npx companies.sh add stubbi/companies/financial-services

That's it — your Paperclip instance now has an 11-agent finance company on the board, with org chart, teams, and skills wired up.

How does this work? · What's an Agent Company? · Browse the catalog

An Agent Company is a fully-configured team of AI agents — org chart, teams, skills, governance — packaged so you can import it into Paperclip and have a working AI workforce in minutes. Each company in this repo is:

• Domain-specific. Finance, legal, research, ops — not generic prompt wrappers.
• Skill-loaded. Every agent ships with the workflows it needs to do its job.
• Runtime-agnostic. Works with Claude Code, Codex, Gemini, OpenCode, Cursor, Pi, and Hermes via Paperclip's adapter chain — Claude is the reference runtime, others run with varying skill polish.
• Boundary-aware. Agents know what they can and cannot decide; outputs are staged for human sign-off where regulation requires it.
• Provenance-honest. Community ports are pinned by upstream commit SHA, hashed for drift detection, and preserve upstream license + attribution. Nothing is vendored or forked.

npx companies.sh add stubbi/companies/<company-slug>

companies.sh resolves the path to a directory in this repo, validates the company manifest, and imports the team into your Paperclip instance. Each company directory is self-contained:

<company-slug>/
├── COMPANY.md          # company-level manifest (generated)
├── teams/              # team manifests (generated)
├── agents/             # one AGENTS.md per role (generated)
├── skills/             # one SKILL.md per skill — generated for upstream-referenced,
│                       # hand-authored for port-original
├── manifest.yaml       # canonical source — edit this, run `make build`
├── images/             # org chart (generated)
├── LICENSE             # company-level license
└── NOTICE              # upstream attribution where applicable

Companies that port an external repo (like financial-services) ship a make bump SHA=<new-sha> workflow that re-fetches upstream by content hash and re-generates the manifest. The pinned SHA is the contract.

Community port of anthropics/financial-services (Anthropic's "Claude for Financial Services") into the Agent Companies format. Runtime-agnostic; Claude is the reference runtime, Codex / Gemini / OpenCode / Hermes / Cursor / Pi are supported via the adapter chain with varying skill polish.

npx companies.sh add stubbi/companies/financial-services

The CEO handles intake triage, cross-team coordination, escalation routing, and weekly summaries. Specialists cover coverage & advisory, research & modeling, fund admin & finance ops, and KYC. Every output is staged for human sign-off — agents do not execute trades, post to a ledger, or approve onboarding.

Company README →

Boundaries. Nothing in financial-services constitutes investment, legal, tax, or accounting advice. These agents draft analyst work product (models, memos, research notes, reconciliations) for review by a qualified professional. They do not make investment recommendations, execute transactions, bind risk, post to a ledger, or approve onboarding. You are responsible for verifying outputs and for compliance with the laws and regulations that apply to your firm. See the company's Boundaries section for full detail.

Community port. Not affiliated with or endorsed by Anthropic. "Claude" is a trademark of Anthropic, PBC.

Community port of Imbad0202/academic-research-skills (the ARS pipeline) into the Agent Companies format. Runtime-agnostic via the Paperclip adapter chain — Claude is the reference runtime, others run with varying skill polish.

npx companies.sh add stubbi/companies/academic-research

Agents	5 (1 CEO + 4 stage owners across 4 teams)
Skills	4 referenced upstream + 4 port-original (CEO-owned)
License
Source	Imbad0202/academic-research-skills at pinned commit 58dad474

The CEO handles intake triage, pipeline orchestration, checkpoint coordination, and escalation. Stage owners cover Research (Stage 1), Writing & Revision (Stages 2 / 4 / 4' / 5), Review (Stages 3 / 3'), and Integrity & Pipeline (Stages 2.5 / 4.5 / 6). Every output is staged for human sign-off — agents do not submit to journals, sign authorship statements, or attest to research integrity. The pipeline's two integrity gates and 7-mode AI failure checklist are mandatory and not skippable.

Company README →

⚠️ Non-commercial use only. Upstream is licensed CC BY-NC 4.0 and this port preserves that restriction. Personal research, academic work, and non-commercial collaboration: yes. Commercial services, paid offerings, work-for-hire: no. See the company's LICENSE and NOTICE.

Boundaries. Nothing in academic-research constitutes peer-reviewed scholarship on its own. These agents draft research artifacts (literature searches, methodology blueprints, outlines, drafts, review reports, formatted manuscripts) for review by a qualified human researcher. They do not submit to journals, sign authorship statements, make editorial decisions, or attest to research integrity on the user's behalf. See the company's Boundaries section for full detail.

Community port. Not affiliated with or endorsed by Imbad0202 or the ARS authors.

10-agent industrial-research lab modeled on Bell Labs (1925–1984). Original synthesis — no upstream port. Separated patron (CEO) and instigator (Director of Research); five-archetype researcher coverage plus boundary roles (Systems Engineer, Librarian) that every modern Bell Labs imitator has dropped. Two-track operation (directed + curiosity, curiosity protected by policy); forced cross-archetype traversal encoded as policy via the Hallway, weekly Colloquium, and Director's walks. Output is the Technical Memorandum.

npx companies.sh add stubbi/companies/bell-labs

The CEO is patron, not boss — onboarding interview, mission protection, intake routing, monthly summaries. The Director of Research is the lab's taste organ — daily Hallway walks, Pierce-style instigation questions (rate-limited), continuation reviews. Researchers cover theorist, experimentalist, inventor, translator, materials/empiricist, mathematician (25%-time consultant), plus a Systems Engineer who brokers problems from the user's real "network" and a Librarian who pushes prior memoranda into active threads. Every researcher runs a Directed queue and a self-owned Curiosity queue; the curiosity queue is protected by policy and make check.

Company README →

Boundaries. This is a research lab. It drafts memos, prototypes, proposals, and handoff docs for human review. It does not ship to the user's production systems on its own authority, file actual patents, submit to actual journals, or claim peer-review. The wild-duck / curiosity track is freedom of method, not freedom of fence. See the company's Boundaries section for full detail.

Name homage. Not affiliated with or endorsed by Nokia / Nokia Bell Labs. "Bell Labs" is a trademark of Nokia.

Community port of anthropics/claude-for-legal (Anthropic's "Claude for Legal") into the Agent Companies format. Covers 12 practice-area plugins — commercial, corporate, employment, privacy, product, regulatory, AI governance, IP, litigation, plus law-student / legal-clinic / legal-builder-hub on the education side. Runtime-agnostic via the Paperclip adapter chain; Claude is the reference runtime.

npx companies.sh add stubbi/companies/legal-services

The CEO handles intake triage, cross-practice coordination, escalation routing, and weekly summaries. Specialists cover commercial & transactions (commercial / corporate / employment), regulatory & compliance (privacy / AI governance / regulatory / product), IP & disputes (ip / litigation), and legal academy & tooling (law-student / legal-clinic / legal-builder-hub). Every output is staged for attorney sign-off — agents do not file briefs, send demand letters, issue legal holds, or take legal positions on behalf of any party.

Company README →

Boundaries. Nothing in legal-services constitutes legal advice. These agents draft work product (memos, redlines, claim charts, deposition outlines, review reports, policies, classifications) for review by a qualified attorney. They do not file briefs, send demand letters, issue legal holds, or take positions on behalf of any party; every output is staged for attorney sign-off. See the company's Boundaries section for full detail.

Community port. Not affiliated with or endorsed by Anthropic. "Claude" is a trademark of Anthropic, PBC.

• Top-level wrapper is MIT — the README, layout, top-level docs, and CI scaffolding.
• Each company ships its own LICENSE and NOTICE. Read those before deploying — community ports preserve upstream license terms (typically Apache-2.0 for Anthropic ports).
• Regulated domains (finance, legal, medical) have a Boundaries section in the company README. Read it before assigning real work.
• Bumping a port: cd <company> && make bump SHA=<new-sha> rewrites the manifest, regenerates all artifacts, and re-validates content hashes against the new upstream commit.
• Validation: every company has make build, make check, and make test targets. CI runs make test and make check on every PR.

This is a personal catalog with one maintainer and one taste. I'm happy to merge well-scoped fixes — typos, link rot, schema drift, upstream SHA bumps. For new companies, the cleanest path is your own <your-username>/companies repo: companies.sh installs are repo-agnostic, so you don't need a PR here. Self-published catalogs let you keep your own bump cadence and curatorial voice.

If you do want to contribute to this catalog, see CONTRIBUTING.md.

• Paperclip — open-source orchestration for zero-human companies. The runtime these companies plug into.
• paperclipai/companies — official Paperclip catalog (16+ companies). Companies submitted there go through maintainer review.
• VectifyAI/PageIndex — vectorless tree-traversal retrieval over long PDFs. Useful when your agents work with long professional documents.

Top-level wrapper: MIT. Individual companies have their own licenses — see each company's LICENSE file. Where individual companies operate in regulated domains (finance, legal, medical), each company's README documents the appropriate boundaries — read those before deploying.