diff --git a/plans/reports/research-260911-1328-agents-refresh.md b/plans/reports/research-260911-1328-agents-refresh.md new file mode 100644 index 0000000..515154f --- /dev/null +++ b/plans/reports/research-260911-1328-agents-refresh.md @@ -0,0 +1,143 @@ +# Research Report: awesome-coding-agents list refresh (additions + stale cleanup) + +**Conducted:** 2026-09-11 13:28 (+07) · **Repo:** `tiennm99/awesome-coding-agents` · **Tracked now:** 29 + +## Executive Summary + +List has a **hard gap at the top**: `deepseek-ai/deepseek-harness` (219.6k ★, DeepSeek's official agent harness, shipped Aug 2026) is untracked and would rank **#1**, above `anomalyco/opencode` (206.6k). `xai-org/grok-build` (xAI's official coding agent, 26.7k) is also missing. Both are first-party vendor agents — same tier as claude-code/codex/gemini-cli, which are tracked. + +14 repos meet the README's own inclusion criteria (coding agent itself, ≥10k ★, open source, actively maintained) and are absent. 7 tracked repos have no push in >3 months; 3 of them are dead beyond debate (bolt.new 21 mo, plandex 11 mo, trae-agent 7 mo). + +Also found one data defect: `data/agents.yml` still lists `yetone/avante.nvim` while GitHub canonical is `avante-corp/avante.nvim`, violating the file's own header rule. + +## Methodology + +- Sources: live GitHub GraphQL (all 29 tracked + 380 candidate slugs probed), 2 web searches, 2 curated landscape lists (`joylarkin/AI-Coding-Landscape`, `bradAGI/awesome-cli-coding-agents`), 17 candidate READMEs read for scope classification. +- Star/push data: live as of 2026-09-11. Staleness cutoff: `pushedAt < 2026-06-11`. +- Scope filter = README criteria verbatim: agent that **writes, edits, or reviews** code; frameworks/wrappers/observability/workspaces *for* agents excluded. + +--- + +## 1. Additions (meet all criteria, verified) + +| ★ | owner/repo | category | pushed | what it is | +|--:|---|---|---|---| +| 219.6k | `deepseek-ai/deepseek-harness` | cli | 09-11 | **DeepSeek's official agent harness (`dsh`)**, plugin architecture. Would be **#1** | +| 68.3k | `openinterpreter/openinterpreter` | cli | 09-09 | Open Interpreter, repositioned as a coding agent for open models | +| 40.9k | `Hmbown/Codewhale` | cli | 09-11 | Rust terminal coding agent (ex-deepseek-tui), 30+ providers | +| 35.5k | `esengine/DeepSeek-Reasonix` | cli | 09-11 | Go binary; CLI/TUI + desktop + VS Code over one local engine | +| 33.1k | `Gitlawb/openclaude` | cli | 09-09 | TS coding-agent CLI, multi-provider, MCP | +| 30.6k | `can1357/oh-my-pi` | cli | 09-11 | Coding agent with LSP/IDE wired in (TS+Rust) | +| 26.7k | `xai-org/grok-build` | cli | 09-09 | **xAI's official coding agent** (TUI, headless, ACP) | +| 22.2k | `alibaba/open-code-review` | cli | 09-11 | Alibaba's AI code-review agent CLI (reviews code ⇒ in scope) | +| 20.5k | `PrimeIntellect-ai/prime-agent` | cli | 09-11 | Self-improving coding/long-horizon agent harness | +| 19.5k | `1jehuang/jcode` | cli | 09-11 | Coding agent TUI, self-updating, YC-launched | +| 16.5k | `HKUDS/DeepCode` | research | 09-09 | Multi-agent code generation; TUI/desktop/web clients | +| 11.9k | `CodebuffAI/freebuff` | cli | 09-11 | Codebuff's free terminal coding agent | +| 11.3k | `MoonshotAI/kimi-cli` | cli | 09-01 | Moonshot's official Kimi Code CLI agent | +| 10.3k | `Kuberwastaken/claurst` | cli | 09-02 | Rust multi-provider terminal agent (at the 10k floor) | + +YAML patch (append to `data/agents.yml`): + +```yaml + - owner: deepseek-ai + repo: deepseek-harness + category: cli + notes: DeepSeek official harness (dsh); developer preview, breaking changes expected + - owner: openinterpreter + repo: openinterpreter + category: cli + - owner: Hmbown + repo: Codewhale + category: cli + - owner: esengine + repo: DeepSeek-Reasonix + category: cli + - owner: Gitlawb + repo: openclaude + category: cli + - owner: can1357 + repo: oh-my-pi + category: cli + - owner: xai-org + repo: grok-build + category: cli + - owner: alibaba + repo: open-code-review + category: cli + - owner: PrimeIntellect-ai + repo: prime-agent + category: cli + - owner: 1jehuang + repo: jcode + category: cli + - owner: HKUDS + repo: DeepCode + category: research + - owner: CodebuffAI + repo: freebuff + category: cli + - owner: MoonshotAI + repo: kimi-cli + category: cli + - owner: Kuberwastaken + repo: claurst + category: cli +``` + +## 2. Stale cleanup (no push in >3 months) + +| tracked repo | ★ | last push | age | archived | verdict | +|---|--:|---|--:|---|---| +| `stackblitz/bolt.new` | 16.5k | 2024-12-17 | 21 mo | no | **remove** — abandoned, `bolt.diy` fork also stale (2026-02-07) | +| `plandex-ai/plandex` | 15.6k | 2025-10-03 | 11 mo | no | **remove** | +| `bytedance/trae-agent` | 12.1k | 2026-02-05 | 7 mo | no | **remove** | +| `AntonOsika/gpt-engineer` | 55.1k | 2025-05-14 | 16 mo | yes | keep only if the "historical significance" carve-out stands; else remove | +| `RooCodeInc/Roo-Code` | 24.3k | 2026-05-15 | 3.9 mo | yes | **user call** — archived; historical value as Cline fork lineage | +| `voideditor/void` | 28.8k | 2026-06-02 | 3.3 mo | yes | **user call** — archived, empty description | +| `Aider-AI/aider` | 48.9k | 2026-05-22 | 3.7 mo | no | **user call** — hardest one; 2026 roundups still call it a top OSS agent, but `pushedAt` shows no commits in 3.7 mo | + +Watch list (inside 3 mo, trending stale): `TabbyML/tabby` (2026-06-30), `Pythagora-io/gpt-pilot` (2026-06-18). + +## 3. Rejected candidates (why) + +Out per criterion 1 — orchestrators, runtimes, wrappers, skill packs, general assistants: + +`NousResearch/hermes-agent` 244.3k (general personal agent, not coding-specific) · `ultraworkers/claw-code` 195k (Rust `claw` harness, but README self-describes as an "agent-managed museum exhibit"; star count disproportionate to substance — **flagged, user call**) · `bytedance/deer-flow` 82.3k (general SuperAgent) · `ruvnet/ruflo` 72k (meta-harness/swarm) · `code-yeongyu/oh-my-openagent` 68.9k (plugin bundle for existing hosts) · `stablyai/orca` 66.2k, `superset-sh/superset` 14.1k, `gastownhall/gastown` 18k, `BloopAI/vibe-kanban` 28.1k, `different-ai/openwork` 23.5k, `getpaseo/paseo`, `iOfficeAI/AionUi` (fleet/orchestration/workspace) · `herdrdev/herdr` 37.6k (runtime agents run on) · `sipeed/picoclaw` 30k (personal assistant on $10 hardware) · `musistudio/claude-code-router` 37.2k, `farion1231/cc-switch` 132k, `router-for-me/CLIProxyAPI` 51.3k (routers) · `trycua/cua` 22.5k (computer-use) · `FoundationAgents/MetaGPT` (framework) · skills/prompt repos (`obra/superpowers`, `anthropics/skills`, `addyosmani/agent-skills`, …). + +Below the 10k floor: `aws/amazon-q-developer-cli` 2.0k · `smallcloudai/refact` 3.5k (archived) · `superagent-ai/grok-cli` 3.5k · `just-every/code` 4.0k · `codestoryai/aide` 2.2k (archived). + +Dead + below relevance: `smol-ai/developer` 12.2k (2024-04-07) · `Doriandarko/claude-engineer` 11.2k (2024-12-12) · `stitionai/devika` 19.6k (2025-09-25) — all >10k ★ but long abandoned; adding them would contradict the cleanup in §2. + +No repo to track (closed source): Cursor, Antigravity (the closed Gemini CLI successor already noted in `agents.yml`), Kiro, Devin, Amp, Factory Droid, Augment. + +## 4. Data defect + +`data/agents.yml` lists `owner: yetone` / `repo: avante.nvim`; canonical slug is now **`avante-corp/avante.nvim`** (API redirects, README already renders the new name). The file header mandates the current canonical slug + a `history.go` migration. Fix: + +1. `data/agents.yml` → `owner: avante-corp` +2. `history.go` `canonicalKeyMigrations` → add `"yetone/avante.nvim": "avante-corp/avante.nvim"` + +Without step 2, avante's star history detaches and Δ7d shows `—`. + +## Next steps + +1. Apply §1 YAML patch (14 adds) + §4 fix; run `go run . -check` then the updater. +2. Decide the 4 user-call rows in §2 (aider, Roo-Code, void, gpt-engineer) and `claw-code` in §3. +3. Consider adding to CONTRIBUTING: an explicit **staleness rule** (e.g. no push in 6 months ⇒ drop unless historically significant), since "actively maintained" is currently unquantified — that is what let bolt.new sit for 21 months. + +## Unresolved questions + +1. `Aider-AI/aider` — keep despite 3.7 mo of no pushes (name recognition) or drop? +2. Archived-but-significant (`void`, `Roo-Code`, `gpt-engineer`) — keep marked, or is the carve-out only for gpt-engineer? +3. `ultraworkers/claw-code` (195k ★, functional Rust harness, joke-flavored README) — include, exclude, or exclude with a documented reason? +4. Staleness threshold to codify: 3 months (your ask) or 6 months (less churn)? +5. `alibaba/open-code-review` is review-only, not an editing agent — in scope per criteria wording; confirm intent? + +## Sources + +- [Best Open Source CLI Coding Agents in 2026 — Pinggy](https://pinggy.io/blog/best_open_source_cli_coding_agents/) +- [Best Terminal AI Coding Agents in 2026 — amux](https://amux.io/blog/best-terminal-ai-coding-agents-2026/) +- [Best CLI AI Tools in 2026 — kilo.ai](https://kilo.ai/articles/best-cli-coding-agents) +- [AI-Coding-Landscape](https://github.com/joylarkin/AI-Coding-Landscape) · [awesome-cli-coding-agents](https://github.com/bradAGI/awesome-cli-coding-agents) +- Live GitHub GraphQL API (stars, pushedAt, isArchived) — 2026-09-11