π-PAI v4.0 — Personal AI Infrastructure for Pi

A Pi Coding Agent extension implementing Daniel Miessler's PAI framework (9.6K ⭐), synced with PAI v4.0.3. Includes the Ralph Wiggum iteration technique, damage control, and 5 features ported from Miessler's full system.

Install

pi install npm:pi-pai

What's New in v4.0

Synced with Miessler's PAI v4.0.3

| Feature | Source | Implementation | |---------|--------|---------------| | v4 Algorithm | PAI v4.0.3 | OBSERVE → PLAN → DECIDE → EXECUTE → VERIFY (was 7-phase, now 5) | | Sentiment tracking | PAI Observability | Every rating gets sentiment (positive/neutral/negative) + trend analysis | | Agent personas | PAI 14 agents | 7 personas: architect, engineer, pentester, designer, reviewer, researcher, qa | | Self-evolution | PAI self-upgrade | Detects repeating learning patterns (3+ occurrences), triggers /pai evolve | | Plans convention | PAI Plans dir | Auto-creates .pi/plans/, lists plans via /pai plans |

Commands

/pai — Goal-Driven Algorithm

# Setup
/pai mission Build a profitable trading system
/pai goal Deploy live stat-arb strategy
/pai challenge Overfitting risk on historical data

# Run the v4 algorithm loop
/pai loop Deploy live strategy    # Start: OBSERVE
/pai isc Strategy achieves Sharpe >1.5 on 5-year backtest
/pai next Observed: spread mean-reverts at 3.2 std
/pai next Plan: backtest 2020-2025, then paper trade 2 weeks
/pai next Decision: go with mean-reversion, tight stops
/pai next Executed: deployed to paper trading
/pai next Verified: 58% win rate, Sharpe 1.8

# Templates
/pai template trading    # Pre-built mission + goals + challenges
/pai template saas|devops|research|agent

# Agent personas (NEW in v4)
/pai agent architect Design the auth system for a multi-tenant SaaS
/pai agent pentester Review this API for security vulnerabilities
/pai agent designer Create the onboarding flow for mobile
/pai agent reviewer Review the feature branch against main

# Sentiment & trends (NEW in v4)
/pai trend              # Rating trend: avg, recent, sentiment distribution
/pai evolve             # Self-evolution: repeating pattern report

# Other
/pai plans              # List .pi/plans/ directory
/pai status             # Full status with all v4 features
/pai learn <insight>    # Record a learning
/pai done g0            # Complete a goal
/pai block g1           # Block a goal
/pai reset              # Clear everything

/rate — Sentiment-Aware Ratings

/rate 9 Clean architecture, fast execution  # → ⭐9 😊 positive
/rate 3 Missed edge cases, slow             # → ⭐3 😞 negative → auto-captures learning
/rate 6                                     # → ⭐6 😐 neutral

Ratings track: score (1-10), context, timestamp, sentiment (inferred from score + keywords).

Widget shows trend: ⭐7.2 📈8.0 (15 ratings) — improving/declining/stable.

/ralph — Deterministic Iteration

/ralph Build a REST API with auth, tests, and docs
# Agent iterates up to 50 times until RALPH_DONE
/ralph stop

🛡️ Damage Control

Guards via YAML rules (damage-control-rules.yaml):

• Blocked patterns: rm -rf, git reset --hard, git push --force, DROP TABLE
• Confirm-first: git push --delete, git branch -D
• Zero-access: .env, ~/.ssh/, ~/.aws/, *.pem
• Read-only/No-delete: configurable per project

🔧 Agent Tools

| Tool | Description | |------|-------------| | pai_status | Full status with v4 algorithm, trends, personas, patterns | | pai_learn | Record insight with sentiment | | pai_rate | Rate 1-10 with sentiment + trend tracking |

📊 Live Widget

  🎯 Build a profitable trading system
  Goals: 2⚡ 0🚫 1✓ │ 5 learnings │ ⭐7.2 📈8.0 (15)
  Loop: ● ● ◉ ○ ○ [DECIDE] standard 42s
  ⚠️ 2 repeating pattern(s) — consider /pai evolve

Agent Personas

| Persona | Focus | |---------|-------| | architect | System design, scalability, API design, failure modes | | engineer | Production code, error handling, types, tests | | pentester | Security: injection, auth bypass, SSRF, secrets, deps | | designer | UX/UI, accessibility, responsive, interaction patterns | | reviewer | Code review: correctness, edge cases, P1/P2/P3 findings | | researcher | Deep investigation, evidence-based, source citations | | qa | Test planning: happy paths, edge cases, boundaries, regression |

Lineage

| Project | Author | What | |---------|--------|------| | Personal AI Infrastructure | Daniel Miessler | PAI v4.0.3 framework (9.6K ⭐) | | pi-ralph | Whamp | Ralph Wiggum iteration technique | | pi-vs-claude-code | disler | Damage control + extension patterns | | Pi Coding Agent | Mario Zechner | The platform |

v3 → v4 Migration

The algorithm changed from 7 phases to 5:

• Old: OBSERVE → THINK → PLAN → DEFINE → EXECUTE → MEASURE → LEARN
• New: OBSERVE → PLAN → DECIDE → EXECUTE → VERIFY

THINK+PLAN merged into PLAN. DEFINE became DECIDE. MEASURE+LEARN merged into VERIFY.

Active loops will reset on upgrade. All other state (goals, learnings, ratings) carries forward.