External finance skill index
Every field below came from a live gh api repos/<o>/<r> call on 2026-09-04. Skill counts are
literal SKILL.md matches in each repo's git tree, not README claims. The machine-readable catalog
is ../../../../catalog/external-skills.json (139 rows) — grep it rather than guessing:
python -c "import json;d=json.load(open('catalog/external-skills.json',encoding='utf-8'));\
print('\n'.join(f\"{r['full_name']:<45}{r['stars']:>6}★ {r['license'] or 'NONE':<12}{r['skill_count']:>4} skills\" \
for r in d['repos'] if 'backtesting' in r['tags']))"
1. The size of the ecosystem
| Repositories indexed | 139 |
Raw SKILL.md files |
4,851 |
| After removing mirrors and vendored copies | 4,556 |
| Finance-dominant | 3,127 — the fairest single number |
⚠️ Volume is not depth. quantskills alone is 230 repos and 1,094 skills, but 803 of those
(73%) sit in three auto-generated factor repos.
2. 🚨 A third of it is legally unusable
| Licence | Repos |
|---|---|
| MIT | 76 |
| Apache-2.0 | 14 |
| MIT-0 | 1 |
| 🚨 NONE | 33 |
| 🚨 NOASSERTION | 11 |
| GPL-3.0 | 2 (+ the quantskills org, 192 of 230) |
| AGPL-3.0 | 1 |
| archived | 2 |
91 safe (65%) · 44 legally unusable (32%). No licence is more restrictive than GPL — under Berne it means all rights reserved, so you may not copy, adapt or redistribute it at all.
🚨 And it hits exactly the repos you would reach for first:
| Repo | Stars | Licence |
|---|---|---|
okx/onchainos-skills |
329★ | NONE |
lzwme/finance-quant-skills |
321★ | NONE |
ALAGENT-HKU/x2strategy |
270★ | NONE |
marketcalls/vectorbt-backtesting-skills |
200★ | NONE |
Check the licence before you copy a single file. Stars are not permission.
3. 🚨 Twelve repos advertise skills and ship none
~1,333 stars sit on repos with zero SKILL.md files, including
quant-sentiment-ai/claude-equity-research (709★ — actually a slash-command plugin) and
angieruiz17/claude-fintech-skills (144★ — a TypeScript app). The word "skills" in a repo name is
marketing, not a format claim. The catalog records skill_count: 0 for each.
4. 🚨 Neither awesome-list has a finance section
ComposioHQ/awesome-claude-skills— 74,466★, and its only finance-adjacent entry is an "Invoice Organizer".travisvn/awesome-claude-skills— 14,966★, nothing.anthropics/skills— 20 skills, zero finance, confirmed.
There is no curated finance section anywhere. If you are looking for a discovery route, you would be creating that section, not joining it.
5. Where to send someone instead of duplicating
Of 139 repos, only 9 overlap quantitative research methodology, 3 of those are GPL or unlicensed, and the permissive-and-still-maintained remainder holds 13 stars in total.
| Need | Go to | Licence |
|---|---|---|
| Leakage-safe quant ML, ML-pipeline framing | ml4t/skills (Stefan Jansen, 61 skills) |
Apache-2.0 |
| Crypto/DeFi/Solana execution, MEV, prediction markets | agiprolabs/claude-trading-skills (68) |
MIT |
| RIA compliance, KYC/AML, Reg BI, GIPS, practice ops | JoelLewis/finance_skills (91) |
MIT |
| Personal/SMB bookkeeping and tax | openaccountant/skills (44) |
MIT |
| ⚠️ A-share factor mining | quantskills/* |
GPL-3.0, Pandadata-locked |
⚠️ ml4t/skills is the closest peer and worth reading. Two structural notes, verified: it has
no .claude-plugin/ manifest while using category directories, so Claude Code will not discover
its skills at .claude/skills/<category>/<skill>/; and it uses non-spec frontmatter keys
(when_to_use, dependencies) that hard-error on claude.ai upload. Its content is strong; its
packaging targets a different client.
6. The three gaps nobody covers
- Research integrity with any traction. 12 repos tag it; the credible ones have <15 stars or are GPL. Meanwhile 26 repos ship discretionary TA — ICT/SMC, Chan Theory, and one bundling Vedic astrology with options Greeks.
- Fixed income, rates and derivatives-pricing correctness.
fixed-incomeappears on 2 of 139 repos,derivativeson 3. Nothing covers day counts, calendars, curve bootstrapping, OIS discounting, or the QuantLib traps. - Point-in-time and survivorship-safe fundamental data, vendor by vendor. 21 repos wrap
market-data APIs; not one documents when the data was knowable.
point-in-timeappears in exactly 2 places, neither vendor-specific.
7. How to use this index
- Before writing a new skill, grep the catalog for its tags. If something permissive already covers it, cite it — this library's position is that duplicating a good MIT skill is waste.
- Before recommending a repo, check
licenseandskill_countin the catalog. A third of the ecosystem fails one of those two. relationon each row iscomplementary(80),overlapping(9) oravoid(50), with the reason incaution.- The catalog is a snapshot dated 2026-09-04. Star counts and licences change; re-verify before acting on a licence claim.
92 of these install through this marketplace already
As of 2026-09-09 the fin-skills marketplace federates 92 third-party packs exposing 2389
skills by their own GitHub source, so the answer to "is there a skill for X" is often an install
command rather than a link. Every one installs disabled, and their claims are not verified by
this repo - what was checked is the licence, that the repository is live and not archived, that
real SKILL.md files exist at the pinned commit, and the skill count. None was installed or read
for correctness.
They are vendor-official packs (Alpaca, Kraken, Longbridge, OKX, HTX, Pionex, BloFin, CoinStats, Upstox, J-Quants, Nansen), Anthropic's own financial-services marketplace and its A-share port, and community packs covering equity research, real estate, tax and accounting, macro, and market data for China, Japan and India.
🚨 Several can place live orders once API keys are set, and their descriptions say so. Read
../broker-execution-apis/SKILL.md before installing one of those.
⚠️ Sizes vary by two orders of magnitude. The largest single pack is
algo-trading-skills at 501 skills - about 25 times the default
listing budget on its own. Federation is per plugin, so a pack's skills cannot be cherry-picked.
Install deliberately and raise skillListingBudgetFraction first.
The eight largest:
| Marketplace name | Upstream | Skills |
|---|---|---|
algo-trading-skills |
https://github.com/HimanshuJ16/Algo-Trading-Skills | 501 |
ftshare-skills |
https://github.com/FTShare-Lab/FTShare-skill | 215 |
yuping322-finskills |
https://github.com/yuping322/finskills | 107 |
vibe-trading-skills |
https://github.com/HKUDS/Vibe-Trading | 90 |
doramagic-skills |
https://github.com/tangweigang-jpg/doramagic-skills | 82 |
tradermonty-trading-skills |
https://github.com/tradermonty/claude-trading-skills | 74 |
trading-skills |
https://github.com/agiprolabs/claude-trading-skills | 68 |
octagon-skills |
https://github.com/OctagonAI/skills | 67 |
The full list with what was verified for each, at which commit, and the candidates left out with
reasons, is in ../../../../catalog/federation-notes.md (wave 1) and
../../../../catalog/federation-notes-wave2.md (wave 2).