autonomous-inventor
dr-armani/autonomous-inventor
A Claude skill that invents hardware and automation designs through a structured process.
Instalação
npx convez-ai dr-armani/autonomous-inventor1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
🔧 Autonomous Inventor
A Claude skill that invents hardware and automation designs through systematic contradiction resolution — not trial and error.
Anchor on a bad form. Iterate with TRIZ. Reframe the architecture. Loop until the hard problems are designed out — not patched. Then optimize, and optionally file a patent.
Why this exists
Most invention methods are linear: brainstorm → prototype → test → repeat. They add parts, tune parameters, and call it done.
This skill does something different. It names the contradiction generating the hardest problems, resolves it architecturally using TRIZ, stress-tests the result, and loops back to reframe again if a structural flaw surfaces. One good reframe eliminates clusters of problems that more iteration never could.
When the design converges, it generates a Python optimization script for numerical trade-offs and — optionally — a complete USPTO provisional patent application ready to file.
Design quality is proportional to how precisely you name the contradiction.
Method
9 phases · 2 feedback loops · TRIZ throughout
0 Contradictions → 1 Baseline → 2 Iterate → 3 Reframe (TRIZ) → 2B Re-iterate*
▲ │
│ loop if structural │
└──── 4 Stress-test ◀┘
│
5 Specialize → 6 Optimize → 7 Final design → 8 Patent†
* Phase 2B runs only when the reframe shifts the whole architecture. The 3↔4 loop is the engine — expect to traverse it more than once. † Phase 8 is offered after Phase 7, or run upfront if requested.
| Phase | Name | What happens |
|---|---|---|
| 0 | Contradiction inventory | Name technical and physical contradictions before designing. Vague contradictions produce vague reframes. |
| 1 | Baseline | Humanoid anchor — deliberately wrong. Every later part must earn its place. Skipped with a reason when absurd. |
| 2 | Iterate (~10 cycles) | − remove · + add · ± adjust, each annotated with TRIZ principles. Exposes which contradictions are actually hard. |
| 3 | Reframe | State contradiction precisely → choose separation strategy → apply TRIZ → name what's eliminated. The multiplier step. |
| 2B | Re-iterate (conditional) | 5–10 fresh cycles on the new architecture if the reframe invalidated more than half the Phase 2 decisions. |
| 4 | Stress-test | Brutal failure-mode analysis across 6 dimensions. Triggers another reframe if a structural flaw surfaces. |
| 5 | Specialize + autonomy | Define the bounded operating domain. Mechanics solves; AI decides. "Refuse safely" and "detect-and-escalate" are features. |
| 6 | Optimize | Tune within the fixed architecture: FMEA, efficiency, cost/weight, manufacturability, final TRIZ sweep — plus a Python scipy.optimize script for numerical trade-offs. |
| 7 | Final design | Name · components · full cycle · non-obvious moves · TRIZ citation log · reframe history · residual limits · niche |
| 8 | Provisional patent | Complete USPTO provisional application — cover sheet, background, summary, ASCII drawings with reference numerals, detailed description, claims, abstract, filing checklist. |
TRIZ integration
The skill embeds the full TRIZ 40 Inventive Principles woven throughout every phase — not bolted on at the end.
| Where | How TRIZ is used |
|---|---|
| Phase 2 | Each −/+/± move annotated with principle(s) — builds auditable design rationale |
| Phase 3 | Contradiction → separation strategy → goal-to-principle lookup table |
| Phase 4 | P22 (Blessing in disguise) and P11 (Cushion in advance) as stress-test lens |
| Phase 5 | P10, P15, P24, P25 guide the mechanics-vs-AI split |
| Phase 6 | Final sweep with P2, P6, P20 to surface remaining redundancy |
| Phase 7 | TRIZ citation log in the final design output |
All 40 principles grouped by cluster — designed to be scanned mid-invention:
Segmentation & Structure: Segmentation · Extraction · Local quality · Asymmetry · Merging · Universality · Nesting · Anti-weight
Dynamics & Adaptability: Preliminary anti-action · Preliminary action · Cushion in advance · Equipotentiality · Inversion · Spheroidality · Dynamics · Partial/excessive action
Geometry & Dimension: Another dimension · Mechanical vibration · Periodic action · Continuity · Skipping · Blessing in disguise · Feedback · Intermediary
Surface & Interface: Self-service · Copying · Cheap short-living · Mechanics substitution · Pneumatics/hydraulics · Flexible shells
Physical & Material Effects: Porous materials · Color changes · Homogeneity · Discarding & recovering · Parameter changes · Phase transitions · Thermal expansion · Strong oxidants · Inert atmosphere · Composite materials
Full table with core moves for each principle is in SKILL.md.
When one element must simultaneously be X and not-X:
| Strategy | How | Example |
|---|---|---|
| In space | X in one zone, not-X in another | Hard at the contact face, soft behind it |
| In time | X at one moment, not-X at another | Rigid during torque, compliant during alignment |
| By condition | X in one state, not-X in another | Stiff under load, flexible under none |
| By system level | X at part level, not-X at system level | Each spindle floats; the ring is rigid |
| Tension | Try first |
|---|---|
| Strength vs. weight | P1, P7, P8, P40 |
| Precision vs. speed | P10, P15, P21, P28 |
| Stability vs. mobility | P5, P8, P12, P13 |
| Torque vs. reaction force | P5, P13, P14, P25 |
| Sensing reliability vs. environment | P3, P10, P15, P23 |
| Grip force vs. surface damage | P3, P9, P16, P24 |
| Complexity vs. reliability | P2, P6, P25, P27 |
| Range vs. accuracy | P15, P16, P17, P23 |
| Cost vs. performance | P6, P26, P27, P35 |
| Autonomy vs. safety | P11, P23, P24, P25 |
Python optimization (Phase 6)
When the design has coupled numerical parameters, the skill generates a ready-to-run scipy.optimize script:
# Objective function, physical constraints, and bounds
result = differential_evolution(objective, bounds, seed=42, tol=1e-8)
# Results table: parameter | optimal value | units | note
# Sensitivity block: ±10% perturbation → % impact on objective
Covers: motor sizing, gear ratios, beam dimensions, cost/weight/speed trade-offs, thermal margins. Multi-objective (Pareto) sweep when two objectives conflict. Skipped with a one-line reason if no meaningful numerical parameters exist.
Provisional patent (Phase 8)
After Phase 7, the skill optionally generates a complete USPTO provisional utility patent application establishing a 12-month priority date (35 U.S.C. 111(b)):
| Section | Content |
|---|---|
| Cover sheet | USPTO Form SB/16 data |
| Background | Field, prior art, problem statement drawn from Phase 0 contradictions |
| Summary | Broadest inventive concept, advantages linked to Phase 3 reframe payoffs |
| Drawings | ASCII line drawings with reference numerals; draftsperson-ready descriptions |
| Detailed description | Component-by-component, operating cycle, 2–3 alternative embodiments (enables POSITA per 35 U.S.C. 112) |
| Claims | ≥1 independent + 3–5 dependent claims (apparatus and method formats) |
| Abstract | ≤150 words |
| Filing checklist | Patent Center URL, fee note, drawings spec (37 C.F.R. 1.84), 12-month non-provisional deadline |
This is a drafting aid, not legal advice. Review with a registered patent attorney or agent before filing.
Core principles
| Principle | Why it matters |
|---|---|
| Contradictions are the signal | Where something is hard to design, a contradiction hides. Name it; don't just add parts. |
| Loop on the reframe | The 3↔4 cycle is where good designs become great. One reframe is rarely enough. |
| Mechanics before AI | Compliance, self-location, and passive alignment beat precision perception. |
| Quantify what matters | Loads, torques, tolerances, masses, cycle time, cost. Estimates beat hand-waving; flag them. |
| Honesty over salesmanship | A design with stated limits beats one that claims everything. |
| Function dictates form | Reject anthropomorphism unless a humanoid feature genuinely earns its place. |
Worked example
Task: change a flat tire autonomously.
Contradiction (Phase 0): high torque needs a heavy reaction base — but heavy hurts mobility. The base must be both heavy (stability) and light (portability).
Reframe (Phase 3): react torque internally through a multi-spindle nutrunner ring [P5, P13]; re-reference the whole machine off the hub rather than the floor [P13]; let mechanical compliance absorb alignment [P10, P15]. Eliminates jack instability, grounded-torque dependence, and sub-millimeter sensing — in one architectural move.
Re-iterate (Phase 2B): + multi-spindle ring · + hub-clamp reaction collar · + RCC wrist · − ground outriggers · − heavy ballast
Optimize (Phase 6): parallel-stage the spare while lifting [P20]; FMEA hardens the lift lock. Python script optimizes spindle count vs. ring mass vs. motor cost via differential_evolution on a weighted objective (cycle time × mass × cost).
Patent (Phase 8): independent claim on hub-clamp reaction collar as torque reaction mechanism; ASCII figures of ring assembly and hub clamp; 4 dependent claims on spindle count, chuck geometry, RCC compliance, and autonomous control mode.
Final design — PitCrew-A. TRIZ citation log: P2 · P5 · P10 · P13 · P15 · P20 · P22 · P25
Installation
One-click:
- Download
autonomous-inventor.skill - Open Claude → Settings → Capabilities → Skills (how to install)
- Upload the file
From source: copy SKILL.md into your Claude skill directory.
Usage
Triggers automatically on any request to invent or design equipment:
"Invent a device to harvest ripe strawberries autonomously."
"Design equipment to clear snow from residential sidewalks."
"Come up with a machine that sorts recycling at the curb."
"How would you build something to inspect underwater pipelines?"
"I need a contraption that automatically loads pallets onto trucks."
Add "and file a patent" to any prompt to run Phase 8 after the design converges.
Repository
autonomous-inventor/
├── LICENSE
├── README.md
├── SKILL.md ← full method + TRIZ appendices (human-readable)
└── autonomous-inventor.skill ← packaged for one-click install
License
MIT — see LICENSE.
Built by Daniel Armani
CEO, Texas Blockchain Center · Author, Better Governance (Springer Nature, 2026)