Weight Space Programming
Inspect, explain, repair, or construct neural-network weights directly through reasoning and tensor edits without training. Use for weight-space interpretation, targeted checkpoint surgery, reasoning-loop diagnosis, and hand-written neural policies or algorithmic transformer circuits. Do not use for ordinary fine-tuning, inference optimization, or prompt tuning.
Install
npx skills add xero00000/weight-space-programming --skill weight-space-programminggit clone --depth 1 https://github.com/xero00000/weight-space-programming.git /tmp/weight-space-programming
mkdir -p ~/.claude/skills
cp -r /tmp/weight-space-programming/. ~/.claude/skills/weight-space-programmingPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Weight-Space Programming Translate a behavioral specification into an explanation or a concrete tensor intervention. Use the methods in [Takehi et al. (2026)](https://rikiyat.github.io/blog/weight-space-programming/), while treating interpretations as hypotheses and improvements as measurements. A skill supplies a method; it does not confer the capabilities reported for the paper's editor. ## Establish the experiment Identify the checkpoint revision and format, architecture and tensor orientation, input/output semantics, observed failure or desired behavior, and available compute. Check model code and configuration before reasoning about tensor names. Preserve the original checkpoint and create candidates separately. Choose the evidence regime from the user's request: - **Weights only:** inspect tensors and algebraic compositions; do not execute the network, individual layers on inputs, or synthetic probes. Separate weight-space evidence from behavioral speculation. If only a subset is visible, describe the subset. - **Bounded probes:** use a declared budget for generations, activations, or environment episodes. Count examples, batch sizes, tokens and calls, since a paper's “forward pass” may contain hundreds of sequences. Keep the same decoding and environment settings for paired comparisons.
Details
- Skill name
weight-space-programming- Repository
- xero00000/weight-space-programming
- License
- MIT
- Repo updated
- 2 November 2026
FAQ
What does the Weight Space Programming skill do?
Inspect, explain, repair, or construct neural-network weights directly through reasoning and tensor edits without training. Use for weight-space interpretation, targeted checkpoint surgery, reasoning-loop diagnosis, and hand-written neural policies or algorithmic transformer circuits. Do not use for ordinary fine-tuning, inference optimization, or prompt tuning.
How do I install the Weight Space Programming skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add xero00000/weight-space-programming --skill weight-space-programming. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Weight Space Programming popular?
It lives in xero00000/weight-space-programming, which has 9 GitHub stars and 2 forks.
Related skills
4d V21
9Ganbin/4d-development-skill
Comprehensive 4D v21 development expert with embedded official documentation. Covers ORDA patterns, entity classes, data model classes, queries, classic methods, data types, error handling, forms, events, web server, and REST API. Use when working with 4D files (.4dm), 4D language questions, 4D project structure, entity classes, ORDA queries, database operations, form development, web server configuration, REST API usage, or any 4D-specific syntax issue. Includes full official 4D v21 documentation for on-demand retrieval. Triggers on: .4dm files, 4D code, ORDA, entity selection, dataclass, 4D
User Stories
9felipereisdev/user-stories-skill
Analyzes a codebase and breaks down a goal into implementable user stories with clear acceptance criteria. Use when planning a new feature, estimating work, or creating a development roadmap. Explores the code before asking anything; when the goal is ambiguous, asks at most 5 blocking multiple-choice questions with a recommended answer, and records assumptions and open questions in the plan. Outputs execution-ready stories (schema v2: type, status, a verify check per criterion, dependencies with reasons, interfaces, files to reuse, file conflicts for parallel work) to .context/[goal-slug]-plan
Comms How
23jpoehnelt/skills
Explain how a subsystem works, its architecture, runtime flow, and where new code belongs. Use for "how does X work", walkthroughs before changing something, and placement or ownership questions like "where should this live" or "which package owns this". Use comms-why for motivation.
Sega Genesis Sgdk
9haroldo-ok/sega-genesis-sgdk-skill-for-claude
Create, debug, optimize, and port games for the Sega Genesis / Mega Drive in C using SGDK (the Sega Genesis Development Kit). Use whenever the user wants to make a Genesis / Mega Drive game, port an existing game (from DOS, C, a Palm/6502/Java codebase, or any platform) to the Genesis, build or debug an SGDK project, fix a ROM that boots to a black screen or crashes, wire up graphics/sprites/tilemaps/scrolling, convert assets (tiles, sprite sheets, VGM/XGM music, S3M/MOD tracks, WAV samples) with rescomp, set up XGM/XGM2 sound, or add host + headless-emulator tests. Trigger even when the user
Effect
23jpoehnelt/skills
Apply Effect v4 conventions for services, layers, tagged errors, boundary runners, resources, retries, deadlines, HTTP, and tests. Use when writing or reviewing TypeScript that imports effect, adds a service or error class, chooses where to run a program, bounds external I/O, or tests Effect code.
Programasweights
9programasweights/skills
Compile a natural-language spec into a tiny neural function that runs locally with ProgramAsWeights (PAW). Use it for fuzzy text-in / text-out tasks that a regex can't handle but that are too slow, costly, or overkill to send to a full LLM on every item - classify, categorize, label, or tag text (sentiment, urgency, intent, topic, spam, support tickets, ALERT vs QUIET log lines); extract fields from messy text (emails, names, dates, IDs, invoice numbers); repair or normalize formats (broken JSON, dates); fuzzy or typo-tolerant matching, near-duplicate detection, and deduplication; map a misspe
Browse the most popular Claude skills, all software development skills, or write your own with the skills generator.