wallacedobbs428/proactive-agent

Proactive Agent v3.0.0

Framework that transforms AI agents from reactive task-followers into anticipatory partners.

Core Purpose

The mindset shift: "What would genuinely delight my human that they haven't thought to ask for?"

Three Core Pillars

Proactive

The agent asks "what would help?" rather than waiting for requests, surfaces reverse-prompted ideas, and monitors outcomes.

Persistent

Using Write-Ahead Logging (WAL), Working Buffer protocols, and Compaction Recovery, the agent survives context truncation by capturing critical details before responding.

Self-Improving

The agent self-heals issues, attempts 10 different approaches before requesting help, and evolves safely using ADL/VFM guardrails that prioritize stability over novelty.

Key Mechanisms

WAL Protocol

Captures corrections, decisions, proper nouns, preferences, and specific values immediately in SESSION-STATE.md — treating chat history as temporary buffer, not permanent storage.

Working Buffer

Logs every exchange after 60% context consumption, surviving the "danger zone" between memory flush and compaction.

Relentless Resourcefulness

Non-negotiable commitment to attempting 5-10 methods before considering asking for help.

Security & Governance

• Hardened policies against external skill installation vulnerabilities
• Warns against AI agent social networks as "context harvesting attack surfaces"
• Prevents context leakage through shared channels