Ward — Development skill for Claude Code

Ward

A fast, local CLI that integrates with [Claude Code hooks](https://docs.anthropic.com/en/docs/claude-code/hooks) to block PII, secrets, and credentials from leaking through your AI coding workflow. Ships as a single Rust binary — no runtime dependencies.

What It Does

Ward scans every prompt you send and every tool call Claude makes, blocking sensitive data before it leaves your machine.

Subcommand	Hook Events	Purpose
ward pii	UserPromptSubmit, PreToolUse, PostToolUse	Block SSNs, credit cards, emails, phone numbers
ward leaks	UserPromptSubmit, PreToolUse, PostToolUse	Block API keys, cloud credentials, tokens, passwords, private keys, connection strings
ward log	All events	Structured event logging to ~/.ward/events.jsonl

On UserPromptSubmit and PreToolUse, a match **blocks** (exit 2) before anything leaves your machine. On PostToolUse the tool has already run, so blocking is impossible — instead ward **redacts**: it rewrites the tool output via the hook's `updatedToolOutput` field, replacing each secret with a `[WARD LEAKS REDACTED: ]` (or `WARD PII`) marker before Claude ever sees it. If a match can't be masked in place, the whole output is withheld.

Detection Coverage

PII Patterns

• Social Security Numbers (with ISO date false-positive prevention)
• Credit card numbers (spaces and dashes)
• Email addresses
• US phone numbers

Leak Detection — 3 Tiers

**Tier 1: Provider-Specific Keys** (32 patterns, high confidence)

• AWS (access key, secret key, Bedrock), GCP, Azure
• Anthropic, OpenAI
• GitHub (PAT, fine-grained, OAuth, app tokens), GitLab
• Slack (bot, app, user, webhook, legacy tokens)
• Stripe, SendGrid, Twilio, Linear, Databricks
• Heroku, Vercel, Grafana, Sentry, Datadog
• HashiCorp Vault, NPM, PyPI

**Tier 2: Structural Patterns** (medium confidence)

• PEM-encoded private keys (RSA, EC, OPENSSH)
• JWT tokens
• Connection strings (postgres, mysql, mongodb, redis, amqp,