Prompts, rulebooks, how-tos, and lessons for building things with language models.
Guides distilled from how real people think and work. For humans to read.
How to Tell Visual Stories with a Camera
how-to/how-to-collier-landry.md
55 transferable rules on visual storytelling, craft, career, and creative resilience, distilled from Collier Landry's record as a director of photography and filmmaker.
How to Design Minimal Safe C++ APIs
how-to/how-to-dimov.md
Reference for evaluating C++ designs the way Peter Dimov does - API minimalism, safe defaults, undefined-behavior elimination, contracts, type invariants, and generic programming.
How to Build AI Tools and Prompts
how-to/how-to-falco.md
Build AI tools, super prompts, and plan files that produce reliable output across runs and fresh contexts.
How to Design C++ Libraries for Review
how-to/how-to-glen.md
Distilled from Glen Fernandes' record on C++ library design, allocator models, pointer utilities, alignment, API hygiene, and community review practice.
How to Build Software with AI Without Cutting Corners
how-to/how-to-greg-bildson.md
63 directives on steering AI coding sessions, durable plans and records, design judgment, security and data governance, teams, network health, and business strategy, distilled from Greg Bildson's written record.
How to Design Precise Technical Interfaces
how-to/how-to-klemens.md
Reference for evaluating technical designs through precise naming, ownership enforcement, minimal API surfaces, evidence-based evolution, and verifiable review criteria.
How to Build Software With Coding Agents Without Losing Control
how-to/how-to-mario-zechner.md
98 directives on refusing scope, keeping architecture in human hands, delegating bounded agent work, reviewing generated output, engineering context, and owning the harness, distilled from Mario Zechner's record.
How to Investigate and Hold Power Accountable
how-to/how-to-nowosielski.md
96 rules distilled from a documentarian's two decades of investigative journalism - developing sources, verifying evidence, parsing official language, detecting cover-ups, and surviving adversarial pressure.
How to Build AI Agents That Work in Production
how-to/how-to-peak-ji.md
63 directives on KV-cache discipline, memory and compression, agent behavior and trust, action space, evaluation, autonomy, and product strategy, distilled from Yichao "Peak" Ji's record.
How to Evaluate Language Change Proposals
how-to/how-to-spicer.md
100 directives on language change drawn from a long record of committee argument - silent breakage, special-case cost, specification, implementability, evidence discipline, and consensus timing.
How to Navigate C++ Committee Design Decisions
how-to/how-to-ville.md
Reference for evaluating designs the way Ville Voutilainen does - rationale, committee process, implementation experience, compatibility, language-design principles, and safety.
How to Evaluate C++ Language and Library Designs
how-to/how-to-vinnie.md
Reference for evaluating C++ designs through evidence, user value, minimal scope, semantic interfaces, layering, generic composition, dependency control, and measurable cost.
How to Build and Run Web Applications
how-to/how-to-wiles.md
87 directives on building and running web applications drawn from consulting practice - query measurement, doing less work, testing, automation, conventions, observability, and project structure.
Style and format references for applying to prompts and plans.
Writing C++
rulebooks/cpp-rulebook.md
Reference for writing or maintaining C++ - project layout, naming, headers, RAII, smart pointers, error handling, modern features, templates, concurrency, testing, tooling, and performance.
Writing Paper Abstracts
rulebooks/create-abstracts-rulebook.md
Write the abstract for a WG21 paper as one actionable finding stated first, the surface a delegate reads before anything else.
Writing Fiction
rulebooks/fiction-rulebook.md
Rules for writing fiction that reads as if a human wrote it. Read this document to learn the craft; give it to a model to apply it.
Writing HTML and CSS
rulebooks/html-css-rulebook.md
Reference for writing, reviewing, or cleaning up HTML and CSS - semantic markup, forms, media, modern layout, custom properties, cascade layers, accessibility, responsive design, and performance.
Writing JavaScript
rulebooks/javascript-rulebook.md
Reference for writing, reviewing, or cleaning up modern JavaScript - language syntax, modules, async, errors, DOM and platform APIs, Node.js, security, testing, and linting.
Maintaining Clang and LLVM
rulebooks/maintain-clang-rulebook.md
Reference for making changes to Clang and LLVM source - architecture, conventions, core APIs, extension recipes, testing, and build/debug workflow.
Maintaining GCC
rulebooks/maintain-gcc-rulebook.md
Reference for making changes to GCC source - architecture, conventions, core APIs, extension recipes, testing, and build/debug workflow.
Writing News Stories
rulebooks/news-rulebook.md
Write or audit a news article, feature, analysis, briefing, or press release for structure, leads, sourcing, style, and objectivity.
Writing Technical Papers
rulebooks/papers-rulebook.md
Draft a WG21 paper for a delegate who reads in passes and stops when a pass fails: show, then assert.
Writing Unambiguous Model Instructions
rulebooks/prompts-rulebook.md
Write or audit a prompt, plan, tool, or rule file for unambiguous instructions and efficient context management.
Revising Model Prose to Human Standard
rulebooks/prose-rulebook.md
Rules for revising model-generated prose so it reads as human-written, in sequential editing passes from structure to wording.
Writing Python
rulebooks/python-rulebook.md
Reference for writing or maintaining Python - project layout, packaging, naming, idioms, typing, errors, API design, imports, dependencies, documentation, testing, tooling, performance, concurrency, and anti-patterns.
Reading Technical Papers
rulebooks/read-papers-rulebook.md
Evaluate a paper's quality in three sequential passes, each with exit criteria, from the general idea to the depth a verdict needs.
Structuring Reports
rulebooks/reports-rulebook.md
Write or audit a report of any type for structure, evidence, uncertainty, sourcing, and format discipline.
Writing Rust
rulebooks/rust-rulebook.md
Reference for writing or maintaining Rust - project layout, naming, ownership, API design, errors, crates, documentation, testing, tooling, performance, async, and unsafe.
Writing TypeScript
rulebooks/typescript-rulebook.md
Reference for writing, reviewing, or cleaning up TypeScript - configuration, type discipline, imports, errors, async, validation, naming, testing, linting, and publishing.
Vibe-Coding with Minimal Prompting
rulebooks/vibe-rulebook.md
Turn a described idea into finished, tested software - plan in levels of resolution, build one testable commit at a time with two work subagents, review-and-fix once, verify on a schedule, and drive to completion.
General-purpose self-contained prompts.
The AI Reasoner
tools/ai-reasoner.md
Takes an argument about whether LLMs can reason and delivers a steelmanned rebuttal from the position that reasoning is interpolation of learned patterns.
The Architect
tools/architect.md
Exacting interactive designer that turns an idea, an existing codebase, or an old design document into a design document any proficient developer or LLM can build from without inventing a decision.
The Booksmith
tools/booksmith.md
Conversational tool that writes your story by reusing the sentence structures of existing published works.
The Briefer
tools/briefer.md
Given an institution or system, produces a structural diagnosis with prognosis, compound dynamics, and conditional predictions across three time horizons.
btc-talk
tools/btc-talk.md
Takes a Bitcoin governance discussion and delivers a diagnosis of reasoning pathologies grounded in quoted evidence, tiered challenger tests, and bidirectional fairness checks.
The Cabinet
tools/cabinet.md
On load, becomes a filing clerk that reads your workspace schema, classifies documents, routes them to correct directories, and maintains frontmatter and naming conventions.
Chatlight
tools/chatlight.md
Export a Cursor or Claude Code chat to clean markdown - skips thinking bubbles, renders the conversation as the user saw it.
CoC Blocker
tools/coc-blocker.md
Transforms conversations into a dossier of ISO Code of Conduct violations with attributed quotes, severity ratings, cross-source patterns, and an amplification map.
Vauban the Converger
tools/converge.md
Given a goal, produces either an architecture of inevitability that survives adversarial testing at every step, or an honest stop-and-redirect when the goal cannot be made inevitable.
The Diligence
tools/diligence.md
Writes a hire/hire-with-conditions/avoid verdict from a firm's public record, 30 diagnostic tests, and compound risk analysis.
Dokuman
tools/dokuman.md
Points at any repository, folder, or file set and extracts its capabilities into tiered progressive-disclosure documentation - the thirty-second pitch, the five-minute orientation, the full mechanical exposition.
The Dramaturg
tools/dramaturg.md
Takes prose and delivers a catalog of machine-generated syntactic patterns with keep/rewrite verdicts, or rewrites the text directly with each edit checked by an internal quality gate.
The German
tools/german.md
Transforms a tool specification into a scored audit of 65 structural rules with specific edits prescribed for each failure.
The Consultant
tools/hubris.md
Given a person who holds power, produces a clinical hubris syndrome assessment scored against Owen's fourteen criteria.
Adaptive Interview Generator
tools/interview.md
Transforms a role description into a reusable AI interview protocol with scenario templates, scoring rubrics, and a state machine.
Make How-To
tools/make-how-to.md
Build a how-to manual that teaches how a named person thinks and works, from their written record - parallel search, parallel distillation into plain-bullet rules, then mechanical assembly, dedupe, trim, and thematic grouping.
The Mentographist
tools/mentograph.md
Takes a willing subject through a story-walk conversation and delivers a verbatim timestamped transcript capturing how they think.
The Nash Equilibrium
tools/nash.md
Writes a structured game-theory document - players, moves, payoffs, equilibria - from freeform conversation about strategic interactions.
The Prompt Normalizer
tools/normalize-prompt.md
Given a prose tool definition, produces a machine-parseable pipeline spec with typed state fields and a dependency DAG.
refactor-rust
tools/refactor-rust.md
Reviews the last week of Rust commits and applies API and hygiene fixes on one candidate branch, verified in a fresh challenger context against cargo public-api and the workspace gates before landing.
Refine Plan
tools/refine-plan.md
Rewrites a plan through three passes - number, audit, compress - into a tighter version of itself.
The Research Desk
tools/research.md
Given a knowledge gap, produces a research brief with sourced finding cards from parallel web foragers.
The SIGINT Analyst
tools/sigint-analyst.md
Takes intercepted communications and delivers a dual-hypothesis intelligence report weighing civilian good faith against political tradecraft.
Skillgate
tools/skillgate.md
Score an engineer's prompting quality from their chat transcripts - reads how each prompt treats the model's previous reply and writes one self-contained report per person.
The Staker
tools/staker.md
Generates a stakeholder analysis report on any institution, organization, or cohort.
The Pedagogue
tools/tutor.md
Transforms a learning goal into a chain of self-contained tutor files, each carrying a real teacher's voice, strict milestones, and vetted URLs.
The Voice
tools/voice.md
Given a name, produces a self-contained voice file that, when loaded, speaks as that person - vocabulary, emotional architecture, life arc, and all.
What to Steal
tools/what-to-steal.md
Profiles a codebase, surveys the popular projects sharing its stack, dives their architecture at pinned commits, tags every idiom's human-or-machine provenance, and reports what to steal ranked by payoff.
Boost Library Review
tools/code/boost-review.md
Given a C++ library repository, produces a competitive design review scored against historical Boost rejection patterns.
The Jannie
tools/code/code-cleanup.md
Takes source code and delivers a numbered repair list where each entry is a two-sentence diagnosis and fix.
Code Review
tools/code/code-review.md
Transforms source files into a challenge-hardened review report with findings sorted by must-fix, should-fix, and nice-to-have.
The Reviewer
tools/code/lib-review.md
Writes a quality-verdicted design diagnosis from an open-source project's source code, domain context, and competitive landscape.
Interview Protocol: Entry-Level Agentic Pipeline Engineer
tools/interview/interview-agentic-pipeline-engineer.md
Takes a candidate through a 4-state adaptive interview scoring 10 cognitive dimensions, then produces a structured assessment with composite score and hiring signal.
Interview Protocol: Python Pipeline Developer
tools/interview/interview-python-pipeline-developer.md
Transforms a candidate conversation into a structured assessment report scoring 9 weighted dimensions through adaptive scenario generation.
The Novelist
tools/novelist/novelist.md
Writes structured story bibles from manuscripts or conversation, then authors chapters against them.
The Audit
tools/novelist/audit.md
Given a story bible, produces per-chapter findings on density, structure, constraint compliance, continuity, and schema.
The Critic
tools/novelist/critic.md
Takes a manuscript and delivers a four-pillar book review with an audit trail of every craft question asked.
The Edit
tools/novelist/edit.md
Transforms a flagged chapter into a tightened one where only unambiguously better changes survive.
The Workshop
tools/novelist/workshop.md
Walks a chapter through an interactive flag-by-flag session where the author writes every fix and the tool attacks every insertion.
Pen of William Gibson
tools/novelist/pen/pen-of-william-gibson.md
On load, becomes William Gibson's prose voice.
Writing Instructions
tools/novelist/prompt/anti-tic.md
Constrains every sentence against thirty-plus forbidden AI writing patterns.
Reviser
tools/novelist/prompt/remove-tic.md
Takes a draft chapter and delivers a clean revision with tics removed and dead weight cut.
Architect
tools/novelist/prompt/structure.md
Transforms a chapter specification into a scaffold of beats, omissions, register shifts, and silences.
Writer
tools/novelist/prompt/system.md
On load, becomes the prose renderer that converts bible specs into chapter fiction.
Novelist Tools
tools/novelist/CLAUDE.md
Guards the novelist pipeline by mapping every tool coupling so no change silently breaks another.
The Tutor: Applied Async Concurrency Patterns
tools/tutor/tutor-async-concurrency-patterns.md
Given a developer, produces mastery of asyncio gather, semaphore-capped fan-out, and error-classified resilience through 8 milestones in David Beazley's voice.
The Tutor: pydantic_ai-Orchestrated LLM Pipelines
tools/tutor/tutor-pydantic-ai-orchestration.md
Transforms a prepared developer into a practitioner of pydantic_ai pipeline orchestration across 8 milestones in Samuel Colvin's voice.
The Tutor: Pydantic v2 Data Modeling
tools/tutor/tutor-pydantic-v2-modeling.md
Takes a developer and delivers mastery of Pydantic v2 through 7 milestones in Raymond Hettinger's voice with no advancement without demonstrated understanding.
The Tutor: Python for Beginners (Arabic)
tools/tutor/tutor-python-for-beginners-arabic.md
Writes a complete Python foundation in Modern Standard Arabic across 7 mastery-gated milestones in Mosh Hamedani's voice.
The Tutor: Python for Beginners
tools/tutor/tutor-python-for-beginners.md
Given an absolute beginner, produces a working Python programmer through 7 milestones in Mosh Hamedani's patient contrast-based voice.
The Tutor: Rust Programming
tools/tutor/tutor-rust-programming.md
Takes an experienced developer new to Rust through 8 milestones in Steve Klabnik's compiler-as-mentor voice - ownership, borrowing, lifetimes, traits, and fearless concurrency.
Voice of Carl Gustav Jung
tools/voice/voice-of-carl-jung.md
On load, becomes Jung - analytical psychologist who speaks through semicolon-cascades, qualifying pivots, and mythic overlays drawn from alchemy and the collective unconscious.
Voice of Elias Canetti
tools/voice/voice-of-elias-canetti.md
On load, becomes Canetti - the anatomist of crowds and power who speaks in sovereign declarations, paratactic stacks, and body-to-politics analogies.
Voice of Franz Kafka
tools/voice/voice-of-franz-kafka.md
Transforms the model into Kafka - the Prague clerk who writes through chain-buts, hypotactic nests, and self-prosecution, placing himself at the bottom of every scale he constructs.
Voice of Ludwig von Mises
tools/voice/voice-of-ludwig-von-mises.md
On load, becomes Mises - the Austrian economist who speaks in adversative pivots, chain-deductions, and apodictic certainties defending the unhampered market.
Voice of Niccolo Machiavelli
tools/voice/voice-of-niccolo-machiavelli.md
Writes every response as Machiavelli - the exiled Florentine secretary who chains clauses with antithesis pairs and escalates every anecdote to a permanent law of politics.
Voice of Friedrich Wilhelm Nietzsche
tools/voice/voice-of-nietzsche.md
On load, becomes Nietzsche - the philologist-philosopher who speaks through dash-pivots, anaphoric chains, and mock-grandiosity, elevating every observation to superlative principle while undercutting it with comic domesticity.
Voice of Sigmund Freud
tools/voice/voice-of-sigmund-freud.md
On load, becomes Freud - the Viennese conquistador of the unconscious who builds arguments through hedge-then-assert and epistemic layering, landing paragraphs on a single aphoristic sentence.
Voice of Thomas Jefferson
tools/voice/voice-of-thomas-jefferson.md
Transforms the model into Jefferson - the Virginia polymath who suspends main clauses behind sixty words of subordination and drops axioms as self-evident premises.
Voice of Alexis de Tocqueville
tools/voice/voice-of-tocqueville.md
On load, becomes Tocqueville - the aristocratic liberal who analyzes democracy through antithesis-pivots and concession-ratchets with composed melancholy about equality's cost.
Voice of Wilfred Ruprecht Bion
tools/voice/voice-of-wilfred-bion.md
Writes every response as Bion - the tank commander turned psychoanalyst who teaches through progressive analogies and devastating questions, certain only that he does not understand.
Voice of William Gibson
tools/voice/voice-of-william-gibson.md
On load, becomes Gibson - the author-consciousness who renders emotion as body-event, character as pocket-dump, and technology as surface texture across four era registers.
Tools specific to the C++ Standards Committee (WG21).
Advocatus Diaboli
tools-wg21/advocatus.md
Given a WG21 paper, produces a sealed tribunal verdict with surviving objections, certified strengths, and a citation resolution table.
Assay
tools-wg21/assay.md
Takes a WG21 proposal and delivers a thesis-derived structural verdict with compound dynamics, citation verification, and per-lens findings.
The Auditor
tools-wg21/auditor.md
Writes a compliance-checked paper and an inspection report from a WG21 manuscript's raw markdown.
Entropotros
tools-wg21/entropotros.md
A debate host that holds one position on the C++ committee and defends it with the committee's own record, and teaches the same analysis to a receptive learner.
The Herald
tools-wg21/herald.md
Given a name in WG21, produces a heraldic pedigree with order, rank, epithets, and a sealed verification from the public record.
Is This C++?
tools-wg21/is-this-cpp.md
Given a C++ proposal, produces a scored verdict on whether it embodies the language's design principles.
The Papersmith
tools-wg21/papersmith.md
Writes WG21 papers through a six-step pipeline (commission, research, skeleton, body, surface, review) and reviews any paper through a reusable Review Process.
Reform Reviewer
tools-wg21/reform-reviewer.md
Takes a reform document and delivers a green/red framing report with rewrites and a verdict.
Review Paper
tools-wg21/review-paper.md
Writes a feedback report from a committee paper by testing every claim against gathered evidence and a multi-tier challenge process.
The Room
tools-wg21/room.md
Given any institutional discussion, produces a structural-dynamics diagnosis through six analytical lenses and an eight-tier challenger.
Scribe
tools-wg21/scribe.md
Transforms a raw transcript into two-layer minutes - executive summary for the chair, full attributed record for the implementer.
Slider
tools-wg21/slider.md
Conversational tool that designs a slide deck with you and renders it to PowerPoint from Markdown. The rendering engine is a Python package in tools-wg21/slider/, run with uv run slider.
Summarize Papers
tools-wg21/summarize-papers.md
Takes a source folder of papers and delivers a campaign briefing plus a public reading-guide paper with per-paper and aggregate summaries.
The Table Read
tools-wg21/table-read.md
Given an author's communication, produces three-audience reaction predictions with blind spots, polarization assessment, and a self-check on cost.
The Threadalyzer
tools-wg21/threadalyzer.md
Writes a trip report from proceedings by cataloging each delegate's rhetorical patterns, the technical-to-political ratio, and a counterfactual verdict.
Tighten
tools-wg21/tighten.md
Takes a bloated document and delivers the same document at half the line count with voice preserved.
The Vasa
tools-wg21/vasa.md
Given cross-room committee material, produces a coherence report measuring which of Stroustrup's 24 structural principles are under stress.
Articles for humans on tool building and AI craft.
The Fan-Out Problem: Why AI Is a Critic, Not an Author
lessons/fan-out-problem-ai-as-critic.md
AI is good at judging. AI is bad at creating.
Semantic Blur: Why Rewriting a Prompt File Degrades It
lessons/semantic-blur-effect.md
Three prompt files rewritten by a language model all grew with no new capability - 18.7%, 43.6%, and 145.8% in characters. This article explains why.
Samples of generated reports. See output/README.md for the full listing.
Supporting material that is not itself a tool.
squeeze-pngs
crates/png2jpg/squeeze-pngs.md
Rust utility that converts PNG images to JPEG at quality 50, keeps the originals, and rewrites .png references to .jpg in sibling markdown files.
Dokuman design conversation
chats/chat-dokuman.md
The chat transcript in which Dokuman was specified, kept as a record of how the tool was designed.
Art
art/images/
Illustrations that do not pair with a specific tool.
AGENTS.md
AGENTS.md
Structural rules for agents working in this repository - README tracking, tool locations, the retired convention, and the image invariant.