byte-codebase-harnesslisted

# Byte Codebase Harness Codebase Harness makes a repository navigable before Byte OS plans or edits it. It adapts Anthropic's large-codebase guidance for Claude Code into a provider-neutral setup that also works for Codex. ## Inputs Inspect: ```text repo root top-level directories existing CLAUDE.md existing AGENTS.md existing .claude/settings.json package/build/test files .gitignore and generated/artifact directories .byte-os/STATUS.md if present ``` ## Workflow 1. Map the codebase. - List top-level directories and give each a one-line purpose. - Identify likely generated files, build outputs, vendored code, dependency directories, fixtures, and large artifacts. - Identify main language stacks and likely LSPs. 2. Create dual-agent context files. - Root `CLAUDE.md`: concise Claude Code entry context. - Root `AGENTS.md`: concise Codex entry context. - Keep root files lean: project map, critical commands, dangerous areas, conventions, and pointers. - Do not put specialized workflows in root context; use skills or module files instead. - Treat `AGENTS.md` as Codex's live codebase harness, not a generic README. It should help Codex know where to start, what to avoid, which scoped commands to run, and which Byte OS artifacts contain deeper context. - Maintain parity between `CLAUDE.md` and `AGENTS.md` where concepts overlap, while adapting syntax and behavior to each tool. 3. Add module-local context where useful. - For the 2-3 most relevant o