Internal maintainer documentation. Public and contributor docs live in
docs/gitbooks/en/ (navigation in docs/gitbooks/en/SUMMARY.md); agent and contributor rules
live in AGENTS.md (CLAUDE.md is a symlink to it); per-domain design notes
sit beside the code as crates/openhuman-core/src/<domain>/README.md.
Every file in this directory is listed below. Links here are checked in CI by
the markdown-link-check job in .github/workflows/pr-quality.yml.
CONTRIBUTING-BEGINNERS.md— step-by-step first-PR guide for newcomers; linked from the rootREADME.mdandCONTRIBUTING.md.SUPPORT.md— where to ask for help; routes each kind of problem to a GitHub Discussions category.community/discussions.md— how maintainers organize and triage those Discussions categories.
The first two are referenced from .github/PULL_REQUEST_TEMPLATE.md.
TEST-COVERAGE-MATRIX.md— feature-ID to test-path matrix. Feature IDs are validated againstscripts/feature-ids.jsonbyscripts/check-coverage-matrix.mjsin thepr-qualityworkflow; test paths are not validated, so keep them current by hand. Update rows in the same PR that changes a feature.RELEASE-MANUAL-SMOKE.md— manual smoke checklist run on every release cut; the only accepted substitute for a 🚫 matrix row.prompt-evals.md— the agent-prompt comprehension suite: the case list, how to run it, and how to read a regression. Coupled both ways toscripts/prompt-eval.sh,scripts/prompt-eval/cases.jsonandtests/in_process/agent_prompt_comprehension_e2e.rs.JAPANESE-UI-VALIDATION.md— the ja-JP UI pass: what to look at, and the layout failures Japanese text exposes that English does not.
ci-self-hosted.md— the only explainer for the lane architecture: howci-fast.ymlandci-fast-hosted.ymlcallci-lanes.yml, what each lane covers, which checks land in which lane, and how the Hetzner EX63 microVM runners are routed. Read this before adding a check to CI.
Resource-footprint work for the embeddable core. The profiling binaries
(library-profile, rss-bench), their driver scripts and the benchmark
method (profile/docs/library-benchmarking.md) live in
tinyhumansai/openhuman-benchmarks, with the agent-scale, life-scenario
and onboarding stress suites. The core keeps only the rss-bench feature they
build against.
library-minimal-recipe.md— the--no-default-features --features "skills,flows"recipe for a headless library build, with measurements.dep-audit/<date>.md— archived dependency-audit snapshots. These are machine-generated:pnpm dep:auditwrites a fresh report totarget/dep-audit/REPORT.md, andpnpm dep:audit --snapshotarchives it here. Readscripts/dep-audit/README.mdbefore acting on one, and date any claim you take from a snapshot.
Domain-local design notes live beside their code. Two exceptions sit here because they coordinate several repositories at once and cannot be owned accurately by one source directory.
specs/memory-v2.md— the Memory v2 contract: the model, scope, the per-turn lifecycle, background jobs, config, the agent tool, the RPC table and the UI. The most referenced document in the tree: the Rust core, the embed facade, the frontend, a Playwright spec, a root integration test and five gitbooks pages all cite it. Treat it as current architecture, not a plan.specs/agent-runtime-upstream-boundary.md— the normative ownership boundary for making the core a host of the TinyAgents harness rather than a second agent runtime, with deletion criteria.plans/— the agent-runtime migration work plans, plus one dated measurement.plans/README.mdcarries the per-plan landed/outstanding status, verified against the tree, and the commands to re-check it. Read that before trusting any individual plan's own status line.
Translations of the root README.md, linked from its language bar:
README.ar.md, README.de.md, README.ja-JP.md, README.ko.md,
README.tr.md, README.ur-pk.md, README.zh-CN.md.
Nothing compares a translation's structure against the root README, so they
drift silently; pnpm i18n:check covers UI strings only. Check the root
README's section list by hand when you touch one.