Key Takeaways
- +104,870 stars across 180 tracked repos (+6.1% in 30 days) — third consecutive month of slowing growth (12.4% → 8.3% → 6.1%)
- 46 projects last shipped in March or April, and 33 of them are
claw-named. The naming wave has run its course - Ecosystem crossed 1.8 million stars
- microsoft/SkillOpt grew +61% (9,357 → 15,046), the month's steepest climb among large repos
- Seven new repositories added, including Harvard's ToolUniverse and a Technion framework published in NEJM AI
- bio-xyz/BioAgents was archived and moved to the graveyard — the first resident that is archived rather than deleted
The Numbers
| Metric | 6/26 | 7/26 | Δ |
|---|---|---|---|
| Tracked repositories (projects + hubs) | 180 | 187 | +7 |
| Total stars (180 repos tracked in both snapshots) | 1,715,902 | 1,820,772 | +104,870 (+6.1%) |
| Projects in graveyard | 2 | 3 | +1 (BioAgents) |
Repositories marked stable | 24 | 70 | +46 |
That jump in stable is a reclassification rather than 46 projects dying this month — see The Claw Cohort below.
Top Movers (6/26 → 7/26)
By absolute stars added:
| Repository | 6/26 | 7/26 | Added | % |
|---|---|---|---|---|
| NousResearch/hermes-agent | 203,342 | 220,574 | +17,232 | +8% |
| earendil-works/pi | 65,695 | 77,661 | +11,966 | +18% |
| anthropics/skills | 155,355 | 164,196 | +8,841 | +6% |
| Yuan1z0825/nature-skills | 23,420 | 31,202 | +7,782 | +33% |
| microsoft/SkillOpt | 9,357 | 15,046 | +5,689 | +61% |
| Imbad0202/academic-research-skills | 34,583 | 39,529 | +4,946 | +14% |
| blader/humanizer | 26,198 | 31,079 | +4,881 | +19% |
| alirezarezvani/claude-skills | 19,096 | 23,202 | +4,106 | +22% |
By percentage (start ≥ 100 stars):
| Repository | 6/26 | 7/26 | Added | % |
|---|---|---|---|---|
| qinheming/BIoClaw | 225 | 416 | +191 | +85% |
| microsoft/SkillOpt | 9,357 | 15,046 | +5,689 | +61% |
| TIGER-AI-Lab/OpenResearcher | 792 | 1,099 | +307 | +39% |
| jaechang-hits/SciAgent-Skills | 211 | 283 | +72 | +34% |
| Yuan1z0825/nature-skills | 23,420 | 31,202 | +7,782 | +33% |
| yaojingang/yao-meta-skill | 1,615 | 2,125 | +510 | +32% |
| InternScience/ResearchClawBench | 172 | 225 | +53 | +31% |
| Master-cai/Research-Paper-Writing-Skills | 4,381 | 5,522 | +1,141 | +26% |
The Claw Cohort
The clearest signal in July's data is not in the growth table. It is in the push dates.
46 tracked projects last shipped code in March or April 2026 and have been silent since. 33 of them carry claw in the name: LabClaw, MedgeClaw, MathClaw, BIoClaw, DrClaw, ClawPhD, MolClaw, MatClaw, EpiClaw, ScienceClaw, PythonClaw, and two dozen more.
The distribution is tight:
| Last shipped | Projects |
|---|---|
| March 2026 | 26 |
| April 2026 | 20 |
That is not a gradual decay — it is a cohort. These projects were largely created during the naming rush that followed OpenClaw's rise, shipped for a few weeks to a few months, and stopped. Eight separate projects named PaperClaw is the clearest artifact of that period; we built eight disambiguation pages because readers genuinely could not tell them apart.
Two things worth stating plainly about this cohort:
It is not 107K stars of abandonment. The 46 projects hold 107,416 stars between them, but karpathy/autoresearch alone accounts for 92,051 of those — 86%. Strip it out and the remaining 45 projects share 15,365 stars, averaging 341 each. The cohort is numerous rather than large.
Silence is not failure. A skill pack that covers its domain is finished. autoresearch at 92K stars with no commits since March is not a dead project — it is a reference implementation that reached a stable state. We have reclassified all 46 as stable, which removes them from our dormancy alerts, and we now report the opposite signal instead: a stable project that starts shipping again. That check caught ai4nucleome/biomaster in its first run, which resumed work on 2026-07-14 after months of quiet.
The projects still shipping every week are disproportionately the infrastructure layer — Hermes, pi, nanobot, openclaw itself — and the skill hubs. The domain-specific agent wrappers are the part that went quiet.
Growth Is Decelerating, Consistently
| Month | Growth | Window |
|---|---|---|
| May | +12.4% | 4/30 → 5/28 |
| June | +8.3% | 5/28 → 6/26 |
| July | +6.1% | 6/26 → 7/26 |
Three months, a clean downward line. Part of this is arithmetic: the base is 21% larger than it was in May, so equal absolute growth reads as a smaller percentage. In absolute terms July added +104,870 stars against June's +125,831 — a 17% decline, milder than the percentage figures suggest.
The honest framing is that the ecosystem is maturing rather than stalling. Attention is consolidating into a handful of infrastructure projects and skill hubs while the long tail stops growing.
New Arrivals
Seven repositories entered the directory in July. Five came from our trending pass, which we repaired this month after finding it had been implemented but never actually invoked:
mims-harvard/ToolUniverse (1,593⭐) — Harvard MIMS Lab. Gives any reasoning model access to 600+ scientific tools and MCP servers across biomedicine, chemistry, and clinical data. The tool-layer bet rather than the agent-layer bet.
OpenNSWM-Lab/FAROS (2,368⭐) — blueprint-driven AutoResearch runtime covering idea generation through peer review, declaratively rather than as a hand-wired agent graph.
Technion-Kishony-lab/data-to-paper (814⭐) — published in NEJM AI. Every claim in the generated paper traces back to the data and code that produced it. Added as stable; the last push was July 2025 and the work is paper-anchored.
Kaimen-Inc/Co-Scientist (203⭐) — open reimplementation of DeepMind's AI Co-Scientist, using a generate-debate-evolve tournament loop.
ai4s-research/ai4s-skills (157⭐) — added as a skill hub. Covers topic exploration through integrity audit.
Also added: synthetic-sciences/openscience (2,788⭐) and xuzhougeng/wisp-science (447⭐).
Into the Graveyard
bio-xyz/BioAgents was archived around 2026-07-22 and now has a graveyard entry. It is the first resident that is archived rather than deleted, which makes it unusual in a useful way: the code stays readable forever.
It also kept gaining stars after it stopped shipping — 129 when we began tracking it in April, 179 today, with the final commit on 2026-06-15. The bio-xyz org remains active and CoreAgent was pushed the day before we wrote the entry, so this looks like a version boundary.
What to Watch in August
- Does the claw cohort stay quiet? If August passes with no revivals, the March–April wave is definitively closed
- Hermes at 240K — it added +17K in July, down from +32K in June. The deceleration we have predicted for three months is finally visible in the numbers
- Does DeepMind's science-skills gain on nature-skills? nature-skills grew +33% in July and now sits at 31.2K
- microsoft/SkillOpt at 25K — +61% in a month is the kind of pace that either continues sharply or stops abruptly
- More archived-not-deleted projects — BioAgents may be the first of a pattern as the 2025-era cohort ages
Methodology
- Star counts: GitHub stars snapshotted daily into
docs/stars-history.csv - Growth window: 2026-06-26 → 2026-07-26 (30 days)
- Comparison basis: the 180 repositories present in both snapshots. Repos added mid-window are listed under New Arrivals and excluded from growth totals
- Top-mover filters: percentage rankings require start ≥ 100 stars
- Dormancy: measured from GitHub
pushed_at. Projects markedstableare excluded from dormancy alerts, since a project that stops shipping has usually finished what it set out to do - Data note: skill-hub star counts were absent from our refresh job until this month, so hub figures before 2026-07-26 come from the daily snapshot rather than the hub listing. The snapshot is the authoritative series and is what every number here is drawn from
Previous report: Ecosystem Report: June 2026 — Hermes Crosses 200K — 180 repos, +8.3%, Hermes crossed 200K on 6/23, DeepMind entered the skill-hub category.
Last updated: 2026-07-26.
