npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

@veritio/claude-code

v0.4.7

Published

Capture Claude Code agent activity as Veritio evidence via hooks, with a local file sink or ingest POST.

Readme

@veritio/claude-code

Part of Veritio, the open-source evidence layer — see the docs and Veritio Cloud.

Capture Claude Code agent activity as Veritio evidence — passively, via hooks — and query it back through an MCP server.

The hook maps each Claude Code event to the @veritio/core provenance recorder and appends a hash-chained, redacted evidence trail to a durable local store (and, optionally, to a Veritio ingest endpoint). Capture is out-of-band, so the trail does not depend on the agent choosing to report. A companion MCP server lets a human or another agent list sessions, inspect a session's provenance graph, and export a verifiable bundle.

Privacy: raw prompts, tool argument payloads (Bash commands, MCP arguments — which can carry secrets), and file contents/diffs are never persisted. Stable ids and content hashes travel. Raw file paths can be retained temporarily in local per-session state to pair a pre-image with its post-image; they are not sent as evidence metadata. Redaction runs before anything reaches a sink.

What is captured

| Claude Code event | Veritio record | |---|---| | SessionStart | agent.session.started (+ caused_by edge to the enforcing human) | | UserPromptSubmit | agent.prompt.recorded (prompt hash only) | | PreToolUse (Edit/Write/MultiEdit) | pre-image content hash cached for the matching PostToolUse | | PostToolUse / PostToolUseFailure | agent.tool.called (succeeded/failed) + a code change with before/after hashes | | Stop | a git status turn-scan records Bash-driven file changes the edit hooks miss | | SessionEnd | finalizes per-session state |

Configure the hook

The hook runs under Bun (the published dist/ consumes the Veritio SDK, which is bundler/Bun-resolved). Add to your project's .claude/settings.json (${CLAUDE_PROJECT_DIR} is provided by Claude Code):

{
  "hooks": {
    "SessionStart":       [{ "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }],
    "UserPromptSubmit":   [{ "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }],
    "PreToolUse":         [{ "matcher": "Edit|Write|MultiEdit", "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }],
    "PostToolUse":        [{ "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }],
    "PostToolUseFailure": [{ "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }],
    "Stop":               [{ "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }],
    "SessionEnd":         [{ "hooks": [{ "type": "command", "command": "bun ${CLAUDE_PROJECT_DIR}/node_modules/@veritio/claude-code/dist/hook.js" }] }]
  }
}

The hook always exits 0 — a logging hook never blocks the agent.

Install an exact reviewed version in repositories and hosted runners; do not use an unversioned bunx command. The bundled plugin pins @veritio/[email protected]. Treat it as unavailable until exact registry readback succeeds; the plugin must fail closed rather than fall back to 0.4.5.

The veritio CLI device-login helper is not published yet. From a repository checkout, build it with bun run --cwd cli build and invoke bun cli/dist/index.js login claude; do not assume a global veritio binary.

Configuration (environment)

Read only at the process boundary; no credential is embedded in the hook.

| Variable | Default | Purpose | |---|---|---| | VERITIO_LOCAL_DIR | ~/.veritio/claude-code | Local evidence store directory (always written) | | VERITIO_TENANT_ID | local | Tenant scope on every record | | VERITIO_ACTOR_ID | local_developer | Stable id of the enforcing human (never an email) | | VERITIO_AGENT_ACTOR_ID | agent_claude_code | Stable id of the agent actor | | VERITIO_ENVIRONMENT | development | Scope environment | | VERITIO_WORKSPACE_ID | — | Optional workspace scope | | VERITIO_INGEST_URL + VERITIO_INGEST_KEY | — | If both set, also POST records to a Veritio ingest endpoint (e.g. Veritio Cloud), so captured sessions surface in the hosted Sessions UI. The server re-redacts. | | VERITIO_INGEST_TIMEOUT_MS | 10000 | Abort bound (ms) for one ingest POST, constrained to 1..30000. The hook still exits 0 because the local store is authoritative for capture. | | VERITIO_SPOOL_HARD_BATCHES | 250 | Optional lower queue batch ceiling. It cannot raise the compiled-in hard ceiling. | | VERITIO_SPOOL_HARD_BYTES | 50000000 | Optional lower queue byte ceiling. It cannot raise the compiled-in hard ceiling. |

Durable delivery queue

Remote delivery has three durable outcomes:

  • retry: transport failures, HTTP 429, and legacy 5xx responses remain in the pending queue.
  • pause: an explicit server pause (including recognized legacy quota codes) opens a sticky circuit and moves the entire queue to held. Later hooks append locally and make zero remote attempts.
  • reject: permanent failures are moved to quarantine with the redacted payload intact; they are never silently deleted.

Ordinary hook invocations never replay a backlog. Recovery requires a separate operator command and every replay epoch is bounded by batches, records, encoded bytes, and elapsed time. This prevents an endpoint recovery, quota upgrade, or misclassified provider failure from turning many routine hooks into an uncontrolled egress drain.

The queue lives under <localDir>/spool/ with pending, held, and quarantine states. The hard ceiling is 250 batches or 50 MB. At the ceiling, existing entries are retained, pending entries move to held, and the newest remote-delivery copy is refused with a visible stderr signal. Older flat spool files upgrade into a manual hold; they do not auto-replay.

Use the operator CLI to inspect and control delivery:

veritio-claude-code-spool status
veritio-claude-code-spool pause --reason "provider transfer alarm"
veritio-claude-code-spool quarantine

# At most one request, 500 records / 1 MB / 5 seconds.
veritio-claude-code-spool canary

# Every drain budget is mandatory and is checked before each request.
veritio-claude-code-spool drain \
  --max-batches 10 \
  --max-records 5000 \
  --max-bytes 10000000 \
  --max-elapsed-ms 15000

# Resume only moves held entries back to pending; it performs no network I/O.
veritio-claude-code-spool resume --acknowledge "provider headroom reviewed"

status, pause, resume, and quarantine work without ingest credentials. canary and drain require both ingest variables. CLI output contains queue metadata, never event payloads or credentials. Spool payload files contain the same redacted, hash-only batch prepared for the wire. The queue is currently TypeScript-only; another capture adapter must reproduce the same disposition and replay-permit semantics (see .claude/rules/02-sdk-parity.md).

GitHub-hosted Claude Code

Anthropic's GitHub Action has its own paid model/API and runner exposure. Bound that workflow separately with a narrow event trigger, GitHub concurrency, a job timeout-minutes, and Claude's --max-turns. Veritio's hook and spool do not cap Claude tokens or GitHub runner minutes.

If the action is configured to load repository Claude hooks, install the exact reviewed @veritio/claude-code version before it runs and pass Veritio ingest credentials only through GitHub Secrets. Omit those credentials for local-only artifact capture. Never run the stress suite or a spool drain from a pull-request workflow; CI uses synthetic hooks and no paid provider credentials.

Query + export (MCP)

The package also ships a read-only MCP server (veritio-claude-code-mcp, dist/mcp.js) over stdio with three tools:

  • veritio.list_sessions(day?) — summarized sessions (enforcing human, agent/model, branch, change count, outcome).
  • veritio.get_session(sessionId) — a session's events + projected provenance graph.
  • veritio.export_session(sessionId) — a verifiable evidence bundle (records + hash-chain verdict).

Register it with your MCP client (reads the same VERITIO_LOCAL_DIR):

{ "mcpServers": { "veritio-provenance": { "command": "bun", "args": ["/abs/path/node_modules/@veritio/claude-code/dist/mcp.js"] } } }

Cross-language parity

This adapter is TypeScript-only today (it matches the TS-only provenance recorder). The hook→recorder mapping table above is the language-neutral contract; a Python/Go capture adapter must reproduce it, including the metadata.sessionId stamp (see .claude/rules/02-sdk-parity.md).