Work items the owner has requested, taken in order, highest first.
Every item is standing: none is ever closed, and every section keeps being built out and re-verified indefinitely.
The owner edits this file to steer the section; agents append ? for tom: bullets and work the items, and never mark one done, strike it through, or remove it.
- Build the research index (standing) - create research notes per the AGENTS.md format, one category at a time, each category seeded with the tools practitioners actually meet. Seed categories, in rough order:
- Harnesses: Claude Code, Codex, Gemini CLI/Antigravity, OpenCode, aider, Crush, Amp, Junie.
- Surfaces: Cursor, VS Code + Copilot, Windsurf, JetBrains/Junie, Zed.
- Orchestration: parallel-agent dashboards and worktree managers (Conductor, dmux, Emdash, and neighbors).
- Protocols: MCP, ACP, AGENTS.md, A2A.
- Context engines: code search and context packing for agents.
- Skills: skill formats, registries, and marketplaces.
- Retrieval: RAG patterns and tooling for code and knowledge bases.
- Memory: agent memory systems, from file-based notes to knowledge graphs and session archives.
- Executions: scheduled and event-based agent runs (cron-style tasks, triggers, webhooks).
- Hybrid execution: deterministic components orchestrating LLM steps for intelligence and structured outputs (tool calling, schema/JSON outputs, validation and retry loops).
- People and publications: the people and websites shaping the domain (Karpathy, swyx/Latent Space, Willison, Yegge, Orosz/Pragmatic Engineer).
Standing work already under way (still open; keep updating and extending each indefinitely):
- Agentic coding tools landscape - started 2026-08-22, published as agentic-coding-tools-landscape; it is the essay layer over the harness/surface/orchestration notes, so keep cross-linking new notes and re-verifying it indefinitely.
- Model selection for coding tasks - started 2026-08-24, published as model-selection-for-coding-tasks; keep folding in new models and price moves indefinitely.
- Context management patterns - started 2026-08-24, published as context-management-patterns; keep re-grounding its patterns in new harness documentation indefinitely.
- People and publications category - seeded 2026-08-29, with the companion people-and-publications-feature-matrix; keep scanning for and adding new voices and publications indefinitely, not only when prompted.
? for tom: The 2026-10-01 banned-term sweep reworded “real” section headings in the main corpus but left this section’s roughly 330 “real” uses (mostly “real-world” compounds and contrastive uses like “adoption is real, stability is not”) in place, and the daily verification has been treating them as the most-appropriate-term exceptions since. Should the section get the same rewording treatment, or is the corpus-only scope the intent? Rewording at this scale needs its own directed run with Changes bullets per touched note, so I have not done it inside a daily refresh.
? for tom: The sandboxing refresh flagged that E2B, the hosted sandbox API the category’s notes already cite as a standing comparison baseline, has never been profiled as a note of its own. Add it to the sandboxing category (it would clear the citation bar easily), or does it stay comparison-only?
? for tom: Following up on the “real” question above: the skills-library banned-terms list now also bans “real” (and “load-bearing”) as of 2026-10-06, while this section’s AGENTS.md still lists only shape/honest/load bearing and the roughly 334 retained uses remain under the documented-exceptions treatment. Confirm and a directed run will reword them with per-note Changes bullets.
? for tom: The retrieval entrant scan rejected Haystack (26.7k stars, active) because the category’s founding scope carries exactly two frameworks and a third framework column is a scope call, not an entrant signal. Should retrieval add Haystack as a third framework column, or does the two-framework scope stand?
? for tom: The trackers scan rejected METR’s time-horizon series as a tracker (a single trend series, not a leaderboard) but flagged it as a possible people-and-publications voice (an evals nonprofit publishing original measurement research). Should METR be profiled in people and publications?
? for tom: The evaluation awesome-list cross-check surfaced Opik (comet-ml/opik, 22.4k stars, Apache-2.0, active) as an uncovered observability-and-eval platform between the category’s two existing columns in adoption (Langfuse 35.5k, Phoenix 11.7k), with promptfoo (25.8k) as a second eval framework next to deepeval and AgentOps (5.9k, quiet since June) the weaker sibling. I rejected all three as scope calls under the retrieval Haystack precedent. Should evaluation-review add a third observability column (Opik), a second eval-framework column (promptfoo), or stand at the current structure?
? for tom: The orchestration framework family now covers the multi-agent systems (CAMEL, AG2, AutoGen, AutoGPT, CrewAI, Agno, Dify, Mastra, LangGraph, and more), and the cross-category scan surfaced the single-vendor agent-building SDKs (openai-agents-python 29.9k, smolagents 29.7k, google adk-python 21.7k, semantic-kernel 28.6k, strands 8.7k, claude-agent-sdk 8.2k). I rejected them as SDKs for building agents rather than multi-agent orchestration systems, a family rejection recorded in the log. Should any of them get orchestration columns anyway?
? for tom: Today’s one new star is p-e-w/heretic (33.8k stars, fully automatic censorship removal for model weights). It fits none of the 23 categories (it modifies models rather than providing access or running agents), so it is logged as out-of-scope. Should model-modification tooling have a home in the section, or does the rejection stand?
? for tom: Three mega-viral repositories surfaced with no clean category: garrytan/gstack (135.6k stars, Garry Tan’s personal Claude Code setup as 23 opinionated agent tools), JuliusBrussee/caveman (110.3k, a skill-plus-proxy claiming 65 percent token cuts by removing articles), and Panniantong/Agent-Reach (92.9k, a CLI giving agents read and search over Twitter, Reddit, YouTube, GitHub, and Bilibili). Profile any of them (skills, context engines, or retrieval would be the nearest homes), or do personal bundles and viral novelty stay out?
? for tom: The people-and-publications scan keeps passing over vendor engineering blogs (Anthropic Engineering, OpenAI’s harness-engineering write-ups) because the category has only admitted independent or practitioner-owned publications. Should a vendor engineering blog be admissible as a voice, or does that stay out of scope?
? for tom: The skills scan surfaced scale records above every prior pack rejection: K-Dense-AI/scientific-agent-skills (47.8k stars, 177 validated science skills), cathrynlavery/diagram-design (44.3k), EveryInc/compound-engineering-plugin (25.4k), and anthropics/claude-plugins-official (37.5k, the vendor-official plugin directory). The standing exclusion covers single skills and packs and vendor in-product directories. Do skills-scale records like these change the policy, or does the formats/registries/marketplaces scope stand?
? for tom: openai/symphony (27.6k stars, Apache-2.0, quiet since September 15) turns project work into isolated autonomous implementation runs, squarely the orchestration category’s subject, and I logged it with the mission-control tail as a family rejection rather than writing it amid today’s 45 notes. Want it profiled next run?
? for tom: The 2026-10-10 scan surfaced tinyhumansai/openhuman (41,802 stars, pushed 2026-10-11, added to both orchestrator directories this week): a Rust library plus desktop harness with built-in memory, token compression, and tool search. It is assistant-runtime-shaped and would sit 2.4x above leon (17.6k), the absorbed family’s largest named rejection, but at OpenClaw-class scale. Profile it, or does the family rejection hold at this scale?
? for tom: GCWing/OpenBitFun (2,412 stars, active, MIT) was rejected from surfaces as a general-purpose Cowork-style agent workspace whose coding support is one mode among several, which reads as assistant-runtime-shaped; but admitting it as a fourth Cowork desktop would re-litigate the standing absorbed-family rejection from the same run. Which shelf (or rejection) stands?
? for tom: Four categories acquired their first credible dedicated awesome lists this run and AGENTS.md changes are yours: yibie/awesome-jev (2,255 stars, hybrid execution), varun1505/awesome-software-factories (17 stars but on-subject and active, software factory), webfuse-com/awesome-autoresearch (2,563 stars, automated research), and walkinglabs/awesome-harness-engineering (4,368 stars, harnesses). Wire any or all into the category-source list?
? for tom: The skills scan surfaced yusufkaraaslan/Skill_Seekers (15,118 stars, active, doc-to-skill converter): a generator kind the category has never admitted, standing beside the formats, registries, and marketplaces scope. Admit the generator kind, or does the standing scope hold?
? for tom: The retrieval family boundaries turned away three scaled tools this run under standing precedents: CocoIndex (11.7k, active, Apache-2.0) under the pathway data-infrastructure rejection, Crawl4AI (85.1k, active) under the Firecrawl crawler rejection, and DSPy (38.6k, active) under the framework family of the still-open Haystack question. Do any of the three deserve their own columns, or do the family scopes stand?
? for tom: The AMO directory’s source-of-truth file (src/data/orchestrators.ts) holds roughly a dozen slugs with no README rendering (ctx, open-agent-relay, agent-coordinator, meshfleet, commonground-kernel, speakeasy-mcp-gateway, agentrq, hyper, and friends) that predate the 2026-10-07 cross-check, which walked the README rendering rather than the data file; the 10-07 “all 67 entries” count was the README’s. Want a directed run to walk the ts-only slugs once?
? for tom: The evaluation-review scan keeps resolving a security-testing cluster the category has never admitted: AgentFence (61 stars, quiet since August), AgentDojo (906 stars, ETH Zurich, quiet since June), and the stronger garak and DeepTeam that a first admission would then owe, all judging attack robustness rather than generated output. Should evaluation-review add a security-evaluation family, or does the output-judging scope stand?
? for tom: The skills pack-scale records grew again today: addyosmani/agent-skills at 105,086 stars and mattpocock/skills at 286,377 stars (both above the 47.8k K-Dense figure in the standing question), and morluto/rea (“REA Reverse Engineer Anything”, 109,906 stars, a 768-point Hacker News launch) resolved as harnesses-kind and held under the same pack exclusion with a skills routing flag. Do any of these change the pack policy, or does the formats/registries/marketplaces scope stand?
? for tom: kodustech/awesome-ai-code-review (49 stars, no license, maintained by Kodus, stale since 2026-06-28) is the only dedicated awesome list the code-review category has ever acquired, but it is vendor-run and stale. Wire it into the category-source list, or does code-review stay on the broad sources?
? for tom: The harnesses wiring question has a name collision: walkinglabs/awesome-harness-engineering (4,373 stars) has been stale since 2026-08-19, while ai-boost/awesome-harness-engineering (4,786 stars, pushed 2026-10-10) is the list you starred on 2026-04-26. If the harnesses list gets wired at all, which one is the intended target?
? for tom: The single-repository star-history chart convention (from your 2026-10-07 note) lives only in log history, not in agents/AGENTS.md, and three waves of workers missed it today before a repair pass added 18 charts. Codify it in the Article format section (one line), or retire the convention? (done 2026-10-12: codified in the Article format section at your direction.)
? for tom: Today’s only new star is TomzxCode/agentic (your own task-and-worktree manager for agent work, public 2026-10-10, 1 star, 0 forks, MIT, ACP harness runs). I rejected it as a note on the citation bar: no third-party footprint, five sources unreachable without counting the repository against itself, and profiling the section owner’s own tool at zero adoption reads like an advertisement. Re-review trigger recorded on independent adoption or third-party coverage. Want an owner-tools exception (disclosed ownership, bar still applied), or does the rejection stand?
? for tom: The browser-automation family keeps scaling past the section (browser-use 117,786 stars, stagehand 25,653, skyvern 23,190, steel-browser 7,791, ClawBench 1,026), and no category covers browser control; the 2026-10-13 scan class-rejected them with the nearest shelf (hybrid execution via cua-s1) judged a whole-family admission. Leave them out of scope, or open a browser-automation family?
? for tom: Update to the OpenBitFun shelf question above: a second same-profile candidate now exists, Orkas (Orkas-AI/Orkas, 2,188 stars, MIT, active, surfaced by the slavak source 2026-10-16), a desktop commander directing sub-agents and coding CLIs, held 2026-10-16 under the same absorbed cowork-desktop family pending your call on that shelf.
? for tom: Update to the autoresearch wiring question above: a second candidate list now exists, WecoAI/awesome-autoresearch (about 1,069 stars, use-case focused), beside webfuse-com/awesome-autoresearch (2,565). Wire either, both, or neither?
? for tom: The executions category acquired its first credible dedicated list sighting: serenakeyitan/awesome-agent-loops (214 stars, CC-BY-4.0, /loop-/goal-/schedule recipes for Claude Code and Codex), on-subject but stale since 2026-06-22 and a recipe collection rather than a tool list. Wire it into the category-source list with the staleness caveat, or leave executions on the broad sources?
? for tom: The trackers category’s nearest scaled list is onejune2018/Awesome-LLM-Eval (659 stars, MIT, eval tools, papers, and leaderboards), but it is stale since 2025-11-24 and academic-eval-focused. Wire it with the staleness caveat, or leave trackers-and-leaderboards on the broad sources?
? for tom: Update to the autoresearch wiring question above: two more dedicated lists surfaced this run, openags/Awesome-AI-Scientist-Papers (176 stars, pushed 2026-10-12, paper links) and Omni-Scientist/Awesome-AI-Scientist (109 stars, pushed 2026-10-07), both report-only while q#26 stands beside the webfuse-com and WecoAI candidates.
? for tom: The sandboxing refresh found the E2B question above already resolved on disk: E2B has held a full note and matrix column in sandboxing since 2026-10-06 (Added bullet dated in the category index). Confirm and treat q#2 as answered, or say what remains open about it.
? for tom: Update to the openhuman question above (q#13): tinyhumansai/openhuman repositioned from “personal AI super intelligence” to “the fastest, cheapest, most efficient open-source agent harness” (fleet-scale Rust framing, GPL-3.0, 42.1k stars as of 2026-10-17), and its README now carries a self-run SWE-bench comparison table against Claude Code, Codex, OpenClaw, and Hermes Agent, the vendor-benchmark pattern this section discounts; still zero independent coverage, so the absorbed-family hold stands, but the harness framing may matter to whichever shelf you pick.
? for tom: The skills-library banned-terms list also names “posture” (as of the 2026-10 list), while this section’s AGENTS.md bans only shape/honest/load bearing; the section carries two uses, both in thesis lines (claude-code-action’s “security posture” and claude-code-routines’s “default posture”). Want a directed pass to reword both with per-note Changes bullets, or does the section’s three-term list stay the section rule?
? for tom: Update to the code-review wiring question above (q#21): a second candidate now exists, zhuyansen/awesome-claude-code-review (created 2026-10-10, pushed 2026-10-17, refreshed every 8 hours, 185 categorized repos with security grades, plus an original 16-tool end-to-end eval with a public results file, but 1 star and run by the Agent Skills Hub site). It supersedes the stale kodustech list on every axis but age and starred history. Wire it (with the 1-star caveat), wire neither, or stay on broad sources?
? for tom: Update to the harnesses wiring question above (q#22): a third kind of candidate now exists, Gloriaameng/Awesome-Agent-Harness (372 stars, CC-BY-4.0, pushed 2026-10-14), the companion list to an “Agent Harness for LLM Agents” survey (168 references, 34 systems) rather than a maintained tool directory, and three more same-named awesome-harness-engineering lists (7, 2, and 4 stars) deepen the name collision. Wire the survey companion, any of the name-collision lists, or none?
? for tom: The memory category now has a second credible dedicated list: IAAR-Shanghai/Awesome-AI-Memory (1,281 stars, created 2025-12-01, pushed daily), a broad AI-memory knowledge base spanning research, frameworks, benchmarks, and practice, beside the wired TeleAI-UAGI list. Wire it as a second source, or stay single-source?
? for tom: The assistant-runtimes category acquired its first credible dedicated list: machinae/awesome-claws (506 stars, updated 2026-10-19, 34 projects, mostly the absorbed claw-family tail). Wire it as a category source (its tail resolves under the standing family rejection), or leave assistant-runtimes on the broad sources?
? for tom: IronClaw (nearai/ironclaw, 12,662 stars, pushed 2026-10-18) surfaced on that list as a claw successor held under the absorbed runtime family, but it sits above the member floor (Octop about 9.2k) and beside the openhuman scale call (q#13). Does the family rejection hold at IronClaw’s scale, or does it join the q#13 ruling?
? for tom: The context-engines category found a second credible dedicated list: yzfly/awesome-context-engineering (213 stars, created 2025-07-28, pushed 2026-10-15), beside the wired Meirtz list. Wire it as a second source, or stay single-source? (Wiring would surface roughly ten in-category candidates for resolution in that run.)
? for tom: Update to the memory second-source question above: the IAAR-Shanghai/Awesome-AI-Memory scan recommends wiring it scoped to its open-source-systems and benchmarks sections only (the 19k-line papers section is high-volume and low-yield for tool notes), which would surface roughly eight non-member candidates (langmem, memobase, MIRIX, ReMe, MemoryOS, EverMemOS, Memori, Memary) for resolution in that run.
? for tom: Today’s token-reduction pass scripted the coordinator mechanics inside agents/worklists/: worker reports now go to /tmp/opencode/agent-reports/## Pricing section the note-format order names (individuals and free publications where pricing does not apply, and the format says notes whose Pricing says pricing does not apply get no price history, not that they can drop the section). Add “does not apply” stubs to the 15 with per-note Changes bullets, or amend the section’s format note so people notes may omit Pricing?
? for tom: Route cuga-project/cuga-agent (CUGA, IBM’s enterprise generalist agent harness, 1,045 stars, Apache-2.0): rejected by harnesses as not coding-harness kind (web/API automation surfaces, AppWorld/WebArena benchmarks, no repository loop) and by control-planes as not a plane. Admit it under assistant-runtimes, hold it out of scope, or extend the harnesses charter to generalist (non-coding) agent harnesses?
? for tom: The control-planes category found a second dedicated list: systempromptio/awesome-ai-agent-governance (52 stars, pushed 2026-10-19, detailed entries), but it is vendor-run (systemprompt.io’s own products fill several sections). Wire it as a second source with the vendor-run caveat, or stay on agentrust-io/awesome-ai-governance alone?
? for tom: Update to the autoresearch wiring question above (q#26): the candidate set now also holds yibie/awesome-autoresearch (784 stars, pushed 2026-10-18, structured categories, active) and AI4Scientist/awesome-autoresearch (170 stars, well-taxed taxonomy list), beside webfuse-com (2,565), WecoAI (1,069), and the two report-only paper lists. Wire any of them?
? for tom: The task-management scan surfaced two dedicated lists owned by other categories, both report-only: kyrolabs/awesome-ade (Agentic Development Environments, created 2026-08-07, nearest homes harnesses or surfaces) and bradAGI/awesome-cli-coding-agents (1,383 stars, updated 2026-10-19, nearest home harnesses). Wire either as a mapped source, or do those categories stay on the broad sources?