triz-engineering-solver

A Claude / AI-agent skill for solving engineering problems with Altshuller's TRIZ (Theory of Inventive Problem Solving). Replaces compromise-driven brainstorming with algorithmic problem solving over a corpus of patent-derived patterns.

What it does

Given an engineering contradiction — "improving X makes Y worse" or "this part must be both A and not-A" — the skill produces 3–5 concrete inventive concepts with explicit ideality scores rather than compromise solutions. It uses:

• Ideal Final Result (IFR) framing
• The 39 engineering parameters and 40 inventive principles distilled by Altshuller from ~200,000 patents
• The full 39×39 Contradiction Matrix (1190 populated cells, anchor-verified against Altshuller 1985)
• Su-Field analysis with the canonical 76 Standard Solutions (5-class taxonomy, Su-Field algebra notation)
• Separation principles (space / time / condition / system-level) for physical contradictions
• ARIZ-85C deep procedure when the quick pass fails the ideality bar
• The 8 trends of engineering system evolution for roadmapping questions

When to use

• Engineering trade-offs where improving parameter A degrades parameter B (technical contradiction)
• A single element must exhibit opposite properties (physical contradiction)
• Design bottleneck where conventional optimisation has plateaued
• System redesign aimed at the Ideal Final Result

When NOT to use

• Pure software architecture without a physical analogue
• UX / interaction design
• Business or organisational strategy
• Open brainstorming with no concrete contradiction identified

The skill refuses with reframe when invoked on out-of-scope problems — see examples/anti_example_misframed.md.

Install

For Claude Code:

git clone https://github.com/Antropocosmist/triz-engineering-solver \
  ~/.claude/skills/triz-engineering-solver

For other agent runtimes (Anthropic SDK, OpenAI Assistants, LangGraph, custom): point the agent's system prompt or tool router at SKILL.md and grant filesystem read access to this directory.

Repo layout

.
├── README.md
├── LICENSE                          MIT
├── CONTRIBUTING.md                  anchor-cell verification protocol, style rules
├── SKILL.md                         entry point — workflow, triggers, output template
├── resources/                       lazily-loaded reference data
│   ├── 39_parameters.md
│   ├── 40_principles.md
│   ├── 76_standard_solutions.md     full 5-class taxonomy with Su-Field algebra
│   ├── contradiction_matrix.json    full 39×39 Altshuller matrix (1190 cells)
│   ├── separation_principles.md     decision procedure for physical contradictions
│   ├── ariz_85c.md                  9-part deep-analysis procedure
│   ├── evolution_trends.md          8 trends + S-curve framing
│   ├── glossary.md
│   └── output_template.md           machine-readable output template (use verbatim)
└── examples/
    ├── brake_disc.md                mechanical / thermal contradiction
    ├── battery_pack.md              physical contradiction via separation
    ├── heat_exchanger_fouling.md    process-industry contradiction + ideality drop
    └── anti_example_misframed.md    refuse-with-reframe demo

Design properties

• Deterministic where possible. Decision points are reduced to lookups in structured resources, not free-form reasoning.
• Source-anchored. Every claim, value, or principle traces back to a primary source. Contributions to the contradiction matrix must pass a 5-cell anchor verification protocol (see CONTRIBUTING.md).
• Refuse-with-reframe. Out-of-scope problems are rejected explicitly, with a hint at where the user should go instead.
• No compromise. Concepts with ideality ≤ 1 are dropped, not split-the-difference accepted. If no concept clears the bar, the skill escalates to ARIZ-85C rather than weaken the recommendation.

Status

v1.0 — release-ready. Full Altshuller 39×39 matrix (1190 cells, 5/5 anchors verified). Three worked examples + one anti-example. Su-Field analysis with the full 76 Standard Solutions. ARIZ-85C deep procedure. 8 evolution trends. Workflow documented end-to-end.

Contributing

See CONTRIBUTING.md. Examples from non-mechanical domains (electronics, chemical engineering, optics, biotech) are particularly welcome.

Attribution

• TRIZ methodology: Genrich Altshuller (1926–1998), originator. Primary sources: Creativity as an Exact Science (1979/1984), The Innovation Algorithm (1999). Modern reference: Darrell Mann, Hands-On Systematic Innovation (2002). Su-Field analysis: Yuri Salamatov, TRIZ: The Right Solution at the Right Time (1999).
• Contradiction matrix data: imported from the MIT-licensed kamil-szczepanik/TRIZ-Agents repository (data/tools_sources/triz_matrix.xls, Casey Perno 2007 transcription of Altshuller 1985). Cite the accompanying ICAART 2025 paper if you build on their work: Szczepanik et al., "TRIZ Agents: A Multi-Agent LLM Approach for TRIZ-Based Innovation," 2025 — arXiv:2506.18783.

License

MIT — see LICENSE.

Authors

• Antropocosmist
• Claude