メニュー
PACT governance — D/T/R addressing, envelopes, clearance, governed agents. Custom authz BLOCKED.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Claude Code architecture — artifact design, context, agentic patterns. For CC audit/build.
Conformance Walk — freeze-then-judge on the source→delivered→live axis: one cw_core + Source/Delivered/Live adapter families, coverage vs pass-rate, discrete verdicts. Use for any testable surface.
Kailash security (Python) — validation, secrets, injection, authn/z. Hardcoded secrets BLOCKED.
/onboard procedure: read roster + team-memory + posture + claims + codify lease + rules-changed for a new operator joining a multi-operator COC repo.
/certify procedure: brief → probe → gate at 100%; loops failed questions until pass. NO Claude-assistance during gate phase. Curated bank, not LLM-generated.
/ecosystem-init procedure — write the D6 ecosystem-config, run the disclosure scan before write, establish genesis via runEnrollmentCeremony, scaffold non-Kailash STACK.md.
| name | 29-pact |
| description | PACT governance — D/T/R addressing, envelopes, clearance, governed agents. Custom authz BLOCKED. |
Quick reference for PACT organizational governance patterns.
pip install kailash-pact # Governance framework
pip install kailash>=2.0.0 # Core SDK with trust subsystem
pip install kailash-kaizen>=2.0.0 # For governed Kaizen agents
| Skill | Use When |
|---|---|
| pact-quickstart | Getting started, first GovernanceEngine |
| pact-governance-engine | Engine API, verify_action, compute_envelope |
| pact-dtr-addressing | D/T/R grammar, Address parsing |
| pact-envelopes | Three-layer model, monotonic tightening |
| pact-access-enforcement | 5-step algorithm, clearance, bridges, KSPs |
| pact-governed-agents | PactGovernedAgent, @governed_tool |
| pact-kaizen-integration | Wrapping Kaizen agents with governance |
| pact-mcp-governance | MCP tool governance: enforce, audit, middleware |
| pact-conformance-features | N1-N6: KnowledgeFilter, EnvelopeCache, PlanSuspension, AuditTiers, ObservationSink, cross-SDK vectors |
from pact.governance import GovernanceEngine, GovernanceVerdict
from kailash.trust.pact.config import (
ConstraintEnvelopeConfig, OrgDefinition,
TrustPostureLevel, VerificationLevel,
ConfidentialityLevel,
)
from kailash.trust.pact.agent import PactGovernedAgent
from kailash.trust.pact.audit import AuditChain
# MCP governance
from pact.mcp import (
McpGovernanceEnforcer, McpGovernanceMiddleware, McpAuditTrail,
McpToolPolicy, McpGovernanceConfig, McpActionContext,
)
Five canonical posture levels with autonomy gradient:
| Canonical | Autonomy | Ceiling | Old Name (alias) |
|---|---|---|---|
| PSEUDO | 1 | PUBLIC | PSEUDO_AGENT |
| TOOL | 2 | RESTRICTED | (new, no old) |
| SUPERVISED | 3 | CONFIDENTIAL | SHARED_PLANNING |
| DELEGATING | 4 | SECRET | CONTINUOUS_INSIGHT |
| AUTONOMOUS | 5 | TOP_SECRET | DELEGATED |
Old names work as enum aliases (TrustPostureLevel.PSEUDO_AGENT resolves to PSEUDO). String deserialization of old values is handled by _missing_().
See .claude/rules/pact-governance.md for security invariants.