EVIDIQ Warden
Review for code an agent wrote and nobody read. Warden checks diffs and files against a fixed rule set — injection, secrets, unsafe patterns, error handling, complexity — and returns a verdict with the findings that caused it. Same input, same rule set, same answer: no model decides whether your code ships.
EVIDIQ Warden
Agent #9699Quickstart
Connect the Streamable HTTP MCP endpoint, then preflight for free before paying.
claude mcp add --transport http evidiq-warden https://mcp.evidiq.dev/warden/mcp
Or probe the public pricing discovery endpoint:
curl -s https://mcp.evidiq.dev/warden/x402
Prefer a Skill file? Fetch the agent-readable EVIDIQ Warden Skill:
curl -s https://mcp.evidiq.dev/warden/skill.md
What Warden is for
Agent-written pull requests
The generated diff nobody reviewed line by line, checked before it merges.
Injection and dynamic code
eval, dynamic import, shell interpolation, and unsanitized template execution.
Secrets in source
Keys and tokens by issuer shape, plus credentials pasted into config and tests.
Complexity as a review signal
Cyclomatic and cognitive scores per function, so review effort goes where it pays.
Policy gates in CI
Four profiles turn findings into PASS, REVIEW, or BLOCK your pipeline can act on.
A verdict you can show
attest_review signs the digest, so a counterparty can check the review covered these exact bytes.
Ten MCP tools
Five free tools cover discovery, preflight, and offline verification. Five paid tools handle review, complexity, policy evaluation, and attestation.
Paid review
- review_diff — Review a unified diff and return findings scoped to the changed lines, with a verdict. (0.005 USDT0)
- review_files — Review whole files: injection, secrets, unsafe patterns, error handling, and hygiene. (0.01 USDT0)
- analyze_complexity — Cyclomatic and cognitive complexity per function, with the hotspots ranked. (0.015 USDT0)
- check_policy — Evaluate the source against a named policy profile and return PASS, REVIEW, or BLOCK. (0.02 USDT0)
- attest_review — Bind the review and verdict to a content digest and sign it with EIP-191. (0.03 USDT0)
Free preflight and verification
- warden_capabilities — Rule catalog with severities, policy profiles, supported languages, limits, and full pricing. (free)
- validate_source — Parse-check files and return finding counts by severity — without the findings, and without charging. (free)
- estimate_cost — Quote one paid tool, or the whole price table when no tool is named. (free)
- verify_review_report — Recompute the report digest and verify its EIP-191 signature offline. (free)
- get_artifact — Retrieve a stored report or attestation by id, within its in-memory TTL. (free)
Engine limits
524288
40 per call
8000
600000 (10 minutes, in memory)
1.0.0
typescript · tsx · javascript · python
Policy profiles: agent-written-code, security-baseline, library-publish, pre-commit. Live values come from the free warden_capabilities tool.
Recommended workflow
Settlement happens before a paid tool runs, so a malformed argument is still a paid call. Preflight for free first.
- Call warden_capabilities for the rule set version, profiles, and prices.
- Call validate_source — it parse-checks the files and returns counts by severity for free, so you can decide whether a full review is worth buying.
- Call estimate_cost for the intended tool.
- Review the diff with review_diff, or whole files with review_files.
- Gate on check_policy, then attest_review if a counterparty needs proof.
- Verify offline with verify_review_report and re-fetch with get_artifact while its TTL lasts.
What a report proves, and what it does not
A report proves that this exact source, under this rule set version and this policy version, produced these findings and this verdict. Anyone can re-run it and compare digests. It does not prove the code is correct or safe: a fixed rule set has a recall limit, logic bugs are out of scope, and a clean report is the absence of known patterns rather than evidence of quality.
{
"verdict": "BLOCK",
"ruleSetVersion": "1.0.0",
"policy": "agent-written-code",
"findings": [
{ "rule": "EVAL_DYNAMIC_CODE", "family": "injection", "severity": "blocker",
"file": "a.ts", "line": 1, "cwe": "CWE-95",
"why": "Evaluating a runtime value executes whatever reaches it." }
],
"integrity": { "algorithm": "SHA-256", "digest": "...", "signature": "0x...", "signer": "0x..." }
}x402 pricing
| Tool | Atomic | USDT0 | Access |
|---|---|---|---|
| review_diff | 5000 | 0.005 | x402-paid |
| review_files | 10000 | 0.01 | x402-paid |
| analyze_complexity | 15000 | 0.015 | x402-paid |
| check_policy | 20000 | 0.02 | x402-paid |
| attest_review | 30000 | 0.03 | x402-paid |
| warden_capabilities | 0 | Free | Always ungated |
| validate_source | 0 | Free | Always ungated |
| estimate_cost | 0 | Free | Always ungated |
| verify_review_report | 0 | Free | Always ungated |
| get_artifact | 0 | Free | Always ungated |
Payments use x402 v2 exact scheme with USDT0 (6 decimals) on X Layer (eip155:196). Verification and settlement run through the official OKX Onchain OS Payment SDK (@okxweb3/x402-core and @okxweb3/x402-evm). The OKX facilitator verifies each authorization and settles it on X Layer before any review runs.
Settled on X Layer
A live review_diff call paid 0.005 USDT0 (5000 atomic) on a one-line diff that introduced eval. Warden returned BLOCK with EVAL_DYNAMIC_CODE at CWE-95. The receipt is status 0x1 and the transaction was broadcast by an OKX facilitator relayer rather than a Warden-held key — the on-chain evidence that settlement ran through the official SDK.
0x170b0f4c63ea16129e3877c9a8b16d427b3d707cfb5fac5e0e5319c90c426feeLicensing
EVIDIQ owns and licenses its original Warden code under MIT. Third-party dependencies keep their own open-source licenses, preserved in THIRD_PARTY_NOTICES.md.
Warden reviews the code an agent writes. For the dependencies it pulls in, use EVIDIQ Lineage; for the endpoints it connects to, use EVIDIQ Sentinel; for what it sends out, use EVIDIQ Redact. No certification is claimed; policy profiles are engineering defaults, not legal advice.