InstructLint · instruction drift verification

Check whether your agent instructions match your repo.

Audit AGENTS.md, CLAUDE.md, Copilot, and Cursor instructions against repository evidence. Preview public repos for free; unlock the full report for $1 USDC.

Public GitHub repositories only. Read-only, bounded, no repo changes, and no database persistence.

verification.console

$ instructlint preview https://github.com/acme/repo

scanning instruction surfaces...

HIGH AGENTS.md requires tests, but no test command resolves

MED Cursor rules name pnpm; packageManager is npm@11.5.2

preview ready. evidence + suggested patch unlock: $1 USDC

Public GitHub preview: free

Full evidence + suggested patch: $1 USDC on Base

Output: drift findings with repo proof

Client preview

Paste a public GitHub repository URL.

Get the score, detected instruction files, finding counts, and top drift signals. Evidence and repair patches unlock only in the full report.

Ready for a public GitHub repository URL.

See the score, detected instruction files, finding counts, and top drift signals. Evidence and repair patches stay in the paid report.

Sample findings

What InstructLint flags

High

Instruction promises tests that do not exist

AGENTS.md requires regression coverage before cleanup, but the repo has no test target for the modified package.

Medium

CLAUDE.md references a retired deploy path

The documented release checklist still points agents at a removed Vercel project and stale environment names.

Low

Cursor rule conflicts with package scripts

The editor instruction says to run pnpm, while package metadata declares npm as the package manager.

Methodology

Evidence before recommendations

Supported claims are checked against repository structure, package metadata, scripts, deployment configuration, and instruction hierarchy. Unsupported prose is never guessed. Each response reports whether its bounded preview was truncated.

  1. Discover agent-facing instructions: AGENTS.md, CLAUDE.md, Copilot, Cursor, and nested overrides.
  2. Map supported claims to repo evidence: paths, package scripts, package manager, lockfiles, deploy config, and nested scope.
  3. Score deterministic findings by severity before suggesting the smallest instruction repair.
  4. Return a free public preview first; the paid report unlocks evidence trails and a suggested instruction patch.

Pricing

Simple by design

$0

Public GitHub preview with stable, high-signal drift summary.

$1 USDC

Full evidence trail, confidence notes, and suggested patch.

View payment flow

FAQ

Operational details

Is the preview really free?

Yes. Public GitHub repositories can request a free preview with stable findings. The full report costs $1 USDC.

What happens to the repository data?

The service reads a bounded snapshot from fixed GitHub origins, returns the audit, and stores neither the repository snapshot nor the result in a database.

What does the paid report include?

Full evidence, confidence notes, drift severity, and a suggested patch for the instruction files.

Do you change my repository?

No. The product returns findings and patches for review. Applying them remains your decision.

When does the $1 payment settle?

The x402 client authorizes one USDC payment. Settlement happens only after the full report handler returns successfully; invalid requests and failed audits are not settled.

What if GitHub rate-limits the scan?

The API returns a retryable rate-limit error. No paid report is settled when repository retrieval fails.