Skip to content

Help wanted

Peerborne already contains substantial CRDT, storage, networking, encryption, ACL, key-management, framework, and indexing primitives. The highest-priority work is proving and completing the end-to-end paths that compose them. See the feature and verification audit for current evidence and the contributing guide before starting.

The issue tracker may not have curated beginner tasks. Use Discussions to scope an idea and Issues for reproducible bugs or agreed actionable work. Security-sensitive findings must go through private vulnerability reporting from the repository Security tab.

The initial founder-plus-one flow has a distinct-identity, cross-NAT acceptance test. What remains is persisted KEM/BeeKEM and replay state, offline/delayed acceptance, larger-group add-side updates, and multi-peer proof that a revoked member cannot read or write subsequent content.

Useful work includes deterministic acceptance tests, durable state design for the current format, adversarial cases, and safe UX that never logs keys or private payloads.

Content-addressed blocks and IndexedDB-backed components exist, but document and identity recovery across browser/process restart needs executable coverage. Test key persistence separately from document blocks, reject persisted state in an unsupported format instead of upgrading it, and verify explicit failure behavior when required state is absent.

CRDT adapters and isolated sync components are tested, and the dedicated cross-NAT invitation job asserts live post-join mutation in both directions. Add deterministic partition, concurrent edit, rejoin, and exact convergence assertions without representing transport-only messaging as database convergence.

Storage and pinning concepts are documented, but a complete publisher, durable remote retention flow, and tested restore path are not established. Contributions should specify trust, authorization, retention, encryption, failure, and recovery boundaries.

The relay has an app-managed persistent identity, readiness checks, bounded resource controls, unit coverage, and NAT acceptance paths. Needed work includes hosted restart and WSS deployment smoke tests, identity backup/rotation, in-flight failover, multiple-relay selection, churn, observability, and scale evidence.

The six @peerborne/* workspaces build, but they are unpublished. Prepare publication only with clean tarball inspection, external-consumer ESM and declaration tests on Node 22.19.0 and browsers, dependency/export verification, and reviewed release automation.

The Site workflow generates TypeDoc Markdown from source during yarn workspace @peerborne/site build; generated files are ignored. Improve source API comments or site/astro.config.mjs, not generated Markdown. Add executable snippet/link checks so quick-start and cookbook commands cannot silently drift. Starlight TypeDoc is the only API reference workflow.

Core and index benchmark scenarios now execute reproducibly as ESM. Establish representative datasets, environment reporting, baselines, variance handling, and regression budgets before publishing performance claims.

The signed manifest, expiring advertisement, direct-query, transport-adapter, and candidate-verification layers have isolated coverage. Build collection search membership and dedicated key distribution, register bounded libp2p handlers, implement the authorized document resolver, publish automatically rotated blind-token snapshots, and verify hostile peers, restart, revocation, partition/rejoin, and schema/key evolution. No current example demonstrates this end to end.

browser-test, wiki-swarm, password-manager, and peerborne-note build and pass Chromium startup smoke tests through yarn test:e2e. Peerborne Note is the bounded initial-release demo; it is not proof of durable invitations, persistence, reconnect, revocation, relay availability, pinning, or distributed search. Contributions should state precisely which boundary a new test crosses and avoid production-readiness claims.

Maintainer responses and reviews have no SLA. Small, focused proposals with a reproducible failing case or a clear acceptance criterion are easiest to evaluate.