@grwnd/pi-governance 1.4.1 → 1.5.0

package/README.md

@@ -10,7 +10,8 @@
 
 <p align="center">
   <a href="https://github.com/Grwnd-AI/pi-governance/actions/workflows/ci.yml"><img src="https://github.com/Grwnd-AI/pi-governance/actions/workflows/ci.yml/badge.svg" alt="CI" /></a>
-  <a href="https://www.npmjs.com/package/@grwnd/pi-governance"><img src="https://img.shields.io/npm/v/@grwnd/pi-governance" alt="npm" /></a>
+  <a href="https://www.npmjs.com/package/@grwnd/pi-governance"><img src="https://img.shields.io/npm/v/@grwnd/pi-governance" alt="npm pi-governance" /></a>
+  <a href="https://www.npmjs.com/package/@grwnd/openclaw-governance"><img src="https://img.shields.io/npm/v/@grwnd/openclaw-governance?label=openclaw-governance" alt="npm openclaw-governance" /></a>
   <a href="https://github.com/Grwnd-AI/pi-governance/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-Apache--2.0-blue" alt="License" /></a>
   <a href="https://grwnd-ai.github.io/pi-governance/"><img src="https://img.shields.io/badge/docs-GitHub%20Pages-blue" alt="Docs" /></a>
 </p>
@@ -24,6 +25,7 @@
 - **Role-based access control** — define who can use which tools
 - **Bash command classification** — auto-block dangerous commands (`rm -rf`, `sudo`, `curl | sh`)
 - **Path-level file gating** — restrict read/write to scoped directories
+- **Data loss prevention** — detect and block/mask API keys, tokens, and PII before they reach the LLM
 - **Human-in-the-loop approval** — require sign-off for sensitive operations
 - **Audit logging** — structured JSONL logs of every governance decision
 - **Prompt-level policy** — role-scoped system prompt templates
@@ -43,6 +45,7 @@ That's it. On next session start, governance is active with sensible defaults:
 - All tools allowed
 - Dangerous bash commands blocked
 - Supervised mode (approval required for writes and bash)
+- DLP disabled (opt-in)
 - Audit logged to `~/.pi/agent/audit.jsonl`
 
 ### Configure
@@ -153,6 +156,7 @@ User message → Pi Agent Runtime
               ┌─────┴──────────┐
               │ onBeforeToolCall │  ← RBAC: tool allowed?
               │  → classify bash │  ← Path check
+              │  → DLP scan      │  ← Block/mask secrets & PII
               │  → HITL approval │  ← Audit log
               └─────┬──────────┘
                     │
@@ -160,10 +164,42 @@ User message → Pi Agent Runtime
                     │    └→ Return denial message
                     │
               ┌─────┴──────────┐
-              │ onAfterToolCall  │  ← Audit result
+              │ onAfterToolCall  │  ← DLP scan output
+              │                  │  ← Audit result
               └────────────────┘
 ```
 
+## Data Loss Prevention
+
+DLP prevents secrets and PII from leaking through tool calls to LLM providers. It scans both inputs (before execution) and outputs (before reaching the LLM).
+
+```yaml
+dlp:
+  enabled: true
+  mode: mask # audit | mask | block
+  on_input: block # block tool calls with secrets
+  on_output: mask # redact secrets in tool output
+  masking:
+    strategy: partial # partial | full | hash
+    show_chars: 4
+  severity_threshold: low
+  built_in:
+    secrets: true # AWS keys, GitHub PATs, JWTs, Stripe keys, ...
+    pii: true # SSN, credit cards, email, phone, IP
+  custom_patterns:
+    - name: internal_key
+      pattern: 'grwnd_[a-zA-Z0-9]{32}'
+      severity: critical
+      action: block
+  allowlist:
+    - pattern: '127\.0\.0\.1'
+  role_overrides:
+    admin:
+      enabled: false # admin skips DLP
+```
+
+DLP is **disabled by default** — zero behavioral change for existing users. See the full [DLP guide](https://grwnd-ai.github.io/pi-governance/guide/dlp) and [pattern reference](https://grwnd-ai.github.io/pi-governance/reference/dlp-patterns).
+
 ## Dual Policy Engine
 
 Choose between two policy engines:
@@ -190,12 +226,40 @@ pi-governance works with [OpenClaw](https://github.com/Grwnd-AI) out of the box.
 
 ```
 OpenClaw gateway (WhatsApp, Discord, Telegram, …)
-  └─ Pi embedded runner
-      └─ @grwnd/pi-governance extension
-          ├─ RBAC for MCP tools (create_report, upload_asset, …)
-          ├─ bash command classification
-          ├─ audit logging (JSONL + webhook)
-          └─ HITL approval flow
+  └─ [optional] @grwnd/openclaw-governance plugin → channel identity
+      └─ Pi embedded runner
+          └─ @grwnd/pi-governance extension
+              ├─ RBAC for MCP tools (create_report, upload_asset, …)
+              ├─ bash command classification
+              ├─ audit logging (JSONL + webhook)
+              └─ HITL approval flow
+```
+
+### Get up and running
+
+```bash
+# 1. Install the Pi governance extension
+pi install npm:@grwnd/pi-governance
+
+# 2. Install the OpenClaw identity bridge plugin
+openclaw plugins install @grwnd/openclaw-governance
+```
+
+Create `openclaw-users.yaml` to map channel users to governance roles:
+
+```yaml
+users:
+  whatsapp:+15550123:
+    role: report_author
+    org_unit: field-ops
+  discord:428374928374:
+    role: analyst
+  slack:U04ABCD1234:
+    role: project_lead
+    org_unit: engineering
+default:
+  role: analyst
+  org_unit: default
 ```
 
 Put MCP tool names directly in your policy rules:
@@ -220,7 +284,7 @@ roles:
     token_budget_daily: 500
 ```
 
-Every MCP tool call is audited as structured JSON — see who did what, when, and whether it was approved or denied.
+When a WhatsApp user messages your OpenClaw agent, the identity bridge parses the session key, maps them to a role, and pi-governance enforces the policy — all automatically. Every MCP tool call is audited as structured JSON.
 
 See the full [OpenClaw integration guide](https://grwnd-ai.github.io/pi-governance/guide/openclaw) for MCP tool reference tables, channel identity mapping, and common patterns.
 
@@ -230,8 +294,10 @@ Full documentation at **[grwnd-ai.github.io/pi-governance](https://grwnd-ai.gith
 
 - [Quick Start](https://grwnd-ai.github.io/pi-governance/guide/quickstart)
 - [Team Deployment](https://grwnd-ai.github.io/pi-governance/guide/team-deployment)
+- [OpenClaw Integration](https://grwnd-ai.github.io/pi-governance/guide/openclaw)
 - [YAML Policies](https://grwnd-ai.github.io/pi-governance/guide/yaml-policies)
 - [Bash Classifier](https://grwnd-ai.github.io/pi-governance/guide/bash-classifier)
+- [Data Loss Prevention](https://grwnd-ai.github.io/pi-governance/guide/dlp)
 - [Configuration Reference](https://grwnd-ai.github.io/pi-governance/reference/config)
 
 ## License