The repository has no AGENTS.md and no shared Claude Code configuration. Every agent session re-derives the same facts from README.md, the justfile, and the CI workflow, and the house rules (no em-dashes, Conventional Commits, honest AI-assistance disclosure) live only in reviewers' heads.
Add a small, shared, checked-in set of tooling:
.claude/settings.json plus three hooks under .claude/hooks/: rustfmt on each edited .rs file (edition 2024, matching the workspace), an em-dash content lint on .rs and .md edits, and cargo nextest run for the crates with uncommitted .rs changes at the end of a turn. Each hook is a no-op when its tool is absent, so nothing changes outside the Nix dev shell.AGENTS.md as the agent contract, with CLAUDE.md as a symlink to it. Written in ASD-STE100 Simplified Technical English, one sentence per line, and grounded only in what this repository already states: the L1 venue-agnostic runtime, the extension seam in nexum-world, the crate layout, the pinned Rust 1.94.0 toolchain, and the real justfile recipes.ripgrep and ast-grep in the flake devShell. cargo-nextest is already present..config/nextest.toml with a slow-timeout, so a hung test fails by name after five minutes instead of consuming the whole CI job timeout.- A
.gitignore change so .claude/settings.json and .claude/hooks/ are tracked while settings.local.json and worktrees/ stay out.
Out of scope: the justfile and .github/workflows/ci.yml are unchanged. Both already run cargo nextest run and a separate cargo test --doc, which is correct.
AI Assistance: Claude Opus used for drafting this issue and the accompanying change.
The repository has no AGENTS.md and no shared Claude Code configuration. Every agent session re-derives the same facts from README.md, the justfile, and the CI workflow, and the house rules (no em-dashes, Conventional Commits, honest AI-assistance disclosure) live only in reviewers' heads.
Add a small, shared, checked-in set of tooling:
.claude/settings.jsonplus three hooks under.claude/hooks/: rustfmt on each edited.rsfile (edition 2024, matching the workspace), an em-dash content lint on.rsand.mdedits, andcargo nextest runfor the crates with uncommitted.rschanges at the end of a turn. Each hook is a no-op when its tool is absent, so nothing changes outside the Nix dev shell.AGENTS.mdas the agent contract, withCLAUDE.mdas a symlink to it. Written in ASD-STE100 Simplified Technical English, one sentence per line, and grounded only in what this repository already states: the L1 venue-agnostic runtime, the extension seam innexum-world, the crate layout, the pinned Rust 1.94.0 toolchain, and the real justfile recipes.ripgrepandast-grepin the flake devShell.cargo-nextestis already present..config/nextest.tomlwith a slow-timeout, so a hung test fails by name after five minutes instead of consuming the whole CI job timeout..gitignorechange so.claude/settings.jsonand.claude/hooks/are tracked whilesettings.local.jsonandworktrees/stay out.Out of scope: the justfile and
.github/workflows/ci.ymlare unchanged. Both already runcargo nextest runand a separatecargo test --doc, which is correct.AI Assistance: Claude Opus used for drafting this issue and the accompanying change.