modelrouter
AGENTS.md1 starsChanged 4 months agoMITView on GitHub
ModelRouter is a zero-dependency Claude Code plugin. Runtime logic lives in scripts/ as single-file CommonJS programs. hooks/hooks.json connects those scripts to Claude Code events. Agent definitions are in agents/, while each command or routing capability has its own skills/<name>/SKILL.md. Default user state is copied from defaults/. Keep architectural documentation in docs/ and demo instructions in DEMO.md. Development fixtures live under .route/fixtures/: .router/ is per-repo runtime state (log.jsonl, gitignored), while .route/ is the gitignored dev harness (plans, fixtures, checkpoints). Plugin metadata…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,843 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
Run the relevant script directly after changes, then exercise the matching checkpoint documented in CONTRIBUTING.md using the .route/*.sh …
- Very specific: line 33
Use two-space indentation, semicolons, double-quoted strings, and CommonJS require() with explicit node: prefixes.
- Very specific: line 50
Follow the existing Conventional Commit style: feat:, fix:, or chore:, with a concise imperative summary.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Repository Guidelines
Project Structure & Module Organization
ModelRouter is a zero-dependency Claude Code plugin. Runtime logic lives in
scripts/ as single-file CommonJS programs. hooks/hooks.json connects those
scripts to Claude Code events. Agent definitions are in agents/, while each
command or routing capability has its own skills/<name>/SKILL.md. Default
user state is copied from defaults/. Keep architectural documentation in
docs/ and demo instructions in DEMO.md. Development fixtures live under
.route/fixtures/: .router/ is per-repo runtime state (log.jsonl,
gitignored), while .route/ is the gitignored dev harness (plans, fixtures,
checkpoints). Plugin metadata belongs in .claude-plugin/plugin.json.
Build, Test, and Development Commands
There is no build step and no npm install; scripts use only Node.js built-ins.
claude plugin validate .validates the plugin manifest and layout.for f in scripts/*.js; do node --check "$f" || exit; donechecks JavaScript syntax across every hook script.echo '{"prompt":"fix typo in README","hook_event_name":"UserPromptSubmit","cwd":"."}' | node scripts/inject.jssmoke-tests prompt routing from stdin.node scripts/stats.jsrenders statistics from.router/log.jsonl; setROUTER_NOWwhen a deterministic date is needed.
Run the relevant script directly after changes, then exercise the matching
checkpoint documented in CONTRIBUTING.md using the .route/*.sh scripts with
RUN_LIVE=1.
Coding Style & Naming Conventions
Use two-space indentation, semicolons, double-quoted strings, and CommonJS
require() with explicit node: prefixes. Prefer small functions and Node's
standard library over dependencies. Script filenames use kebab-case
(memory-card.js); skill directories use lowercase command names. Preserve
the hook contract: read JSON from stdin and emit only valid JSON or intentional
user-facing text. Keep Markdown direct and wrap prose near 80 columns.
Testing Guidelines
The repository uses focused smoke checks instead of a test framework or formal
coverage target. Test valid input, malformed or missing local state, and the
empty-data path for any changed hook. Use temporary HOME and fixture data
when a script touches ~/.router so development does not overwrite personal
rules. Add fixtures only when output must remain stable across runs.
Commit & Pull Request Guidelines
Follow the existing Conventional Commit style: feat:, fix:, or chore:,
with a concise imperative summary; append a plan ID such as (T4.2) when
applicable. Pull requests should explain behavior changes, list commands run,
link the issue or plan task, and include screenshots for visible CLI or Claude
Code output. Never commit API keys, personal ~/.router data, or generated
.router/ logs and cards.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.