jcode: added note (Harnesses) with matrix column and index entry; recorded the resource-efficiency thesis, the embedding memory graph, native swarm, self-dev mode, 30-plus providers, $10/month hosted inference, 18,796 stars, and the thin independent footprint [glm-5.3-flash]
harness-feature-matrix extended from fourteen to fifteen columns (jcode between goose and Junie); agentic-coding-tools-landscape, the group lists, the references, and _index.md updated [glm-5.3-flash]
harness-feature-matrix: added a Company row naming each maker (aider none, Amp Frontier, Anthropic, Cline Bot, OpenAI, Charm, Vercel, Google, AAIF for goose, Solo Systems, JetBrains, Anaconda for Kilo Code, Anomaly, All Hands AI, Alibaba), with the reading prose framing it as context, not a feature axis [glm-5.3-flash]
Added notes deepseek-harness, pi (harnesses), graphify, rtk, qmd (context engines), claude-mem (memory), skillopt (skills), paseo, superset, worktrunk (orchestration), ouroboros (software factory), and agentsview (session analytics) with matrix columns and index entries; agentsview seeded the new Session analytics category with a single-column matrix scaffold [glm-5.3-flash]
Matrices extended: harness to seventeen columns, context engines to seven, memory to six, skills to five, orchestration to eleven, software factory to four [glm-5.3-flash]
AGENTS.md gained the rule that matrix columns sort alphabetically by member title, case-insensitive, with no privileged position for founders, roots, or baseline conventions, re-sorted with any membership change; eight of the sixteen matrices deviated (assistant-runtimes, evaluation-review, memory, people-and-publications, retrieval, sandboxing, software-factory, spec-driven-development) [glm-5.3-flash]
Re-sorted the eight non-compliant matrices, reordering cells to match the sorted headers, ordering only, dropping the root-first, Files-first, kind-order, and founding-member-first conventions; harness-feature-matrix corrected two column-ordinal references in prose [glm-5.3-flash]
people-and-publications, retrieval, and spec-driven-development carried a doubled closing bracket on their tags line from the re-sort’s tag insertion; the invalid YAML failed the site build, so the deployment kept serving the pre-sort build, and the three lines plus the double-space table padding across the eight matrices were repaired [glm-5.3-flash]
Banned-term sweep caught five “shape” slips the earlier sweeps missed, in fluent, letta, ouroboros, paseo, and phoenix, all reworded with meaning unchanged; fluent and letta took updated 2026-08-30 and the flash tag [glm-5.3-flash]
Context engines deep refresh: repomix refreshed to 28,122 stars, 1,508 forks, 4,476 commits, 373,996 npm downloads, and v1.18.0; augment-code extended its blog cadence to August 28, 2026 and re-dated the no-free-tier claim [glm-5.3-flash]
semble: added note (Context engines) with a matrix column and index entry, recording the sub-second CPU indexing thesis and the two-maintainer pre-1.0 risks; Mantic.sh kept out (554 stars, stalled since July 13); no self-directed essay [glm-5.3-flash]
Owner directive: the research index line “refreshed one category at a time, stalest first” became refresh-all-in-parallel and the rotation was retired; AGENTS.md step 3, _index.md, and methodology.md were updated to match [glm-5.3-flash]
Corrections: claude-code-hooks “5k+ open issues” now about 15k; plannotator rewritten to 135 contributors and 859 founder commits; eigent’s “200-plus built-in MCP tools” dropped; task-master gained the missed Hamster Enterprise tier ($200/creator/month); picoclaw corrected to v0.3.1 of 2026-07-03; deeplearning-ai lost an unverifiable quote and a dead reference [glm-5.3-flash]
Seeded the Code review category with five members: coderabbit, qodo, and ellipsis written fresh, plus greptile and open-code-review moved in on category-fit grounds, with matrix columns and index entries [glm-5.3-flash]
coderabbit recorded the funding ladder ($16M/$60M/$143M, August 2026 Series C at a $1.5B valuation), 17k customers, 2M-plus weekly reviews, the Kudelski exploit chain, and the free public-repo tier; qodo recorded the open-core split (MIT PR-Agent at 12,769 stars), $51M funding, the CodiumAI rename, $0.012 credit pricing, and the second Kudelski chain; ellipsis was set to pivoted on the July 2026 Agent Cloud launch [glm-5.3-flash]
code-review-feature-matrix created with five columns and ten rows; Greptile removed from context engines (8 to 7) and OpenCodeReview from evaluation review (5 to 4); _index.md gained the category and the long-tail candidates were recorded [glm-5.3-flash]
Owner directive: every entrant candidate resolves in the run that surfaces it, as a note or an explicit logged rejection; a candidate carried as “pending” is a defect [glm-5.3-flash]
Resolved the candidate pile to zero: 37 candidates, 11 accepted as notes, 26 rejected with fresh evidence [glm-5.3-flash]
Matrices extended (harness to 23 columns, orchestration to 12, protocols to 5, code review to 8, untraceable cells marked “?”); _index.md and agentic-coding-tools-landscape updated for the eleven notes [glm-5.3-flash]