~ knotless Search/
↑↓ move↵ open⇧↵ all results

johnandersen777.bsky.social/vuln-notif

0
0
main
Go to filet
Code
README.md

vuln-notif graph viewer

Static, backend-free viewer for the atproto records vuln-notif uploads. It renders a force-directed graph of records and the links between them, caches the records in IndexedDB, and stays live through the Bluesky jetstream.

Build

From the repository root:

python3 -m py_compile spa/build.py
python3 spa/build.py --db .state/prod-week.db
jq '.counts' spa/dist/graph.json

Deploy the built site to wisp.place (credentials from the environment):

export ATP_HANDLE=...
export ATP_APP_PASSWORD=...
./spa/deploy.sh

WISP_SITE defaults to cve; CACHE_DB overrides the SQLite source. The script rebuilds dist/, copies the vendored dependencies from spa/vendor/, and runs wispctl deploy --spa so client-side routes fall back to index.html. No credentials are stored in the script or repository.

Vendored runtime assets (no CDN at load time):

  • D3 7.9.0 (spa/vendor/d3.min.js)
  • Pico CSS 2.1.1 (spa/vendor/pico.min.css)
  • highlight.js 11.9.0 JS + GitHub theme CSS

The only network calls at runtime are the live PDS/PLC/jetstream services.

Options:

python3 spa/build.py --db .state/prod-week.db --out /tmp/graph.json
python3 spa/build.py --db .state/prod-week.db --repo-did did:plc:...

--db defaults to $CACHE_DB, then .state/prod-week.db. --repo-did defaults to the first DID found in the cache. The script only reads SQLite and uses the Python standard library.

spa/dist/config.json is written next to the viewer:

{
  "repo_did": "did:plc:...",
  "generated_at": "2026-01-01T00:00:00Z"
}

graph.json is still written for backward compatibility and offline tooling. The SPA does not read it: the browser builds its own mirror from the PDS.

Serve

python3 -m http.server 8092 --directory spa/dist

Open http://127.0.0.1:8092/. Any static file server works; no backend is involved.

Seeding

On first load the viewer opens the vuln-notif-graph IndexedDB database (version 1) and checks the records store. When it is empty it resolves the repo's PDS and seeds from the read-only XRPC endpoint:

GET <pds>/xrpc/com.atproto.repo.listRecords?repo=<did>&collection=<nsid>&limit=100

It follows cursor until exhausted or until roughly 1000 records are seen per collection. Records, links, and meta values are written in batched IndexedDB transactions, and #status shows seeding collection n/7. Later loads find the cache non-empty and skip seeding entirely, rendering immediately.

PDS resolution uses the DID document at https://plc.directory/<did> and picks the service entry whose id ends with #atproto_pds. A pds_url field in config.json overrides resolution. The PDS must allow browser CORS requests for com.atproto.repo.listRecords; public Bluesky PDS hosts send access-control-allow-origin: *. The viewer never writes to the PDS.

Live updates

The viewer connects to wss://jetstream1.us-east.bsky.network/subscribe as soon as the repo DID is known, in parallel with seeding. The URL carries wantedDids=<did> and one wantedCollections parameter per indexed collection. The connection is opened even while seeding is still running, and live events update both IndexedDB and the graph.

Only kind: "commit" frames are handled. create and update put the record and replace the links whose source is that record; delete removes the record and those links. Identity, account, and other frame kinds are ignored. The time_us of each commit is stored in meta as jetstreamCursor and passed back as cursor= on reconnect. If the socket closes or errors, the viewer reconnects automatically forever with exponential backoff and jitter (1s up to 30s); no user action is needed. #status shows connecting, live, reconnecting, or error, plus node/edge counts and the number of records mirrored in IndexedDB.

IndexedDB

Database vuln-notif-graph, version 1:

  • records, keyed by uri: {uri, collection, rkey, did, cid, record, indexedAt}
  • edges, keyed by [source, target, label]: {source, target, label, collection}
  • meta, key/value: repoDid, pdsUrl, jetstreamCursor, seededAt, lastEventAt

Clear and reseed from the browser console:

await window.__vulnGraph.clear();   // empty IndexedDB and the graph
await window.__vulnGraph.reseed();  // clear, then seed again from the PDS
window.__vulnGraph.counts();        // {records, edges}
window.__vulnGraph.db;              // IDBDatabase handle

Or delete the database in DevTools > Application > IndexedDB and reload.

Offline behavior: a previously seeded database still renders without a network. Seeding fails gracefully and leaves cached records in place; the jetstream keeps retrying in the background. Live-only updates are lost while offline and return when the socket reconnects.

Indexed collections

  • com.nahdig.cve.atprotoV1
  • software.release.project
  • software.release.channel
  • software.release.version
  • software.release.manifest
  • software.release.status
  • com.nahdig.release.projectMeta

com.nahdig.cve.nvdV2, com.nahdig.software, and com.nahdig.purl are not indexed or rendered.

Edges follow src/links.rs: nvd.uri, software[].project.uri, software[].channels[].uri, software[].versions[].uri, software[].manifests[].uri, parent, renamedTo, project, manifest.uri, version.uri, subject, and project meta project. StrongRefs ({uri, cid}) contribute their URI; duplicate links are dropped.

Use

  • Click a node to open its detail modal. The modal shows the collection and rkey, the at-URI, the DID, and the record as highlighted YAML.
  • Open on pdsls.dev opens the record in a new tab.
  • Show connected only hides everything except the selected node, its 1-hop neighbors, and the edges among them. Press it again (or close the modal, or press Escape) to restore the graph.
  • Legend checkboxes hide or show collections.
  • Drag nodes, pan by dragging the background, and scroll to zoom.

Checks

python3 -m py_compile spa/build.py
python3 spa/build.py --db .state/prod-week.db
jq '.counts' spa/dist/graph.json

Deterministic live-ingest smoke test from the browser console:

await window.__vulnGraph.ingest({
  kind: "commit",
  commit: {
    operation: "create",
    collection: "software.release.project",
    rkey: "test-live-node",
    record: { "$type": "software.release.project", name: "test-live-node" },
    cid: "bafytest",
  },
});

There is no npm or Node build step.