The universal link for code. Select the exact code once. One canonical Share becomes a readable review for a teammate or bounded Markdown, JSON, or archive context for an AI agent.
Create in the browser · Open a real example · Read the privacy boundary
Neurcode Share is the open format, CLI, SDK, and viewer for creating a bounded, reviewable code handoff. The hosted service at share.neurcode.com publishes that handoff as a link when you choose to.
From any Git repository:
npx @neurcode-ai/share@0.3.0The Composer binds only to loopback, inventories the exact disclosure, scans for sensitive material, and creates nothing remotely unless you explicitly choose Publish.
For a non-interactive local archive:
npx @neurcode-ai/share@0.3.0 src/auth.ts:20-80 --diff --yes --out review.tar.gz
Select the exact material, review the disclosure cut locally, then publish one
immutable revision and one URL. A person gets inert readable HTML; an AI agent
gets the equivalent bounded Markdown, JSON, or verified archive without the
creator manually repackaging anything. The example uses the released 0.3.0
CLI and the same canonical Share as this
public focused-review Share.
| Workflow | What the recipient gets | Open |
|---|---|---|
| Focused code review | Implementation, complete diff, relevant test, passing evidence, and one review question | View Share |
| Debugging handoff | Failing path, bounded reproducer, observed failure, controls, and an asserted hypothesis | View Share |
| AI-agent handoff | Cited task scope, current work, contract context, evidence, remaining work, and trust guidance | View Share |
See public Share examples for the included and excluded scope plus direct Markdown and JSON delivery.
Local creation needs no account. Export a deterministic archive, inert HTML, Markdown, or agent JSON:
npx @neurcode-ai/share@0.3.0 src/queue.ts --yes --out queue-review.html
npx @neurcode-ai/share@0.3.0 --staged --yes --out staged-review.md
npx @neurcode-ai/share@0.3.0 --diff=main..HEAD --yes --out change.json
npx @neurcode-ai/share@0.3.0 --run "pnpm test queue" --yes --out evidence.tar.gzCommands are bounded by time and output limits. Sensitive paths are denied by default. The disclosure airlock runs before any file is written or uploaded.
Verify exact cited bytes against a selected local repository state, or prepare a reviewed immutable successor without publishing:
npx @neurcode-ai/share@0.3.0 verify review.tar.gz --repo ../project
npx @neurcode-ai/share@0.3.0 refresh review.tar.gz --decision i2=use --output refreshed.tar.gzSee the CLI guide for deterministic JSON, named revisions, staged comparisons, explicit refresh decisions, hosted checks, and authorized comments.
Add --publish after reviewing the local cut. Publishing uses the existing
Neurcode identity and a loopback PKCE flow, so no hosted token is copied into
the terminal:
npx @neurcode-ai/share@0.3.0 src/session.ts --publish --visibility unlistedThe hosted service at share.neurcode.com is
proprietary and is not implemented in this repository. Its documented client
contract is available through @neurcode-ai/share-sdk.
For a person, send the viewer link. They see the question, selected context, provenance, and comments subject to the Share's access policy.
For an AI agent, use the same accessible URL with content negotiation, an adjacent canonical representation, or create a short-lived, revision-pinned agent link in the hosted library and fetch its scoped format:
npx @neurcode-ai/share@0.3.0 fetch 'AGENT_LINK' --stdout mdCapability and agent secrets remain in URL fragments at rest and are sent in request headers, not request paths.
A Share archive contains canonical cut.json, narrative story.json,
provenance pins.json, and content-addressed blobs. The digest excludes
wall-clock capture time but includes the selected content, intent, evidence,
and provenance. Identical meaningful inputs therefore produce identical
identity.
The source-free example demonstrates the metadata a recipient can inspect without publishing real source.
- Capture, selection, scanning, review, and export happen locally.
- No account is required until hosted publishing.
- Paths, archive sizes, expanded bytes, item counts, commands, and timeouts are bounded.
- Archives reject traversal, links, malformed records, digest mismatches, and expansion bombs.
- Renderers treat all captured material as inert text.
- Secret scanning reduces risk but cannot prove that content is safe; the human disclosure review remains authoritative.
See privacy, the threat model, and format compatibility.
| Tool | Exact provenance | Local disclosure review | Bounded evidence | Agent formats | Hosted collaboration |
|---|---|---|---|---|---|
| Chat paste | rarely | no | no | prose | chat-specific |
| Gist | commit-level | no | no | source | link and comments |
| Pull request | strong | diff review | CI-dependent | diff/API | repository workflow |
| Neurcode Share | content pins and digest | yes | yes | Markdown, JSON, archive | optional |
Share complements pull requests when the review context is smaller, unfinished, cross-file, or intended for an external person or agent.
@neurcode-ai/share-format: schema, archive, verifier, renderers, scanner@neurcode-ai/share: focused CLI and local Composer@neurcode-ai/share-viewer: verified local/static rendering helpers@neurcode-ai/share-sdk: public hosted client contracts
- Text/code files only; binary assets are not captured.
- A Share contains an explicit cut, not a live repository mirror.
- Secret scanning is heuristic.
- Hosted publishing depends on the separately operated Neurcode service.
- Browser-only hosted creation is available at share.neurcode.com/new, not in this open-core repository.
Read CONTRIBUTING.md, GOVERNANCE.md, and the release policy. Security reports should follow SECURITY.md.
Apache-2.0 covers Neurcode-owned source here. Trademarks remain reserved as described in TRADEMARKS.md.