read 2,640,895

  • 2026-10-22 | cross-category | notch supervision surfaces, passive notification and approval-forwarding UIs over orchestration members, reject as tails under the mission-control dashboards-and-tails family | perch (asafofer/perch, 22 stars, zero coverage) held on it

  • 2026-10-22 | cross-category | guides, cookbooks, prompt references, and pattern collections are content, not tools, and stay unprofiled regardless of scale | the unmatched broad-source tail (claude-howto 42.5k, claude-code-system-prompts 13k, loop-engineering 11.6k, infrastructure-showcase 10k) reads out on it

  • 2026-10-23 | automated-research | single-harness personal autoresearch ports with no independent coverage stay unprofiled | the webfuse-com/awesome-autoresearch scan’s pi/mlx/webgpu/sudoku-class tail read out on it

  • 2026-10-23 | cross-category | benchmark and evaluation-harness repositories are not research loops and route to evaluation-review | the webfuse-com scan surfaced mle-bench, AgentBench, MLAgentBench, ML-Bench, and mlrbench

  • 2026-10-23 | automated-research | dormant members below 500 stars with no independent coverage close until new coverage lands, per the concord-and-huzzah rule | applied to Sibyl-Research-Team/AutoResearch-SibylSystem, MLR-Copilot, ResearchAgent, ML-Agent, autoresearcher, self_improving_coding_agent

  • 2026-10-23 | evaluation-review | a member’s subject repository moving to a redirecting path is a note-level provenance fact, not a scope change | hermes-labs-ai/hermes-gate now redirects to roli-lpci/hermes-gate; the note’s twin-repo caution was rewritten to the present fact in rotation

  • 2026-10-25 | automated-research | a member repository’s transfer into an institutional organization is a recorded note fact, with the star-history chart and references moved to the canonical path the same run | AutoSci moved from skyllwt to PKU-DAIR (the Peking University DAIR lab of senior author Bin Cui), old URLs redirect and the README still points at them

  • 2026-10-23 | model-access | upstream -pro model ids (OpenAI GPT-5.6/6 Sol/Luna/Terra Pro) are provider-side serving tiers, not aggregator-only lines; the category records them in openrouter with the same-price-more-reasoning-tokens cost trap named | resolved the 2026-10-22 section-root sighting on OpenRouter endpoints evidence (OpenAI flex/fast/ultrafast and Azure endpoints, dated snapshot ids, base-identical pricing)

  • 2026-10-25 | trackers-and-leaderboards | bestvaluemodel admitted, resolving the 2026-09-25 reopen-on-license line | a month of six-hourly refreshes with daily data diffs, still fully derivative of member Artificial Analysis with credit, and the single-source caveat stands

  • 2026-10-28 | trackers-and-leaderboards | the category still keeps no dedicated awesome-list source | the per-refresh check re-ran 2026-10-28; the strongest new candidate (tatn/awesome-ai-benchmarks, 7 stars) sits far below the other categories’ mapped sources, and the BenchGecko and leoncuhk rejections stand

  • 2026-10-25 | context-engines | a pushed-signal that resolves to non-default-branch activity, no main-branch commit and no release, is non-material and leaves the note untouched | entroly, SoL-Pi, and repowise pushes on 2026-10-25 all read out as agent/*-branch activity over mains last touched 2026-10-16 to 2026-10-20

  • 2026-10-25 | executions | version-gated claims cite only gates still present on the live doc page at re-verification | the hooks v2.1.214 if directory-pattern gate and the routines v2.1.213 trust-model gate both aged out of the live docs within a month of being cited

  • 2026-10-25 | hybrid-execution | the Jeff name on the JevBench board belongs to jgeuter’s unrelated Gemma-4-31B fine-tune, so board credits in member notes check the row’s Hugging Face owner before citing | the 10-15 and 10-20 jeff bullets credited the family with a row that was never its, corrected 2026-10-25

  • 2026-10-25 | hybrid-execution | member notes citing JevBench capability standings carry their board revision, since method v1.7’s 10-second rule re-ranked every row on its landing (H2O-Lightning-31B v1.0.7 first at 84.3, Chon-1.0-12B from first to rank 30 in one day) | board v1.7.48 fetched 2026-10-25

  • 2026-10-25 | cross-category | the andyrewlee 2026-10-24 addition bdots holds on the assistant-runtimes absorbed runtime family (renamed upstream to fayazara/coworkers 2026-10-28, 306 stars, the rejection carries by mechanism) | a Cloudflare-native personal multi-bot runtime, the cloudflare-os slot; zero independent coverage, reopen on q#13, OpenClaw-class scale, or coverage

  • 2026-10-25 | cross-category | the ai-for-developers list’s Code Review section resolved in full (twenty-two entries, all rejections or already-covered); its remaining sections and the other broad sources’ back catalogs get a one-off unresolved-entries walk | the 2026-10-25 sweep found ten never-resolved entries hiding in a source the scanner reads, so carry-over-on-commit-history misses back-catalog entries

  • 2026-10-23 | session-analytics | CodexMonitor admitted as the category’s first dormant member; dormancy within two months of a vendor surface launch is the note’s finding, not a rejection reason | the 4,348-star Codex command center went quiet days after Hacker News compared it with the official Codex app

  • 2026-10-23 | session-analytics | ax’s session-graph core makes it session-analytics despite the repository’s observability-plus-memory self-description | the memory half is session-derived introspection (retro loop, AGENTS.md fixes, skill triage), not a cross-project memory layer

  • 2026-10-23 | spec-driven-development | a renamed-and-rebranded candidate re-resolves on current evidence; the old rejection carries by mechanism, not by name | BusinessLens rejected below the bar 2026-10-13, admitted as pdd-kit on the rebrand, the shipped site, and continued daily development

  • 2026-10-23 | task-management | ccpm admitted as the category’s forge-native kind | coordinator handoff resolved; MIT skill running PRDs onto GitHub Issues with worktree parallelism, 8,414 stars, a 175-point Show HN thread, and independent critical coverage clear the citation bar

  • 2026-10-23 | trackers-and-leaderboards | the category keeps no dedicated awesome-list source; the broad sources suffice | the two candidates found 2026-10-23 fail the bar, BenchGecko/awesome-llm-benchmarks stale since creation 2026-03-27, leoncuhk/awesome-llm-bench a daily mirror of member BenchLM

  • 2026-10-23 | cross-category | the andyrewlee 2026-10-22 additions resolve on standing families | Delegatus on the thin-record tail, deepseek-harness-action as wrapper tooling on an unprofiled subject, Harness by Autonomous routed to orchestration on the standalone-ADE precedent (admission resolved by the next run)

  • 2026-10-26 | cross-category | star-history chart markup: only the dark source in a picture block carries theme=dark, never the light source or the img fallback | coordinator handoff; the 2026-10-25 pass fixed the img fallbacks but left light sources carrying theme=dark across five categories, all fixed 2026-10-26

  • 2026-10-26 | cross-category | a subject board’s page drifting under an unchanged revision number counts as a signal and re-verifies board-citing notes out of rotation | the jev rotation re-fetch found JevBench v1.7.48 grown from 220 to 227 ranked open-weights and 47 to 48 API rows since the 10-25 fetch

  • 2026-10-26 | cross-category | Hugging Face download counts are rolling-window figures that swing both ways, so rotation refreshes them with a fresh as-of date and reads no trend into a swing | decider-2b read 334,268 on 10-18 and 164,599 on 10-26 with no release in between

  • 2026-10-26 | skills | the rea.tools scale record grew from the 109.9k recorded 2026-10-13 to 387,014 stars (morluto/rea, fetched 2026-10-26), now the largest known skills-adjacent record and still outside on the standing exclusion pending q#11 and q#20 | the VoltAgent/awesome-agent-skills scan surfaced the repo as a new entry indexing one of its skill directories

  • 2026-10-26 | software-factory | the ahmadyan/awesome-software-factory list (2 stars, single-day push, maintainer self-promotes Agentastic.dev) stays below the dedicated-source bar and is not adopted as a mapped source | its entries cross-check to known members, already-resolved subjects, or other categories’ kinds

  • 2026-10-26 | spec-driven-development | acceptance-test and QA harnesses route to evaluation-review under the primary-artifact rule | vostride/agent-qa (895 stars, FSL-1.1-ALv2, self-published AndroidWorld and AppControlBench results) authors and runs acceptance tests, not spec contracts

  • 2026-10-27 | cross-category | npm’s point/last-month downloads endpoint can serve a stale window (June reads on October 27 across harnesses, protocols, and context engines), so refreshes use dated-range queries (api.npmjs.org/downloads/point/:) and range sums whenever the point response’s window dates lag today | ag-ui, repomix, copilot-cli, and bullet all read stale or wrong figures from point endpoints before range sums corrected them

  • 2026-10-27 | surfaces | OpenBitFun stays out on the Cowork-style general-purpose workspace rejection and the q#14 re-read found the same self-described workspace with a Mini App marketplace, so the question reads closable | surfaces re-check on the governing question

  • 2026-10-27 | surfaces | a vendor with a dual identity files by its desktop editor kind, folding the sibling product in as prose | puppyone admitted as an agent editor even though its own docs pitch a Context File System

  • 2026-10-27 | memory | the Meirtz-list handoff portable-ai-memory/python-sdk, the PAM interchange spec’s SDK, stays below the bar | 1 star, dormant since 2026-02-19, spec repo 12 stars, NOASSERTION license, zero independent coverage, providers do not natively support the format; reopen on traction or coverage

  • 2026-10-27 | evaluation-review | vostride/agent-qa admitted as the acceptance-test column; the 2026-10-12 dormancy hold triggered on the 2026-10-24 push and three releases | citation bar cleared in-run with the docs hub, CI guide, both self-published benchmark pages with per-task evidence, the FSL license text, npm SLSA provenance, and the 50-point launch thread; the vendor-run benchmark caveat rides the note like FrontierHarness Eval

  • 2026-10-27 | cross-category | run-recording, replay, and causal-analysis repositories are session-analytics kind, not evaluation-review | Psionic-labs/Agent-Causality and MSKazemi/novafabric routed there; they re-run and slice recorded runs, they judge nothing

  • 2026-10-28 | assistant-runtimes | openhuman stays held on the absorbed-family shelf, the q13 OpenClaw-class scale trigger as framed still unfired | tinyhumansai/openhuman read 42,378 stars on 2026-10-28 against 42.2k on 10-21 (about +180 in a week) while OpenClaw reads 393k

  • 2026-10-28 | model-access | a member note’s recorded pricing-page rebuild moves that note’s matrix cells in the same run, not only membership changes do | the experiential 2026-10-24 flat-plan rebuild reached the note and price table but left the matrix’s billing-mechanics, top-tier, and price-trajectory cells stale until this pass caught them

  • 2026-10-28 | people-and-publications | pauldambra.dev rejected as a voice: the flagged “Machinists > Tailors” post surfaced at 2 HN points and 0 comments and no independent coverage, so the coverage-window bar is unmet | reopen on independent coverage of any agent-era post

  • 2026-10-28 | software-factory | varun1505/awesome-software-factories (42 stars) remains the only dedicated software-factory list above the bar | the 2026-10-28 search surfaced only single-star single-author lists beside the already-rejected ahmadyan list

  • 2026-10-28 | section-root | the section index’s verified stamps, member counts, and newest-column claims are synced mechanically from each category matrix’s own updated date, header row, and Changes tail whenever the section-root worker holds the index | the 2026-10-27 interrupted run updated category matrices without an index pass, leaving ten counts and twelve newest claims stale against the matrices they describe

  • 2026-10-30 | assistant-runtimes | LangBot admitted as the category’s sixteenth member from the new elyase/awesome-personal-ai-assistants scan | Apache-2.0 IM-bot platform in the AstrBot kind, 18,601 stars above the Octop floor, weekly v4.11 cadence, PyPI Production/Stable, the zero-English-footprint signal stated in the note

  • 2026-10-30 | assistant-runtimes | the elyase list’s claw-named tail (MimiClaw 5.8k, IronClaw 12.7k, OpenBrowserClaw, MyClaw, SubZeroClaw, Clawlet, and Cloudflare’s MoltWorker) holds on the absorbed-family shelf by mechanism, and its non-claw assistants (moltis 2.9k, Picobot 1.3k, Sentient, Gaia, rho, the archived lettabot and clawlet, the dormant personal-ai-assistant and MARVIN) reject on the OpenAgent floor or thin records | one scan, no new rulings needed

  • 2026-10-30 | assistant-runtimes | openhuman reads 42,424 stars on 2026-10-30, about +46 in two days (slower than the +180-per-week recorded 10-28) against OpenClaw’s 393k | the q13 OpenClaw-class scale trigger stays unfired, openhuman held

  • 2026-10-30 | code-review | the vendor-run kodustech/awesome-ai-code-review list stays unwired as a mapped source; the category stays on broad sources | staleness re-confirmed 2026-10-30 (pushed_at still 2026-06-28) and no credible dedicated list has appeared (largest dedicated candidates 1 to 2 stars); updates the 2026-10-24 line’s evidence for open q21

  • 2026-10-30 | control-planes | rotation re-verifications record a ## Changes bullet even with no drift | the work list derives last_verified from the newest Changes bullet, so a silent pass never advances and the same notes re-rotate; seven notes already followed this convention before this run

  • 2026-10-29 | section-root | The section-root essays’ star-count and install-count prose (Orca, Claude Code, OpenCode, DeepSeek Harness, Kimi Code, Cline) stops refreshing under the stars exemption; the existing dated figures stand | the dispatch restates “stars never refresh anywhere” and the essays’ older daily star refreshes contradict it, so the prose counts freeze at their last dated values.

  • 2026-10-29 | section-root | A pure delta-pass re-dates the tracker’s as-of clause only when a fact moved | zero-move runs (this one for five of seven provider pages) leave the as-of anchored to the last change date, matching the 2026-10-21 and 2026-10-23 precedent.

  • 2026-10-29 | trackers-and-leaderboards | the category still keeps no dedicated awesome-list source; the per-refresh check re-ran 2026-10-29 | the only literally-dedicated list (metaory/awesome-llm-leaderboards, created 2026-09-10) reads 2 stars, below even the rejected 7-star tatn candidate, and the credible new benchflow-ai/awesome-evals (1,017 stars) is evaluation-infrastructure kind, not trackers

  • 2026-10-29 | automated-research | the InternScience org scan resolves with one admission, one sibling fold, and the benchmark route | InternAgent admitted with MLEvolve folded in as its optimization subsystem; Dolphin, DiscoveryOS, and Virtual-Scientists closed on the dormant-below-500 rule until coverage lands; SGI-Bench routed to evaluation-review on the benchmark family

  • 2026-10-29 | assistant-runtimes | openhuman stays held on the absorbed-family shelf, the q13 OpenClaw-class scale trigger as framed still unfired | tinyhumansai/openhuman read 42,403 stars on 2026-10-29 against 42,378 on 10-28 (about +25 a day) while OpenClaw reads 393k

  • 2026-10-30 | executions | rotation re-verifies read the subject’s product changelog, not just the repo API: Trigger.dev’s inbound-webhooks beta (2026-10-26) invalidated the note’s and the matrix’s no-webhooks claims between passes, and only the changelog page surfaced it

  • 2026-10-30 | executions | no credible dedicated executions source has appeared as of 2026-10-30; the category keeps riding the broad sources

  • 2026-10-30 | harnesses | the pi port wave routed from the assistant-runtimes scan (Dicklesworthstone/pi_agent_rust 1.9k, joelreymont/pz, philipp-spiess/modern, matteing/opal) resolves on the wrapper-tooling-on-a-member mechanism, the 2026-10-28 Ralph-loop precedent, with pi_agent_rust folded into the pi note as ecosystem evidence | ports, rewrites, and GUIs of an existing member are that member’s ecosystem evidence, not standalone harnesses

  • 2026-10-28 | cross-category | the andyrewlee 2026-10-27 additions resolve on standing families and one handoff | hotline and open-walnut on the thin-record tail, agentglass on the mission-control tail, Scalekit on the closed-tier line, Speakeasy Gram on the MCP-platform kind, CAO (awslabs/cli-agent-orchestrator, 1.5k) handed to orchestration on the OpenHarness precedent

  • 2026-10-28 | cross-category | the six broad sources’ back catalogs are fully walked and carry-over-on-commit-history gaps are closed | 950 never-matched entries resolved in one pass: 93 dead links, the rest on standing families; the largest new records read onto open questions (andrej-karpathy-skills 225.6k content, archify 90.6k single skill, learn-claude-code 79.4k content, claude-hud 28.9k tail, claude-obsidian 15.9k memory-kind handoff)

  • 2026-10-28 | cross-category | same-kind multi-agent frameworks below the admitted tier’s adoption stay out while the tier question is open | swarms 7.3k, agency-swarm 4.6k, langroid 4.1k rejected on it; ChatDev 34.6k (active 2.0) handed to orchestration as the one scale-and-activity exception

  • 2026-10-28 | cross-category | the Ralph-loop cluster is wrapper tooling on the claude-code member, ecosystem-evidence kind | ralph-claude-code 9.7k is the head; pattern writeups are content under the 2026-10-22 ruling

  • 2026-10-31 | cross-category | HN Algolia search citations need the query re-checked at re-fetch, because loose query strings silently stop matching while the cited threads still exist | the bernstein reference (query=bernstein.run&tags=story) returned zero hits while query=bernstein&tags=show_hn returned the cited 1-point Show HN; the claim stood, the URL no longer grounded it

  • 2026-10-31 | orchestration | a member’s product rename records in place (note title kept), with the star-history chart and references moved to the canonical path the same run; a retitle is an owner call | superset-sh/superset renamed to ParadevAI/Paradev, README headlines “Paradev (formerly Superset)”; retitle question appended to the queue

  • 2026-10-31 | surfaces | OpenBitFun’s Cowork-style general-workspace rejection holds on the third q#14 re-read (the README still pitches a general-purpose agent workspace with a Mini App marketplace, now 3,423 stars), so the question reads safe to close | surfaces re-check on the governing question

  • 2026-10-31 | task-management | the dedicated-list check’s incidental candidates resolve on the thin-record tail without a mapped-source scan | rasimme/FlowBoard (96 stars) and DanWahlin/ai-agent-board (67 stars) surfaced in a dedicated-list websearch, both zero independent coverage

  • 2026-10-31 | trackers-and-leaderboards | the category still keeps no dedicated awesome-list source; the per-refresh check re-ran 2026-10-31 | the strongest new candidates are brandonhimpfen/awesome-ai-benchmarks-evaluation (14 stars, active 2026-10-07) and git-git-hurrah/awesome-llm-leaderboards (6 stars, stale since April), both far below the other categories’ mapped sources, and onejune2018/Awesome-LLM-Eval (659 stars) is evaluation-kind and stale since 2025-11

  • 2026-10-31 | trackers-and-leaderboards | the Vectara board’s published rankings are computed by the commercial HHEM-2.3, with HHEM-2.1-Open as the downloadable open variant | the README and model card state it in plain text; the note, three matrix cells, the verification prose, and the index one-liner corrected 2026-10-31

  • 2026-10-31 | cross-category | the andyrewlee and AMO 2026-10-29 to 10-31 additions resolve on standing families | nine thin-record tails held (wsp, volt, borg, lookout, pohunek, yardlet, MercuryCLI, Aixle Flow, sorcerer), alchemist-coder’s AMO-wave rejection carried by mechanism on the second listing, herdr-core and herduck folded into the herdr member as ecosystem evidence

  • 2026-10-31 | orchestration | Commonly (Team-Commonly/commonly, 1,374 stars as of 2026-10-31) admitted on the OpenHarness standalone-ADE precedent, the zero-coverage signal stated | Apache-2.0 self-hostable team pod sharing one task board and conversations with each member’s agents, added to AMO’s Coordination And Team Systems 2026-10-30, v3.0.0 2026-10-28

  • 2026-11-01 | cross-category | the AMO 2026-10-30 to 11-01 additions resolve with one thin-tail rejection and two already-covered listings | Joxo (89 stars, closed-source hosted, plugin-only repo, zero coverage) rejected on the thin-record tail; the #97 Google AX listing is the orchestration ax member’s own repository and #99 is a Herdr entry upgrade, no membership change

  • 2026-11-01 | section-root | Superset is now Paradev (ParadevAI/Paradev, Elastic License 2.0, “formerly Superset” per its own README, CLI still superset); treat any remaining Superset mentions in essays as Paradev | repository README fetched 2026-11-01

  • 2026-11-01 | section-root | Paseo is Apache-2.0, relicensed from the AGPL-3.0 earlier records carry, with Antigravity and Muse Code added to its provider roster | repository README fetched 2026-11-01

  • 2026-11-01 | model-access | a vendor page contradicting itself between its plan table and its own FAQ/example text is recorded with both readings and the disagreement named, never averaged into one number | the experiential 2026-11-01 fetch read 10M/100M/1B tokens-a-day ceilings in the plan table against the FAQ’s retained 9B/36B/54B, so the note and matrix carry both with dates

  • 2026-11-01 | people-and-publications | ludicity (Nik Suresh, Hermit Tech) rejected as a voice on the long-established clause of the 2026-10-13 ruling: the author’s breakout essays predate the coverage window, so the July 2026 AI Mania coverage is not an in-window breakout | the essay itself remains admissible as third-party evidence inside member notes

  • 2026-11-01 | protocols | a community SDK merged to main counts as shipped for the note’s Status even while the README and docs tables lag | ag-ui’s Swift SDK (PR #1512, 2026-10-31) landed in sdks/community/swift with the README table not yet naming it; recorded in Status with the lag noted

  • 2026-11-01 | sandboxing | the arjan/awesome-agent-sandboxes back catalog is fully walked and its carry-over gap is closed | the 2026-10-09 walk had logged only its two admissions, so twenty-two unmatched entries resolved this run on standing families, dormancy, or kind, with two cross-category flags (rivet-dev/sandbox-agent, portofcontext/pctx)

  • 2026-11-01 | sandboxing | a subject’s self-published coordinated-disclosure record is disclosure evidence, not an audit, wherever its audit status is stated | nono now carries fifteen advisories since May 2026 including the CVE-2026-47128 D-Bus systemd-run sandbox escape and four high-severity bypasses, recorded in its note and the matrix’s audit prose

  • 2026-11-01 | cross-category | OpenReview forum and API pages serve a bot-check wall to unauthenticated fetches, so venue-acceptance claims cite the fetched page that records the citation instead of the unfetchable forum URL | the EvoSkill rotation grounded the Agent Skills ‘26 workshop acceptance (ACM CAIS 2026) in the fetched v2 branch README after openreview.net/forum?id=i4b4As7RFX and its api2 endpoint both returned the verification wall

  • 2026-11-01 | software-factory | camplight/awesome-software-factories (8 stars, pushed 2026-10-28) joins the below-the-bar dedicated-list set beside the ahmadyan rejection | single-author list at trivial adoption; varun1505 (55 stars) remains the only list above the bar

  • 2026-11-01 | spec-driven-development | MCP-server spec workflows resolve on the primary-artifact rule; mcp-probe-kit admitted at 41 stars as the protocol-native kind with zero coverage stated | the converge gate’s artifact is a spec-plus-evidence chain, not tracker state, and the admission rides the Mycelium no-membership-floor precedent rather than the thin-tail band

  • 2026-11-01 | spec-driven-development | declarative UI screen specs consumed by a design renderer’s layout engine are design-surface tooling, not spec-driven development | DESIGN:OS Studio (jangtrinh/design-os-studio, 81 stars, PolyForm Shield, zero coverage) rejected on it and reported for surfaces

  • 2026-11-01 | task-management | a repository’s own pricing claim defers to the vendor’s live article when the two disagree | automaker’s README still advertises the Claude Agent SDK credit pool effective 2026-06-15 that Anthropic’s article (updated 2026-10-07) records as paused, with subscription limits still applying

  • 2026-11-01 | task-management | the dedicated-list check found no credible task-management-specific list again, and GitHub topic-page incidental candidates route by primary artifact | builderz-labs/mission-control (control-plane kind) and Chorus-AIDLC/Chorus (harness kind) surfaced in the topics page and routed out of the category

  • 2026-11-01 | harnesses | zcode desktop release versions read from GitHub releases, not the JS-rendered product site | the 2026-11-01 rotation could no longer extract an installer version from zcode.z.ai by direct fetch, so the note cites GitHub releases (v3.17.1, published October 30)

  • 2026-11-01 | cross-category | a repository-level pushed_at can move on side-branch pushes while the default branch sits dormant, so a pushed date read as a liveness signal checks the default branch’s latest commit | hermes-gate’s 2026-10-29 push was a dependabot bump with the default branch quiet since 2026-09-29; kiln, kitaru, and ReviewBench showed the same pattern this run

  • 2026-11-01 | evaluation-review | pypi.org project pages serve a bot challenge to plain fetchers, so version and publish-date claims ground through the pypi.org/pypi//json endpoint | hermes-gate’s cited project page challenge-blocked the re-fetch while the JSON endpoint returned 0.1.7 (2026-09-19); the browser-canonical URL stays cited

  • 2026-11-01 | executions | rivet-dev/sandbox-agent URLs are the rivet-agents member repository itself (renamed upstream; the path 301s to rivet-dev/agents) | resolves the 2026-11-01 handoff without a new note and stops future source listings of the old name from re-surfacing it

  • 2026-11-01 | assistant-runtimes | accomplish-ai/coworker, the kaushikb11 list’s “Openwork” row, rejected as a dead tombstone and confirmed not the renamed fayazara/coworkers rejection | the coworkers repo still exists unchanged (358 stars, pushed 2026-11-01), and the coworker repo holds only a “no longer supported” README, no code, 10,824 vestigial stars, dormant since 2026-08-13

  • 2026-11-01 | automated-research | ARIS’s mainline skill count reads as skills/ directories carrying SKILL.md, not the raw directory count | the raw listing holds 91 directories (skill packs and shared-references bundles included) while 87 carry SKILL.md, so reading 91 as a correction of the recorded 87 would be wrong

  • 2026-11-01 | code-review | the vendor-run kodustech/awesome-ai-code-review list stays unwired as a mapped source; the category stays on broad sources | staleness re-confirmed 2026-11-01 (pushed_at still 2026-06-28) and a repository search for dedicated lists surfaced only single tools (largest 257 stars, no curated list); evidence update for open q21

  • 2026-11-01 | control-planes | builderz-labs/mission-control admitted as the fourth live plane on the task-management routing | the Aegis approval record on done plus the opt-in dispatch sandbox flags govern the task lifecycle, which the mission-control dashboards-and-tails family fails; the changelog’s own no-policy-engine line is the stated boundary, reopen only on a policy engine shipping

  • 2026-11-01 | surfaces | the ai-for-developers list’s two long-unresolved surfaces candidates reject on the thin-record tail | BBARIT (binary distribution ended 2026-10-18, pointer repo at 40 stars over a 13-star source drop, zero coverage) and Bodega One Code (closed source, distribution repos at 47 and 2 stars, licensing undecided, zero coverage)

  • 2026-11-01 | orchestration | the 2026-11-01 topic-page ADE wave admits six members on the standalone-ADE and terminal-family lines with the missing-footprint signal stated per note | agtx, Cezar, Chorus, Happier, nodeterm, and termio admitted from the github.com/topics/ade and topics/parallel-agents sighting; coverage is zero (happier, nodeterm, agtx, termio) or launch-echo (Chorus 3 points, Cezar 15 points)

  • 2026-11-01 | orchestration | single-harness SDKs whose kernel runs its own agent loop with model routing are not multi-agent orchestration regardless of topic-page placement | loushang (zhnt/loushang, 1,939 stars) rejected, no parallel/multi-agent/worktree surface, team workflows a roadmap item

  • 2026-11-01 | orchestration | topic-page candidates below the wave’s own zero-coverage floor (termio at 751 stars) reject on the below-band line while the tier question is open | Pane 595, vicoa 523, wmux 440, agentbox 336, mulmoterminal 256, operator-oss 208, rove 181, tlbx 116, agetor 110, and cleancode 108 rejected 2026-11-01; one tier ruling re-resolves the tail in a single pass

  • 2026-11-01 | orchestration | kyrolabs/awesome-ade is a credible dedicated mapped source for the category (25 stars, 18 member entries, actively pushed, curation depth above the admitted varun1505 bar) | wiring and its back-catalog walk are owner calls, queued 2026-11-01

  • 2026-11-01 | spec-driven-development | compressed spec plugins for one harness are SDD kind, but the caveman-family personal-bundle hold applies | cavekit (JuliusBrussee/cavekit, 1,156 stars) routed report-only from orchestration, frozen August 2026, development moved to caveman

  • 2026-11-01 | cross-category | portofcontext/pctx closed out of scope with no claiming category after four kind declines (sandboxing, protocols, executions, hybrid-execution) | the Code-Mode tool-execution substrate question queued for the owner on the heretic precedent

  • 2026-11-03 | cross-category | the andyrewlee 2026-11-02 additions resolve with two admissions routed and three rejections | kanzo (921 stars, board plus MCP) routed to task-management, Cloudroom (1,311 stars, GUI built on the bb member) routed to orchestration on the standalone-ADE line, Friday (29), Agora (5), and TaskHandoff (5) rejected on the thin-record below-band line while the tier question is open

  • 2026-11-03 | cross-category | the 2026-11-01 protocols handoff resolves one admission and one rejection | Personal Agent Protocol (Sierra and Meta with Genesys, Instinct, Rocket, Shopify, Stripe, Walmart; HN 9 points) clears the bar with the unshipped v0.1 spec as its Status fact and beri.net as the critical source; toqprotocol/toq (6 stars, pushed 2026-03-18, 0-comment Show HN) fails on the thin dormant record, reopen on traction or coverage

  • 2026-11-03 | cross-category | reading and book tools reject on the no-agent-story scope line regardless of stars | tudorms/BookScan (2 stars, dormant) and bookorbit/bookorbit (6,104 stars) serve readers, not engineers working with agents; the browser-automation precedent holds at scale

  • 2026-11-03 | cross-category | the matrix contract’s header width means cell-count parity, not character width; sourced cells are never truncated to fit a column’s padding, a longer cell leaves its row wider | the retrieval worker’s 2026-11-03 ruling request; every recent commit already keeps each row’s own width stable

  • 2026-11-03 | cross-category | jackjin1997/markindex held out of retrieval on the thin-record tail | 1 star, 11 days old, vectorless Markdown retrieval MCP server surfaced by the protocols source; reopen on traction

  • 2026-11-03 | section-root | DeepSeek’s Search-and-Fetch pricing notice kept its free-of-charge wording through the 2026-11-03 read, three days past its stated 2026-11-01 start; the guide and matrix sentences already carry the dated clock, so no cell moved | pricing page fetched 2026-11-03

  • 2026-11-03 | section-root | the Claude Code context-window doc now organizes compaction survival per mechanism and names idle compaction (a session left idle compacts on its own) | docs fetched 2026-11-03

  • 2026-11-03 | retrieval | Danielskry/Awesome-RAG stays a mapped source: quiet since 2026-10-06 but reachable with entry overlap intact, quiet alone does not meet the staleness bar | the 2026-11-03 handoff re-check fetched the API, commit history, and README

Standing family rejections
#

  • 2026-10-12 (ruled 2026-10-07) | orchestration | single-vendor agent-building SDKs rejected as SDKs, not multi-agent orchestration systems | openai-agents-python, smolagents, adk-python, semantic-kernel, strands, claude-agent-sdk; microsoft/agent-framework and deepagents held under it, queue q#7 open
  • 2026-10-18 (ruled 2026-10-07) | orchestration | mission-control dashboards and tails rejected against the absorbed canonical family | claudecodeui, open-multi-agent, background-agents, RunMaestro, agent-office, clodfarm, zeron, OpenAgents Workspace, Ignite, Solenta, Agent Console, UpComputer held; openai/symphony the flagged carve-out, queue q#12 open
  • 2026-10-14 (ruled 2026-10-07) | assistant-runtimes | the absorbed runtime family holds: cowork desktops and claw successors resolve into existing notes, no new columns | rowboat, leon, cloudflare-os, nullclaw, LobsterAI, jev-chat-jarvis, openfang, xiryn, orkas, openpaw held; openhuman scale call open, queue q#13
  • 2026-10-12 | orchestration | the 2023 dormant wave of agent repositories stays rejected | re-confirmed with fresh counts in the kaushikb11/awesome-llm-agents scan
  • 2026-09-27 | cross-category | browser automation stays out of scope and no browser-agents category opens; final, not open to re-review | owner ruling on the browser-use question (116k stars declined)
  • 2026-10-13 (ruled 2026-10-07) | retrieval | data infrastructure, document parsers, vector stores, and crawlers stay out of the category | pathway, CocoIndex, Omnigraph on the data-infrastructure precedent; Crawl4AI on the Firecrawl crawler rejection; DSPy under the framework family; the parser slot (kreuzberg, opendataloader-pdf, marker family) held, queue q#17 open
  • 2026-10-13 (ruled 2026-09-07) | skills | the standing exclusion covers single skills, packs, community waves, and vendor in-product directories; scope is formats, registries, and marketplaces, plus optimizer and discovery tooling | scale records K-Dense 47.8k, rea.tools 109.9k, mattpocock/skills 286k escalate via queue q#11 and q#20
  • 2026-10-13 | assistant-runtimes | the rafska chat-UI set (LM Studio, jan, ChatBox, text-generation-webui, SillyTavern, Page Assist) finally rejected on category kind | trigger recorded, see re-review triggers
  • 2026-10-19 (ruled 2026-09-29) | orchestration | the AMO thin-record tail stays rejected | Wave, Culture, SwarmClaw, NarraNexus, Bunkhouse, handoff, Companies.sh, zachealy1/orchestrator, RustyClaw held on the handoff thin-record profile; the 2026-10-19 AMO wave adds STARK OS, AIWorks, Coterie, SpaceApp, Alchemist Coder, SuperAgent, Sqwakie, and Brow
  • 2026-10-19 (ruled 2026-10-07) | cross-category | personal bundles and viral novelty repositories stay unprofiled pending the owner’s ruling | gstack, caveman, Agent-Reach logged out-of-scope; ECC (278k) and Cubicle Crew (29 stars, pixel-art agent-office visualization) held on the same question, queue q#9 open
  • 2026-10-12 (ruled 2026-09-27) | cross-category | the owner’s own tools stay out at zero third-party adoption | llm-conversations-viewer, tps-viewer, TomzxCode/agentic; owner-tools exception asked in queue q#24
  • 2026-10-07 | cross-category | model-modification tooling fits no category and stays out pending the owner’s ruling | p-e-w/heretic (33.8k) logged out-of-scope, queue q#8 open
  • 2026-10-18 (ruled 2026-10-06) | control-planes | closed commercial governance products stay out | Autoheal precedent holds the awesome-ai-governance closed tier at zero admissions; Aident Loadout (aident.ai hosted MCP credential vault, no public repository) rejected on it 2026-10-18, also not an evaluation instrument (eval worker concurrence)
  • 2026-10-19 | control-planes | the awesome-ai-governance open-source thin tail stays below the admission bar | conarium, verax, cedulon, quilzo, sunglasses, and Mandare (0 stars, 3 weeks old) at 0-10 stars each with no coverage, reopen on traction or independent coverage; arifOS resolved above the bar 2026-10-16; Ironheights (0 stars, a skill scanner that tests rather than governs) added 2026-10-19 on the governs-something ruling
  • 2026-10-12 (ruled 2026-09-21) | hybrid-execution | applications, runtime bridges, and name-alike tooling on a category member stay ecosystem evidence inside its note | jev-ultrafast folded into jev; jev-chat, vllm-jev, LLM2Jev held
  • 2026-09-21 (ruled 2026-09-18) | skills | session-portability tools are out of the skills scope | Skillsync rejected by four category runs
  • 2026-10-19 | cross-category | terminal-output theming mods for member tools are appearance edits with no category kind | prismantis (134 stars) rejected on it; the claude-code list’s new Mods section is where future additions will land
  • 2026-10-21 | memory | the TeleAI list’s thin-tail and closed-hosted new entries stay below the bar | Fraise (8 stars vs the 294 floor, no persistence, 1-point self-published Show HN), MemorySync (closed hosted-only API, no open engine, 22-point self-published Show HN), MoobFS (2 stars, research filesystem, self-declared not for production), and LLMSlim (17 stars, zero coverage) rejected, reopen on traction, coverage, or an OSS engine
  • 2026-10-21 | spec-driven-development | trigger-and-schedule workflow automation is not spec-driven regardless of the list section it sits in | frink’s primary artifact is a triggered agent workflow, reported to orchestration
  • 2026-10-21 | control-planes | Praesidia joins the closed tier | proprietary platform, design-partner-only SDKs, demo-gated pricing, Autoheal precedent
  • 2026-10-21 | assistant-runtimes | Orkas-VideoStudio resolves outside the category, the absorbed-family shelf keeps Orkas and its extracted toolkit is skills-kind | README states it is the extracted video agent of the absorbed Orkas desktop, agent-agnostic install into coding agents; routing flagged to skills
  • 2026-10-21 | assistant-runtimes | openhuman stays held on the absorbed-family shelf pending q13 | 42.2k stars is a tenth of OpenClaw-class so the scale trigger as framed has not fired; evidence gathered for the owner
  • 2026-10-21 | harnesses | generalist enterprise-automation agent frameworks are not harnesses-kind | cuga-project/cuga-agent rejected on it (web/API surfaces, AppWorld/WebArena benchmarks, no repository loop), extending the 2026-10-12 deepagents ruling; the shelf question is queued for the owner
  • 2026-10-21 | cross-category | vendor field reports about a member tool fold into that member’s note as ecosystem evidence | the WarpBench report folds into warp; q#10 (vendor blog as voice) untouched
  • 2026-10-21 | software-factory | factory-kind candidates dormant 6+ months reject on the CodeMachine precedent without re-deriving coverage | Antfarm (2,495 stars, dormant since 2026-02-26) resolved on it; the list’s own unmaintained tag is the matching signal
  • 2026-10-21 | session-analytics | Aident Loadout stays out of session-analytics | an actions-and-credentials gateway with an audit feature, no session recording, indexing, cost, or live-observation cell; the 2026-10-18 control-planes closed-tier rejection stands
  • 2026-10-21 | skills | Orkas-VideoStudio stays out beside its held Orkas desktop sibling | the vendor-sibling fold and the standing pack exclusion both apply to the extracted video-agent toolkit; resolved at coordinator level from the assistant-runtimes worker’s routing
  • 2026-10-21 | orchestration | worktree-manager CLIs slot in orchestration beside Worktrunk, not surfaces or sandboxing | gwq admitted on the terminal-family precedent (worktrees are not isolation boundaries), closing the software-factory and scanner routing flags
  • 2026-10-21 | sandboxing | a repository rename resolves against the prior name’s history only when that name was itself a resolved candidate; a prose mention is not a resolution | the loom-to-Overalls handoff: loom was only a mention in the geoffrey-huntley note, so overalls was judged fresh and admitted (1,557 stars, Firecracker microVM platform, research preview 2026-11-03)
  • 2026-10-23 | cross-category | Mods-section additions in the claude-code list resolve as single-mod kinds under the skills exclusion | review-inbox (2 stars) the first confirmed instance, Molecular Animation (2 stars) out of scope on top of it
  • 2026-10-24 | code-review | the vendor-run kodustech/awesome-ai-code-review list stays unwired as a mapped source; the category stays on broad sources | stale since 2026-06-28, HEAD carries dead, 404, and 1-star entries; evidence recorded for the open q21, its one live entrant Nayjest/Gito admitted 2026-10-24
  • 2026-10-24 | memory | the TeleAI list’s 2026-10-21 to 2026-10-23 additions resolve on standing families | TriMem, MemRec, BetaMem, and EvoKernel are paper-section entries rather than tools, Monolithos is closed-hosted with no open engine, Big Agentic Brain (18 stars) and agent-memory-recall (0 stars) sit below the 294-star floor
  • 2026-10-24 | memory | the awesome-mcp-servers 2026-10-23 wave’s memory-server cluster stays below the bar | open-memex (6 stars), MemCore (1), swarmmemo (2), IHMT-MEMORY (38), compound-memory (1) on the 294-star thin-tail floor, synapse-memory and alethech unverifiable (404); reopen on traction or coverage
  • 2026-10-24 | orchestration | standalone ADEs and multi-machine command centers admit on the dmux/claude-squad/Conductor/Superset line even with vendor-hosted distribution, with coverage quality recorded as the caution | OpenHarness (autonomous-ai/openharness, 1,358 stars as of 2026-10-24) admitted with launch-echo-only coverage and vendor-only benchmarks; hosted-relay accounts and the name collision noted
  • 2026-10-24 | orchestration | evidence-carrying coding agents with per-task worktree workers are orchestration kind, judged past the small-member floor with the missing-footprint signal stated | Swarm Orchestrator (moonrunnerkc/swarm-orchestrator, 112 stars, 672 npm downloads a month, one contributor) admitted; anything below v13 of the package name is a different program and stays rejected elsewhere
  • 2026-10-24 | orchestration | Ivy Tendril (Ivy-Interactive/Ivy-Tendril) admitted on cross-category carry | plan-to-PR supervision of many CLI agents in parallel worktrees with human-gated review; surfaces rejected (no editor surface), software-factory rejected (loop is human-gated, not self-owning)

Owner questions governing scope
#

(q#N is the Nth ? for tom: bullet in queue.md, in file order.)

  • q#4 (2026-10-06) | retrieval | should Haystack become a third framework column, or does the two-framework scope stand
  • q#5 (2026-10-06) | people-and-publications | should METR be profiled as a voice after its tracker rejection
  • q#6 (2026-10-07) | evaluation-review | add a third observability column (Opik), a second eval-framework column (promptfoo), or stand at current structure; AgentOps rides the same question
  • q#7 (2026-10-07) | orchestration | should any single-vendor agent-building SDK get an orchestration column
  • q#8 (2026-10-07) | cross-category | should model-modification tooling (heretic) get a home in the section
  • q#9 (2026-10-07) | cross-category | profile the mega-viral trio (gstack, caveman, Agent-Reach) or do personal bundles and viral novelty stay out
  • q#10 (2026-10-07) | people-and-publications | is a vendor engineering blog admissible as a voice
  • q#11 (2026-10-07) | skills | do the skills-scale records (K-Dense 47.8k, diagram-design 44.3k, compound-engineering-plugin 25.4k, claude-plugins-official 37.5k) change the pack policy
  • q#12 (2026-10-07) | orchestration | should openai/symphony be profiled despite the mission-control tail rejection
  • q#13 (2026-10-10) | assistant-runtimes | does openhuman (41.8k) break the absorbed-family rejection at OpenClaw-class scale
  • q#14 (2026-10-10) | surfaces | which shelf (or rejection) holds for OpenBitFun, a fourth Cowork desktop candidate
  • q#15 (2026-10-10) | cross-category | wire the four first dedicated lists (awesome-jev, awesome-software-factories, awesome-autoresearch, awesome-harness-engineering) into the category-source list
  • q#16 (2026-10-10) | skills | admit the generator kind (Skill_Seekers) or hold the formats, registries, marketplaces scope
  • q#17 (2026-10-10) | retrieval | do CocoIndex, Crawl4AI, or DSPy deserve their own columns, or do the family scopes stand
  • q#19 (2026-10-11) | evaluation-review | add a security-evaluation family (AgentFence, AgentDojo, garak, DeepTeam) or hold the output-judging scope
  • q#20 (2026-10-11) | skills | do the grown pack-scale records (addyosmani/agent-skills 105k, mattpocock/skills 286k, morluto/rea 109.9k) change the pack policy
  • q#21 (2026-10-11) | code-review | wire the vendor-run, stale kodustech/awesome-ai-code-review list or stay on broad sources
  • q#22 (2026-10-11) | harnesses | if a harnesses list gets wired, is the target stale walkinglabs or active ai-boost (name collision)
  • q#24 (2026-10-12) | task-management | owner-tools exception (disclosed ownership, bar still applied) or the TomzxCode/agentic rejection stands
  • q#25 (2026-10-14) | cross-category | leave the browser-automation family (browser-use and peers) out of scope or open a browser-automation family
  • q#26 (2026-10-14) | automated-research | wire either, both, or neither of the two autoresearch lists (webfuse-com, WecoAI)
  • q#27 (2026-10-16) | executions | wire serenakeyitan/awesome-agent-loops (214 stars, stale since 2026-06-22, recipe collection) into the category-source list or stay on broad sources
  • q#28 (2026-10-16) | trackers-and-leaderboards | wire onejune2018/Awesome-LLM-Eval (659 stars, stale since 2025-11-24, academic-eval focus) into the category-source list or stay on broad sources

Re-review triggers
#

  • 2026-10-17 (ruled 2026-09-06) | memory | okf-agent-memory trigger fired and resolved: admitted at 814 stars on both trigger arms, three independent citations in the 392-point documentation-over-recall thread plus a second implementation (fellowgeek/mcp-memory) | supersedes the 2026-10-07 re-rejection
  • 2026-10-12 | task-management | TomzxCode/agentic reopens on independent adoption or third-party coverage | rejected on the citation bar, owner’s own tool
  • 2026-10-13 | assistant-runtimes | the rafska chat-UI set reopens on renewed adoption (inferred) | the final rejection carries a recorded trigger the log does not name
  • 2026-09-24 | people-and-publications | the humanlayer (dex-horthy) rejection is closed: overridden by owner admission, no trigger remains | the 2026-09-08 no-sustained-publication rejection was overridden 2026-09-24
  • 2026-08-30 | cross-category | concord and huzzah reopen only on new independent coverage | closest below-bar calls of the 2026-08-30 candidate pile
  • 2026-10-18 (ruled 2026-09-21) | cross-category | coverage-triggered holds reopen on independent coverage; brooks-lint’s trigger fired 2026-10-18 on the Agent Skills Hub eval (5/8 defects, flagged security blind spot) and resolved with the rejection standing (one mixed third-party measurement, planning-stage kind, no findable primary footprint) | thruwire/foreman, magpie, golive-skill, traycer remain open on coverage
  • 2026-09-25 | hybrid-execution | MoJev reopens on independent traction | uncategorized at 29 stars, self-published
  • 2026-10-15 | hybrid-execution | TypeChat reopens on a first tagged release or npm publication | 2026-10-15 re-confirmed rejection (zero releases ever), but the 10-13 push added three named TypeScript-team maintainer commits and a Programs blog post
  • 2026-10-16 | assistant-runtimes | OpenAgent reopens on independent coverage or crossing the member floor | 5,698 stars but below the category floor (smallest member Octop about 9.2k), zero independent coverage, keyword-stuffed topics
  • 2026-10-16 | software-factory | uzi (vtmocanu) reopens on independent coverage or a stable release | 91 stars, MIT, alpha, nightly RC train, only a self-authored write-up, below the Orbi precedent (213 stars rejected 2026-10-11)
  • 2026-10-25 (ruled 2026-09-25) | trackers-and-leaderboards | bestvaluemodel trigger fired and resolved: admitted on the license, a month of six-hourly refreshes with daily data diffs demonstrated | supersedes the reopen line; the fully-derivative-of-Artificial-Analysis caveat stands in the note
  • 2026-09-25 | trackers-and-leaderboards | llm-list.com reopens on independent coverage | fully derivative of models.dev, OpenRouter, Arena, Artificial Analysis, and Hugging Face
  • 2026-10-24 (ruled 2026-09-27) | hybrid-execution | drawgent reopens on a primary mirror | license and stars unverifiable from tangled.org-only hosting; re-checked 2026-10-24, GitHub repository search returns zero drawgent repositories and tangled.org/drawgent 404s, unmet
  • 2026-10-24 (ruled 2026-09-27) | harnesses | drawgent reopens on a primary mirror | license and stars unverifiable from tangled.org-only hosting; re-checked 2026-10-24, no mirror exists, unmet
  • 2026-10-16 | memory | Mnemoverse reopens on traction or independent coverage | 27 stars, 4 forks, hosted-first thin record far below the category floor (smallest member Emulo at 294)
  • 2026-09-27 | skills | dotskills reopens on sustained traction | curated pack overlapping the Agent-Native column
  • 2026-10-27 (ruled 2026-09-27) | executions | octopool reopens on traction; re-checked 2026-10-27 at 140 stars with zero independent coverage, the bar is not met (traction means adoption evidence, not raw stars in a hyped ecosystem) | weeks-old ecosystem utility, thin docs
  • 2026-10-27 | evaluation-review | transcript-grading batteries with zero adoption stay below the admission floor | forevercrab321-svg/leevar-battery (0 stars, 0 forks, one release, optional hosted scan by the same authors) rejected; HermesGate’s 3-stars-plus-docs-plus-releases admission is the floor to beat; reopen on any third-party signal
  • 2026-09-27 | memory | library-mcp reopens if a writeup ships | dormant five months, thin documentation