trigger	always_on
last_updated	2026-03-23
pc2e_version	1.0

PC2E Agent Governance Framework

Production-ready governance for deterministic AI agents

Quick Start • Documentation • Examples • Contributing

---

Why PC2E?

Enterprise AI adoption is blocked by unpredictable behavior. This framework implements Predict, Communicate, Explain (PC2E) principles to ensure:

• Predictability: Confidence scores for every technical decision.
• Transparency: Core assumptions declared BEFORE major actions are taken.
• Auditability: A clear, reproducible chain of reasoning for all system changes.

Production Results

• ✅ 20+ applications in production environment.
• ✅ 0 regressions in 60 days of operations.
• ✅ 100% task completion rate (all exit criteria strictly met).
• ✅ Deterministic behavior across all operating modes.

---

Quick Start

Integrate the PC2E framework into your AI development workflow in minutes:

1. Clone the Framework

Deploy the governance layer directly into your agent's workspace:

git clone https://github.com/thegeekybeng/pc2e-agent-governance.git .agent

2. Configure Local Context

Map the .agent directory to your AI assistant's context. For VS Code/Cline users, adding the following to your System Prompt ensures mandatory compliance:

"Prioritize rules in the .agent/ directory following the hierarchy: Global > Mode > Workflow."

3. Verify Operational Gates

Run a test task and verify that the agent emits a Confidence Score and Chain of Reasoning as mandated by the PC2E Protocol.

---

Overview

The PC2E Agent Governance Framework is a production-grade orchestration and governance layer designed to ensure predictable, communicative, and verifiable AI agent operations. This framework provides the necessary structure to manage complex multi-agent workflows with zero technical debt and maximum architectural scalability.

Framework Architecture

The framework is organized into four logical tiers to ensure clear separation of concerns and rule precedence:

1. Global Governance (global/) - Universal mandates that apply to all operations.
2. Operational Modes (modes/) - Context-specific execution standards (Orchestrator, Architect, Code, etc.).
3. Standardized Workflows (workflows/) - Optimized procedures for specific task execution.
4. Governance Artifacts (templates/) - Standardized records for decision-making and evaluation.

Precedence Logic

Rules are applied in descending order of authority: Global > Mode > Workflow. In the event of a conflict, the higher-tier rule maintains absolute precedence.

---

Operational Navigation

Fundamental Principles

• PC2E Framework: Review the core philosophy and implementation standards.
• Operational Imperatives: Understand the 4 Core Imperatives (Scalability, Security, Zero Debt, Privacy).

Execution Modes

• Multi-step Planning: Orchestrator Mode
• System Design: Architect Mode
• Implementation: Code Mode
• Troubleshooting: Debug Mode
• Research & Analysis: Ask Mode

Specialized Procedures

• Infrastructure Management: Docker Compose Workflow
• Decision Records: TDR Template
• Technology Evaluation: Scoring Rubric

---

The PC2E Protocol

Every operation within this framework must strictly adhere to the PC2E protocol:

Predict

• Quantify confidence levels (0-100%) for all technical hypotheses.
• Proactively identify missing context before execution.
• Present architectural alternatives for limiting choices.

Communicate

• Declare core assumptions and tool selection rationale explicitly.
• Maintain continuous transparency throughout the execution lifecycle.
• Establish entry gates and pre-flight checklists for every major action.

Explain

• Maintain an auditable Chain of Reasoning: Observation → Hypothesis → Action.
• Document the rejection criteria for alternative approaches.
• Ensure all logic is reproducible and verifiable via system logs.

---

Core Operational Imperatives

All implementations are evaluated against these four non-negotiable imperatives, listed in order of priority:

1. Scalability: Architect for 10x growth without structural regression.
2. Security: Enforce a zero-vulnerability mandate across all layers.
3. Zero Technical Debt: Demand production-quality code from the initial commit.
4. Privacy & Data Minimization: Protect PII and ensure strict path isolation.

---

OWASP LLM Top 10 — Framework Compliance

This framework is itself an AI governance layer and is directly exposed to LLM-specific risks. The table below documents the mitigation status for each OWASP LLM Top 10 category.

#	Risk	Status	Mitigation
LLM01	Prompt Injection	✅ Mitigated	global/prompt-injection-defence.md — 7 defensive layers including system prompt isolation, role separation, and instruction salting
LLM02	Insecure Output Handling	✅ Mitigated	templates/output-schema.md — mode-specific output schemas with required confidence scoring
LLM03	Training Data Poisoning	➖ N/A	Framework contains no training pipelines; consuming projects must evaluate independently
LLM04	Model Denial of Service	⚠️ Partial	global/token-optimisation.md — context budget management; no explicit DoS rate-limiting at framework layer
LLM05	Supply Chain Vulnerabilities	✅ Mitigated	CODEOWNERS — owner review required on all global/ changes; .github/workflows/lint.yml — automated validation
LLM06	Sensitive Information Disclosure	✅ Mitigated	global/privacy-pdpa.md — Blind Execution Standard, PII masking regex, PDPA controls
LLM07	Insecure Plugin Design	✅ Mitigated	CODEOWNERS — mode files (loaded as plugins) require owner approval before merge
LLM08	Excessive Agency	✅ Mitigated	HITL gate in global/governance-framework.md — unconditional human approval for destructive operations
LLM09	Overreliance	✅ Mitigated	80% confidence threshold in global/pc2e-framework.md — mandatory escalation below threshold
LLM10	Model Theft	➖ N/A	Framework governs agent behaviour, not model weights; model protection is infrastructure-layer

---

Documentation

Comprehensive documentation for each layer of the framework is located in the following directories:

• Global Governance: Universal rules, anti-regression patterns, the PC2E core philosophy, and token optimisation.
• Operational Modes: Detailed Entry/Exit gates for Orchestrator, Architect, Code, Debug, and Ask modes.
• Standardized Workflows: Step-by-step procedures for Docker, CI/CD, and system migrations.
• Governance Artifacts: Standardized formats for Tech Decision Records (TDRs) and system logs.

Examples

PC2E in Action: Technical Hypothesis

Before PC2E:

"I will try to fix the database connection by restarting the service."

After PC2E (High Fidelity):

Predict: I have 85% confidence that a service restart will resolve the timeout. Communicate: I am assuming the underlying network config is stable. I will use systemctl restart now. Explain: The logs show a stale socket error (Err: 110), which typically indicates a hung process rather than a configuration failure.

Contributing

We welcome contributions to the PC2E Agent Governance Framework. To propose updates:

1. Identify a gap or inconsistency in the current Operational Modes.
2. Document the proposed change using the TDR Template.
3. Submit a Pull Request following the feat/ or fix/ branch naming conventions.

---

Support & Maintenance

• Rule Conflicts: Refer to the Precedence Logic section.
• Framework Gaps: Document detected inconsistencies and propose updates to the relevant module.
• Version History: See CHANGELOG.md for detailed change logs.

---

Release Notes

v1.0 (2026-03-19)

• Initial deployment of the PC2E-aligned governance structure.
• Consolidation of legacy rule systems into a 4-tier directory hierarchy.
• Standardization of mode-specific entry/exit gates and kebab-case naming.