Skip to content

Discovery report 2026-07-09: new agent candidates #152

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
YugantM wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Discovery report 2026-07-09: new agent candidates #152

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
217 changes: 217 additions & 0 deletions docs/research/new-agent-candidates-2026-07-09.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,217 @@
{
"snapshot_date": "2026-07-09",
"discovery_command": "GITHUB_TOKEN=$(gh auth token) python3 discover_agents.py",
"discovery_totals": {
"repositories_found": 561,
"novel_against_agents_json": 416,
"passed_automated_prechecks": 381,
"after_rubric_text_filter": 287,
"manually_vetted": 18
},
"context_note": "The ai-agent topic space is heavily star-farmed in 2026: the top results by stars are awesome-lists, skills libraries, format specs, and tools-for-agents, plus 2026-created repos with implausible star counts. Candidates below were filtered against docs/strict-inclusion-rubric.md and verified via the GitHub API (creation date, license, repo structure, README-vs-description). HVTrust is robust to inflated stars — since v4.1 popularity ranks BELOW audit signals and does not lift the score — so even a star-farmed repo lands at its true (usually low) trust once real signals fill in. agents.json was NOT changed; this is a discovery report for owner decision.",
"recommended_adds": [
{
"source_url": "https://github.com/Intelligent-Internet/ii-agent",
"snapshot_evidence": {
"repository": "Intelligent-Internet/ii-agent",
"created": "2025-04-07",
"last_push": "2026-04-13",
"stars": 3365,
"license": "Apache-2.0",
"language": "Python",
"topics_or_dirs": "open-source AI agent, runnable/forkable, BYOK",
"readme_evidence": "\"II-Agent is an open-source AI agent built for real work\" — general-purpose agent, run/fork/extend, bring-your-own-key."
},
"reason": "General-purpose first-party AI agent with its own execution codebase, Apache-2.0, credible org (Intelligent-Internet), created before the 2026 star-farm era. Clears the agent test (multi-step execution, tool use, autonomy).",
"confidence": "high",
"priority": "P1",
"exact_agents_json_entry": {
"repo": "Intelligent-Internet/ii-agent",
"name": "II-Agent",
"category": "Agent Frameworks",
"listing_status": "listed",
"tracking_mode": "direct"
}
},
{
"source_url": "https://github.com/Integuru-AI/Integuru",
"snapshot_evidence": {
"repository": "Integuru-AI/Integuru",
"created": "2024-10-22",
"last_push": "2026-06-24",
"stars": 4633,
"license": "AGPL-3.0",
"language": "Python",
"readme_evidence": "Agent identifies the request that performs an action, traverses a dependency graph of network requests, and converts each node into runnable Python code — goal-directed, multi-step, tool-using."
},
"reason": "First-party autonomous agent that reverse-engineers platform integrations into runnable code via graph traversal. Oldest and most credible of the batch (created 2024-10), clear OSS license, real agent logic in-repo.",
"confidence": "high",
"priority": "P1",
"exact_agents_json_entry": {
"repo": "Integuru-AI/Integuru",
"name": "Integuru",
"category": "Coding Agents",
"listing_status": "listed",
"tracking_mode": "direct"
}
},
{
"source_url": "https://github.com/ruc-datalab/DeepAnalyze",
"snapshot_evidence": {
"repository": "ruc-datalab/DeepAnalyze",
"created": "2025-10-11",
"last_push": "2026-07-01",
"stars": 4342,
"license": "MIT",
"language": "Python",
"readme_evidence": "\"Agentic Large Language Models for Autonomous Data Science\" — performs data-oriented deep research and data-centric tasks locally without any workflow or closed-source APIs."
},
"reason": "Autonomous data-science agent from a real academic lab (RUC DataLab). First-party agentic loop, MIT-licensed, deployable locally. Clears the agent test (autonomous multi-step task execution + tool use).",
"confidence": "high",
"priority": "P1",
"exact_agents_json_entry": {
"repo": "ruc-datalab/DeepAnalyze",
"name": "DeepAnalyze",
"category": "Research & Data",
"listing_status": "listed",
"tracking_mode": "direct"
}
},
{
"source_url": "https://github.com/superagent-ai/grok-cli",
"snapshot_evidence": {
"repository": "superagent-ai/grok-cli",
"created": "2025-07-14",
"last_push": "2026-07-06",
"stars": 3219,
"license": "MIT",
"language": "TypeScript",
"npm_package": "grok-dev (published, verified)",
"readme_evidence": "Terminal coding agent: tool rounds (--max-tool-rounds), sub-agents on by default, X/web search, remote control via Telegram. Same class as coding agents already tracked."
},
"reason": "Community-built open-source terminal coding agent (known org superagent-ai), MIT, published npm package, first-party agent loop with sub-agents and tool use. Disclaims xAI affiliation (community project, not a vendor wrapper).",
"confidence": "high",
"priority": "P1",
"exact_agents_json_entry": {
"repo": "superagent-ai/grok-cli",
"name": "Grok CLI",
"category": "Coding Agents",
"npm_package": "grok-dev",
"listing_status": "listed",
"tracking_mode": "direct"
}
}
],
"manual_review": [
{
"source_url": "https://github.com/lsdefine/GenericAgent",
"snapshot_evidence": {
"repository": "lsdefine/GenericAgent",
"created": "2026-01-16",
"last_push": "2026-07-09",
"stars": 13345,
"license": "MIT",
"language": "Python",
"pypi_package": "genericagent (published, verified)",
"structure": "ga_cli, memory, plugins, reflect, frontends; topics: autonomous-agent, self-evolving, skill-tree, memory-system, computer-control, browser-automation"
},
"reason": "General-purpose self-evolving agent with first-party CLI loop, memory, plugins, and reflection — a strong fit on the merits. Held for review only because it was created in 2026 with an implausibly high star count (star-farm era); confirm canonicity (not a renamed clone) before listing.",
"confidence": "medium",
"priority": "P2",
"exact_agents_json_entry": {
"repo": "lsdefine/GenericAgent",
"name": "GenericAgent",
"category": "Agent Frameworks",
"pypi_package": "genericagent",
"listing_status": "listed",
"tracking_mode": "direct"
}
},
{
"source_url": "https://github.com/HKUDS/Vibe-Trading",
"snapshot_evidence": {
"repository": "HKUDS/Vibe-Trading",
"created": "2026-04-01",
"last_push": "2026-07-09",
"stars": 18858,
"license": "MIT",
"language": "Python",
"pypi_package": "vibe-trading-ai (published, verified)",
"structure": "agent/ (first-party), tools/, frontend/; topics: ai-agent, multi-agent, mcp, algorithmic-trading, backtesting"
},
"reason": "Trading agent from HKUDS — the lab behind LightRAG, an existing HVTracker badge adopter — with a first-party agent/ module and MCP support. Real agent, credible lab. Held only because it is domain-specific (finance) and very recently created; confirm it is a general enough agent vs a single-purpose app.",
"confidence": "medium",
"priority": "P2",
"exact_agents_json_entry": {
"repo": "HKUDS/Vibe-Trading",
"name": "Vibe-Trading",
"category": "Research & Data",
"pypi_package": "vibe-trading-ai",
"listing_status": "listed",
"tracking_mode": "direct"
}
},
{
"source_url": "https://github.com/666ghj/BettaFish",
"snapshot_evidence": {
"repository": "666ghj/BettaFish",
"created": "2024-07-01",
"last_push": "2026-05-24",
"stars": 41685,
"license": "GPL-2.0",
"language": "Python",
"structure": "ForumEngine, InsightEngine, MediaEngine, MindSpider, QueryEngine, ReportEngine, SentimentAnalysisModel; topics: agent-framework, multi-agent-system, deep-research",
"note": "README is primarily Chinese"
},
"reason": "Substantial multi-agent public-opinion-analysis system with first-party engines built 'from 0, no framework dependency'. Age-credible (created 2024-07, predates the star-farm era). Held because it is a domain-specific application (opinion analysis) rather than a general agent/framework — owner judgment on whether the multi-agent orchestration clears the 'primary purpose is agent construction / goal-directed agent' bar.",
"confidence": "medium",
"priority": "P2",
"exact_agents_json_entry": {
"repo": "666ghj/BettaFish",
"name": "BettaFish",
"category": "Multi-Agent Systems",
"listing_status": "listed",
"tracking_mode": "direct"
}
},
{
"source_url": "https://github.com/ModelEngine-Group/nexent",
"snapshot_evidence": {
"repository": "ModelEngine-Group/nexent",
"created": "2025-04-28",
"last_push": "2026-07-09",
"stars": 5555,
"license": "MIT",
"language": "Python",
"description": "Zero-code platform for auto-generating production-grade agents"
},
"reason": "Agent-construction platform, MIT, age-credible, actively maintained. Held because 'zero-code platform for generating agents' sits near the UI-app-builder / no-code boundary — confirm it provides genuine agent-framework abstractions (tool dispatch, memory, lifecycle) rather than being a generic no-code builder with incidental agent support.",
"confidence": "low",
"priority": "P3",
"exact_agents_json_entry": {
"repo": "ModelEngine-Group/nexent",
"name": "Nexent",
"category": "Agent Frameworks",
"listing_status": "listed",
"tracking_mode": "direct"
}
}
],
"notable_rejections": [
{"repo": "code-yeongyu/oh-my-openagent", "stars": 65396, "reason": "NOASSERTION (no clear OSS license) + wrapper/TUI orchestrating other coding agents (claude, codex, cursor, opencode) rather than a first-party agent."},
{"repo": "santifer/career-ops", "stars": 59307, "reason": "Implausible star count for a job-search app created 2026-04 (star-farm); domain-narrow application."},
{"repo": "the-open-agent/openagent", "stars": 5366, "reason": "Creation date 2020-05 with a modern LLM/RAG description — canonicity red flag (likely renamed/repurposed repo); verify identity before ever listing."},
{"repo": "elder-plinius/T3MP3ST", "stars": 4094, "reason": "Created 2026-07-02 (days old) with 4k stars — extreme star-farm; too new to assess as a real, maintained project."},
{"repo": "golutra/golutra", "stars": 3752, "reason": "NOASSERTION license (no clear OSS license)."},
{"repo": "snarktank/ralph", "stars": 20964, "reason": "Stale (no push since 2026-02) and 21k stars weeks after a 2026-01 creation — viral/star-farmed, not actively maintained."},
{"repo": "NVIDIA/OpenShell", "stars": 7511, "reason": "Runtime/sandbox for agents — a tool-for-agents without its own autonomous loop (same class excluded in the 2026-07-06 report)."},
{"repo": "*skills / *tools / *templates / OpenClaw-clients / htop-for-agents / slide+ppt+video generators / memory+context layers / format specs (agents.md, design.md)", "reason": "Rubric out-of-scope classes: prompt/skill collections, tools-for-agents, UI/media generators, wrappers around hosted agents, and format specifications. The bulk of the 287 post-filter survivors."}
],
"verification_notes": [
"GitHub metadata (created/pushed/stars/license/language, dir structure, topics) captured 2026-07-09 via the authenticated API.",
"README claims spot-checked against implementation intent for all four recommended adds (descriptions alone were not accepted).",
"Package identities verified against the live registries: PyPI genericagent 200, vibe-trading-ai 200; npm grok-dev 200; PyPI ii-agent 404 and integuru 404 (repo-only tracking). DeepAnalyze/nexent have no published package (repo-only).",
"The three previously owner-rejected repos (LobsterAI, Agent Orchestrator, Sandcastle) were correctly filtered out by discover_agents.py REVIEWED_REJECTED.",
"candidates.json (full 381) is a scratch artifact and intentionally not part of this deliverable."
]
}
88 changes: 88 additions & 0 deletions docs/research/new-agent-candidates-2026-07-09.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,88 @@
# New agent candidates — 2026-07-09

`GITHUB_TOKEN=$(gh auth token) python3 discover_agents.py` found 561 repos,
416 novel against the current 347-entry `agents.json`, 381 passing the
script's automated pre-checks. A rubric-based text filter
(`docs/strict-inclusion-rubric.md`) removed 94 obvious out-of-scope repos
(collections, guides, skills, format specs, tools-for-agents), leaving 287
for review; 18 were manually vetted via the GitHub API.

This is a discovery report only. **`agents.json` was not changed** —
discovery proposes, owner decides (and applies via the add-agent runbook).

## The 2026 star-farm caveat (read first)

The `ai-agent` topic space is now heavily star-farmed. The top results by
stars are **not agents**: awesome-lists (`awesome-llm-apps` 117k⭐),
skills libraries (`agent-skills` 76k⭐, `claude-skills`), format specs
(`agents.md`, `design.md`), tools-for-agents (`beads`, `agent-browser`,
memory/context layers), and OpenClaw desktop clients. Many genuine-looking
repos are 2026-created with implausible star counts (e.g. `career-ops`:
59k⭐ three months after creation).

**Why this is safe for HVTracker anyway:** since scoring v4.1, popularity
ranks *below* audit signals and no longer lifts the score. A star-farmed
repo therefore lands at its true (usually low) trust once real signals
fill in — exposing inflated stars is on-brand for a trust registry, not a
risk to it. The filtering below is about the *inclusion rubric* (is it a
real agent?), not about star count.

## Recommended adds (P1 — verified real agents, credible, OSS-licensed)

| Project | Created | ⭐ | License | Why it passes |
| --- | --- | --- | --- | --- |
| [II-Agent](https://github.com/Intelligent-Internet/ii-agent) | 2025-04 | 3.4k | Apache-2.0 | General-purpose open-source AI agent, run/fork/extend, BYOK; first-party execution codebase. |
| [Integuru](https://github.com/Integuru-AI/Integuru) | 2024-10 | 4.6k | AGPL-3.0 | Autonomous agent that traverses a network-request dependency graph to generate runnable integration code — goal-directed, multi-step, tool-using. Oldest/most credible. |
| [DeepAnalyze](https://github.com/ruc-datalab/DeepAnalyze) | 2025-10 | 4.3k | MIT | "Agentic LLM for Autonomous Data Science" from RUC DataLab (real academic lab); deep-research + data tasks locally, no closed-source workflow. |
| [Grok CLI](https://github.com/superagent-ai/grok-cli) | 2025-07 | 3.2k | MIT | Terminal coding agent (org superagent-ai): tool rounds, sub-agents by default, search; published npm `grok-dev`. Community project, not a vendor wrapper. |

All four predate the star-farm era or come from a credible org, carry a
clear OSS license, and their READMEs confirm a first-party agent loop
(checked against implementation intent, not description alone). The
companion JSON has copy-ready `agents.json` entries.

## Manual-review items (P2/P3 — real but need an owner call)

- **[GenericAgent](https://github.com/lsdefine/GenericAgent)** (2026-01,
13k⭐, MIT, PyPI `genericagent`) — general self-evolving agent with a
first-party CLI loop, memory, plugins, reflection. Strong on the merits;
held only because it is 2026-created with an implausibly high star count
— confirm canonicity (not a renamed clone) before listing.
- **[Vibe-Trading](https://github.com/HKUDS/Vibe-Trading)** (2026-04,
19k⭐, MIT, PyPI `vibe-trading-ai`) — trading agent from **HKUDS**, the
lab behind LightRAG (an existing HVTracker badge adopter), with a
first-party `agent/` module and MCP support. Held because it is
domain-specific (finance) and very recent.
- **[BettaFish](https://github.com/666ghj/BettaFish)** (2024-07, 42k⭐,
GPL-2.0) — substantial multi-agent public-opinion-analysis system with
first-party engines built "from 0, no framework". Age-credible; held
because it is a domain application rather than a general agent/framework.
- **[Nexent](https://github.com/ModelEngine-Group/nexent)** (2025-04,
5.6k⭐, MIT) — zero-code platform for generating agents. Held because it
sits near the no-code-builder boundary — confirm it offers genuine
agent-framework abstractions, not incidental agent support.

## Notable rejections (with reasons)

- `code-yeongyu/oh-my-openagent` (65k⭐) — no clear OSS license
(NOASSERTION) **and** a TUI orchestrating *other* coding agents, not a
first-party agent.
- `the-open-agent/openagent` (5.4k⭐) — 2020 creation date with a modern
LLM/RAG description = canonicity red flag (likely renamed/repurposed).
- `elder-plinius/T3MP3ST` (4k⭐) — created 2026-07-02, days old with 4k
stars; extreme star-farm, unassessable.
- `snarktank/ralph` (21k⭐) — stale (no push since 2026-02); viral, not
maintained.
- `golutra/golutra` — NOASSERTION license.
- `NVIDIA/OpenShell` — runtime/sandbox for agents, no own loop (same class
excluded on 2026-07-06).
- The bulk of the 287 survivors — skills/tools/templates/format-specs/
OpenClaw-clients/media-generators — are rubric out-of-scope.

## If the owner approves adds

Follow the add-agent runbook (`deploy_mechanism` memory / CLAUDE.md): edit
`agents.json` only → one branch/PR → three gates → squash-merge →
(owner-instructed) `railway up` clean worktree → `railway restart` to score
the provisional rows. New agents land grade-D with `scorecard_score:null`
until the daily OSSF scan and a later deploy fill in sub-scores.
Loading