neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
📋 The Spec-Driven Development framework. Define the what, AI handles the how.
Translation pending
Developer toolsJavaScript · TypeScriptGitHub @brabos-aiSelf-installOwner unconfirmed
Paste into Claude or ChatGPT for installation instructions suited to your environment.
Have you tried it? One tap is enough. No sign-in needed.
Loading the full README. The saved excerpt is shown below.
AI-powered development workflows that make you addicted to shipping code. Production-ready commands, scripts, and skills for AI coding assistants, with a single CLI installer.
Official Framework Site: code.brabos.ai — your complete development workflow guide and documentation.
Most AI coding setups are fragmented across custom prompts, scripts, and editor-specific conventions.Read the full README ↗AI-powered development workflows that make you addicted to shipping code. Production-ready commands, scripts, and skills for AI coding assistants, with a single CLI installer.
Official Framework Site: code.brabos.ai — your complete development workflow guide and documentation.
Most AI coding setups are fragmented across custom prompts, scripts, and editor-specific conventions.
Code Addiction standardizes this with:
.codeadd/ (commands, scripts, skills, templates)codeadd) with install, update, modify, providers, uninstall, doctor, and validate# install latest release
npx codeadd install
# install a specific version (e.g., v0.2.20)
npx codeadd install --version v0.2.20
# install from a release channel
npx codeadd install --channel beta
# install at user level (home directory)
npx codeadd install --global
# running install over an existing installation offers Modify / Update / Reinstall / Cancel
# instead of resetting it
# check environment health
npx codeadd doctor
# validate file integrity via SHA-256 hashes
npx codeadd validate
# repair missing or corrupted files
npx codeadd validate --repair
# update to latest release
npx codeadd update
# update to a specific version
npx codeadd update --version v0.2.14
# switch release channels while updating
npx codeadd update --channel stable
# edit providers, features and plugins in one interactive screen
npx codeadd modify
# list, add or remove providers (--global / --user apply here and to modify)
npx codeadd providers list
npx codeadd providers add cursor
npx codeadd providers remove cursor --force
# remove Code Addiction files from your project
npx codeadd uninstall
# list optional features
npx codeadd features list
# list available plugins (see docs for per-plugin setup)
npx codeadd plugins list
# serve the knowledge graph over MCP stdio (docs | artefacts)
npx codeadd mcp --corpus=docs
# show installation configuration
npx codeadd config show
Plugins integrate external MCP tools into ADD commands — extending discovery, planning, and diagnosis with structural code navigation, and adding live browser driving to QA validation (which works out of the box without any plugin). Plugins are disabled by default — enable only after installing the external tool.
See code.brabos.ai/docs#plugins for the full plugin catalog with install prerequisites and step-by-step setup.
The board feature keeps a project's backlog on its own board branch, so a ticket write never touches your code branches. It is disabled by default.
npx codeadd features enable board
New project. Enabling board (with codeadd features enable board, codeadd modify --enable-feature board, codeadd install --enable-feature board, or by accepting it in /add-setup) creates the board: it writes .codeadd/board.json, pushes an orphan board branch to origin and makes the clone under ~/.codeadd/. The project needs an origin with its main branch pushed. Nothing on your code branches is committed or pushed. Afterwards, commit .codeadd/board.json, which is already staged. If the board could not be created, the enable still succeeds with one warning; fix what it names and run node .codeadd/scripts/board-init.cjs.
Migrate a project that still keeps docs/backlog.jsonl in the code. Needs codeadd 1.5.0 or later installed in the project (npx codeadd update). Download scripts/migrations/migrate-board.cjs from this repository and run it at the project root, on any branch:
node migrate-board.cjs
It uses the scripts the project already has in .codeadd/scripts/ (MODULES= says which folder), creates the board branch, and stages the code side without committing: commit those staged changes yourself. Until that commit is merged, write nothing to the old docs/backlog.jsonl; run the script again right before the merge (--reimport applies what was written meanwhile).
ERROR=board-modules-missing: the project has no .codeadd/scripts/ or its copy is older than 1.5.0. Run npx codeadd update and try again.IGNORE_WARNING=board-json-ignored: a rule the script does not recognise (a nested .gitignore, .git/info/exclude) still ignores .codeadd/board.json. IGNORE_RULE= names it. Un-ignore the file.Release step (optional). With it on, a merged ticket waits in awaiting-release instead of going straight to done.
node .codeadd/scripts/backlog-commit.cjs release-flow on # turn it on (off turns it back off)
node .codeadd/scripts/backlog-commit.cjs release v1.2.0 # at release time: close every waiting ticket with the version
Open the board. The board screen is not distributed yet (it is the next piece of work on the backlog). Today you run it from a clone of this repository:
git clone https://github.com/brabos-ai/code-addiction && cd code-addiction
npm run setup
node board/server.mjs --root /path/to/your/project
It prints BOARD_URL= and opens the browser (--no-open skips that). It only reads.
Code Addiction turns complex development into a guided, repeatable flow. Instead of figuring out how to build, you just follow the next command. The AI does the heavy lifting — you stay in control.
Every feature follows a clear path from idea to delivery. Pick the trail that fits your task:
Step Command What happens Output
───────────────────────────────────────────────────────────────────────────────
0. Explore /add-brainstorm Brainstorm ideas (read-only) Initial concept
1. Discover /add-new AI-guided feature discovery about.md
2. Plan /add-plan Technical planning + UX contract plan.md + design.md
3. Code /add-build Subagent-driven implementation Working code
4. Review /add-review Code review + spec audit (+QA) review-NNN.md
5. Done /add-done QA evidence, changelog, docs, merge Merged branch
Steps 1-4 can also run unattended: approve /add-brainstorm with "deliver automatically" and each stage hands off to the next until /add-build runs its own final review and asks whether to open the PR. /add-done always waits for you.
Deploy is out of scope of a delivery: a delivery is code, migration files and tests in one merge, and deploying runs after /add-done, through your own deploy process.
Not sure which command to run next? /add-help answers which command to use and in what order, what each feature and plugin is and how to enable it, how to run from a bot, and the CLI lifecycle. It is a router and finishes no work. New here, or just ran codeadd install or codeadd update? Run /add-setup: it checks your project, offers each feature and plugin with its benefit in one list, and sets up the ones you accept.
Pick the shortest path that fits. Less ceremony, same quality.
COMPLETE (complex features with UI)
brainstorm --> new --> plan --> build --> review --> done
(design is produced inside /add-plan STEP add-plan.ux-design)
STANDARD (features without complex UI)
new --> plan --> build --> review --> done
LEAN (small changes, quick tasks)
new --> build --> done
AUTOMATIC (one approval, no further interaction)
brainstorm --> new --> plan --> build --> publish question
(chosen at brainstorm's approval; build runs its own final review, then asks to publish)
EXPLORATION (don't know where to start?)
brainstorm --> new --> ...pick your flow above
EMERGENCY (critical bug in production)
hotfix --> done
TRIAGE (ambiguous symptoms, unclear path)
diagnose --> hotfix @report --> done
(diagnose can also route to new, or stop at no-action)
ANALYSIS (understand existing codebase)
wiki / audit
That's it. No config files to tweak, no boilerplate to write, no decision fatigue. Type the command, follow the AI, ship the feature. Repeat.
.codeadd/.claude/.codex/.agent/.cursor/.opencode/.codex/ (commands/skills, shared with Codex; agents go to .zcode/)Commands and skills install to every provider you select. Subagents install to Claude Code, Cursor, OpenCode, Codex and ZCode (Antigravity is not yet supported for agents).
AGENTS.md: project instructions read by AI coding assistants (replaces CLAUDE.md at the repo root)cli/: installer CLI published as codeaddmcp/: the knowledge-graph MCP server (two corpora, selected by --corpus), shipped in the npm packageboard/: the read-only backlog board over the project's board branch (read through one clone per project) — a React/TypeScript app plus a zero-dependency Node server; built by npm run build:board, run by npm run board. Not carried by the npm package or the main release ZIPframwork/: framework payload copied into target projects by the installer
framwork/.codeadd/plugins/: plugin asset source tree (fragments and skills per plugin)workbench/: internal-layer source for the framework's own commands, skills and agents, with its own build pipeline (compiles to .claude/, .opencode/, .agents/ and .codex/ at the repo root, gitignored)docs/deliveries/: durable delivery history — closed-out plan archivesdocs/delivered.jsonl: the delivery index every close-out appends to.codeadd/board.json: the one tracked file under .codeadd/; it says where the board branch (the backlog ticket ledger the board reads) livesRequires Node.js >=22.19.0; Node 24 LTS is recommended. The suites run on Node's own runtime — no Bash, no Docker.
npm ci # root tooling (pins the lock)
npm --prefix cli ci # CLI dependencies (vitest)
node scripts/build.js # compile product provider files + sidecars
node scripts/build-workbench.js # compile this repo's own pipeline
npm test # CLI suite (vitest) through scripts/run-tests.js
npm run test:scripts # root native script suite (node --test)
npm run test:all # both, vitest first
npm --prefix cli run test:package # pack + install + execute the CLI bin
npm run release # maintainers: tag, GitHub release, CLI publish
npm test and npm run test:scripts both go through scripts/run-tests.js,
work on a copy of the checkout rather than the tree you are standing in, and run
natively on Windows, macOS and Linux.
Contributions are welcome. Start here: