single-agent-lightweight-growthlisted

# Skill: Single-Agent Lightweight Growth ## Purpose Use this workflow during normal single-agent execution so that the agent can keep improving without enabling the full multi-agent scheduling system. This is the default growth mode. ## Trigger Use for all normal tasks unless multi-agent long-task scheduling is explicitly enabled. ## Principle Growth analysis may be continuous, but durable growth writes are permission-bound. Most small tasks should not produce files, long reviews, or heavy process. ## Authority Levels - `L1 Analyze`: identify a real signal and form an in-memory candidate. Do not create or modify a durable project, global, canonical, Skill, checklist, lint, adapter, tool-install, or public record. - `L2 Project Maintain`: only after a one-time project authorization that names the writable project surface. Record the triggering event, evidence, and authorization reference. The user may switch back to `analysis_only` or revoke the authorization. - `L3 System Promote`: proposing or changing `GLOBAL_MEMORY`, global/canonical rules, Skills, checklists, lint, adapters, installed tools, or public content always requires a separate confirmation. L2 never implies L3. ## Workflow 1. Execute the user's task normally. 2. Verify before claiming completion. 3. At the end, briefly check whether the task produced a high-signal event: user correction, verification reversal, repeated failure, rework, recovery/rollback, a materially successful method, or a tool-fact/a