← ClaudeAtlas

demonstrate-governancelisted

Exercise the four-stage governance pipeline - scope check, secret scan, blocklist, rate limit - through the systemprompt MCP hub, then read back the audited decisions
systempromptio/systemprompt-demo · ★ 3 · AI & Automation · score 66
Install: claude install-skill systempromptio/systemprompt-demo
# Demonstrate Governance Explain the four-stage governance pipeline and **prove two of its stages live**, using only the `systemprompt` MCP documentation hub. Every tool call in this session — including each one below — passes through the same synchronous check before it executes. ## When to Use Use this to show that governance is enforcement rather than description: that a call is stopped *before* it runs, that the reason is specific, and that the decision is written down. ## The pipeline Four policies run in order on every tool call, configured in `services/governance/config.yaml`: | Stage | Policy id | What it blocks | |-------|-----------|----------------| | Scope check | `scope_check` | A non-admin caller reaching for an admin-only tool prefix (`mcp__admin__*`) | | Secret scan | `secret_scan` | Plaintext credentials in any tool input (35+ patterns), any scope | | Blocklist | `tool_blocklist` | Tool names matching a blocked pattern, for non-admin scope | | Rate limit | `rate_limit` | More than 300 calls per 60s for one identity | Scope comes from the caller's live database roles, not from anything the agent says about itself. `admin` callers are exempt from `scope_check` and `tool_blocklist` — but **this terminal caps every caller at `user` scope**, whatever their roles say, so those exemptions never apply here. It is a sandboxed demo surface, not an admin console; an operator signed in as admin sees exactly the enforcement a visitor sees. `secret_scan` and `rate_l