processfork
CLAUDE.md1 starsChanged 6 months agoMITView on GitHub
This is the build-agent contract. Keep it short. Subsystem detail lives in agent_docs/ and is loaded on demand. - Rust edition 2024, MSRV 1.91 (1.85 → 1.88 was the home crate bump in v1.0.0; 1.88 → 1.91 was aws-sdk-s3 in v1.0.2 — both recorded in claude-progress.json/assumptions[]). - Python 3.11+, formatted with ruff and type-checked with mypy --strict. - TypeScript with biome for lint + format. - Run cargo fmt --all and cargo clippy --all-targets -- -D warnings before every commit.…
In this repository5 other files it gives its agents.
- kvcache-formatSkill
- napi-bindingSkill
- pyo3-bindingSkill
- rust-patternSkill
- test-harnessSkill
The file
We split this CLAUDE.md into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 13
Run cargo fmt --all and cargo clippy --all-targets -- -D warnings before every commit.
- Very specific: line 20
Every public symbol gets a doc-comment.
- Very specific: line 30
Before claiming a phase complete, invoke the gate sub-agents in this order: qa-reviewer → security-reviewer → perf-reviewer.
- Very specific: line 32
Never edit claude-progress.json to advance current_phase until the gate passes.
- Distinctive: line 48
Always leave the repo in a state where a fresh window can resume from the three state files above + git status.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.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
ProcessFork — agent rules
This is the build-agent contract. Keep it short. Subsystem detail lives in
agent_docs/ and is loaded on demand.
Toolchain
- Rust edition 2024, MSRV 1.91 (1.85 → 1.88 was the
homecrate bump in v1.0.0; 1.88 → 1.91 was aws-sdk-s3 in v1.0.2 — both recorded inclaude-progress.json/assumptions[]). - Python 3.11+, formatted with
ruffand type-checked withmypy --strict. - TypeScript with
biomefor lint + format. - Run
cargo fmt --allandcargo clippy --all-targets -- -D warningsbefore every commit. The.claude/hooks/post-edit.shhook does this automatically for files you've just touched.
Style bar
- No
unwrap()/expect()outside#[cfg(test)]andexamples/. - Every public symbol gets a doc-comment. Doc examples must compile
(
cargo test --doc). - Errors are typed (
thiserrorin libs,anyhowonly in binaries). tracingfor logs — neverprintln!outside CLI user-facing output.- SPDX header (
// SPDX-License-Identifier: MIT) on every source file.
Verification discipline
- A feature is not done until an end-to-end example in
examples/runs green against real binaries. Unit tests alone are insufficient. - Before claiming a phase complete, invoke the gate sub-agents in this order:
qa-reviewer→security-reviewer→perf-reviewer. All three must PASS. - Never edit
claude-progress.jsonto advancecurrent_phaseuntil the gate passes.
State files (read these first every new session)
claude-progress.json— machine-readable phase state, blockers, assumptions. Update on every phase boundary.claude-plan.md— your scratchpad and handoff to your future self. Keep < 200 lines.agent_docs/<current-phase>.md— full spec for what you're building.
Sessions and context budget
- This is a multi-session build. You may not see the kickoff prompt again.
- At 60% context, write a progress note. At 70%, commit and consider compact. At 85%, finish the current logical unit and stop adding new work.
- Always leave the repo in a state where a fresh window can resume from the
three state files above +
git status.
Out of scope (do not build, even if it seems natural)
Hosted SaaS, web dashboard, Windows native, distributed multi-host fork, custom inference engine, custom model-merge algorithm, telemetry of any kind. See the §2 scope fences in the original kickoff prompt.
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.

