AI coding agent skill for the First Principles Framework (FPF) by Anatoly Levenchuk.
FPF is a transdisciplinary reasoning architecture for systems engineering, knowledge coordination, and mixed human/AI teams.
FPF is a thinking amplifier — it helps you plan deeper and make better decisions by systematically exploring relevant alternatives instead of anchoring on the first idea.
This skill functions as agentic RAG — retrieval-augmented generation driven by the agent itself, with no external vector database or embedding pipeline. The upstream FPF specification is split into a two-level hierarchy (15 directories, 404 files). SKILL.md provides a thinking-verb router that maps the user's intent to the right section, including choosing a practical entry, applying a pattern to its first useful result, discovering Systems that may bear consequences, recovering candidate Methods from Work evidence, disambiguating agent-relative instructions, steering or resuming ongoing Work, making applicable methods noticeable, initiating inquiry from ordinary work, coordinating development states, bridging domain terms, recovery-testing Work-performance configurations, screening mandatory workflow steps for operational relevance, comparing finite configuration changes, appraising advice and evidence demands for a recipient’s decision, clarifying learning/development/evolution claims, probing apparent capability loss before redevelopment, navigating a DPF Suite Reference, revalidating affected uses after source changes, requesting bounded specialist results, selecting representations for one use, constructing comparable ways to obtain one result, assembling framework publication forms or composite methods, synthesizing architecture across non-isomorphic practice structures, developing capability for named Work, synthesizing source ontologies without flattening their local meanings, recovering project/process/case subjects and roles, tracing outside-use dependencies, distinguishing relations from occurrences, transforming epistemes, reasoning about decisions, causal-model construction, actual temporal structure, time, architecture, and structural adequacy, publishing stable multi-view artifacts, governing ontic admission, refreshing SoTA packs, and tracing provenance. Current navigation also covers reader-extractable structural information (C.2.8), Architectural Rationale and profile choices, retargeting with separate bounded-use assertions and current-case judgments, and cross-context reuse with assurance requirements determined by the stated use. The agent then navigates _index.md files to pick the narrowest sub-section and loads only that into context. The agent is the retriever, the router, and the reasoner — all in one loop.
Navigation also covers how constituent actions enact encompassing work (B.1.5.EW), whether constituent-method replacements preserve their receiving uses (B.1.5.RS), first models and theory use, reasoning revision, physical/mathematical/computational connections, measurement construction and repair, object reidentification, predictive information, reusable operations, and branching search. Symmetry comparisons are covered by C.29.1, and clarification of model/explanation uses by E.10.
npx skills add CodeAlive-AI/fpf-problem-solving-skill -gsections/
05-part-a---kernel-architecture-cluster/
_index.md # TOC with descriptions of all sub-sections
01-a-0---generative-search-onboarding-glossary.md
02-a-1---recognize-a-whole-with-parts.md
... # 30 sub-sections total
09-part-c---kernel-extension-specifications/
_index.md
... # 94 sub-sections
... # 15 directories total
The agent reads _index.md first, picks the right sub-section file, and loads only that.
| # | Section | Sub-sections |
|---|---|---|
| 01 | Title page | 0 |
| 02 | Table of Contents | 0 |
| 03 | FPF Readme | 11 |
| 04 | Preface | 22 |
| 05 | Part A — Kernel Architecture | 30 |
| 06 | A.IV.A — Signature Stack & Boundary | 29 |
| 07 | A.V — Constitutional Principles | 49 |
| 08 | Part B — Trans-disciplinary Reasoning | 40 |
| 09 | Part C — Kernel Extensions | 96 |
| 10 | Part D — Ethics & Conflict | 5 |
| 11 | Part E — Constitution & Authoring | 69 |
| 12 | Part F — Unification Suite | 22 |
| 13 | Part G — SoTA Patterns Kit | 15 |
| 14 | Part H — Reserved | 0 |
| 15 | Part I — Annexes | 1 |
Counts follow upstream H2 headings. The Preface has 21 content sections plus the content-free FPF.Preface:End boundary marker.
When the upstream FPF specification changes, two things need updating:
Clone the official upstream ailev/FPF into a temporary skill layout outside this repository, then run the splitter there and replace the tracked sections/ tree with the generated result:
tmpdir="$(mktemp -d)"
mkdir -p "$tmpdir/skill/scripts"
git clone https://github.com/ailev/FPF.git "$tmpdir/skill/FPF"
cp scripts/split_spec.py "$tmpdir/skill/scripts/split_spec.py"
python3 "$tmpdir/skill/scripts/split_spec.py"
rsync -a --delete "$tmpdir/skill/sections/" sections/
rm -rf "$tmpdir"The section files are raw content — SKILL.md is the navigation layer on top.
After regenerating, review whether the thinking-verb router, use cases, or Section INDEX
in SKILL.md need updating to reflect new, changed, or removed content.
See FPF-SKILL-UPDATE-GUIDE.md for the full methodology: what to check, how to validate router entries, and how to run an FPF self-audit on the skill file itself.
- FPF specification: Anatoly Levenchuk — github.com/ailev/FPF
- Skill packaging: CodeAlive-AI
Skill packaging and splitter: MIT. Adapted from First Principles Framework (FPF) by Anatoly Levenchuk, ailev/FPF, upstream commit 5cc5731453a6ae4a2e264b1c4e3fc83b84a29f01. The FPF specification text is licensed under CC BY 4.0; see the upstream licensing scope. Changes: the specification was split into section files with generated navigation indexes, and SKILL.md adds an agent-oriented router and usage instructions. This adaptation is not endorsed by the FPF author.