self-improve

The self-improve skill functions as a meta-learning mechanism for OpenCode AI agents, enabling the system to evolve its own operational knowledge. By monitoring execution successes and failures, this skill allows the agent to update its Markdown documentation, refine prompts, and adjust tool logic in real-time. It is designed for developers who want their agents to become increasingly autonomous and better aligned with the repository's core principles, such as self-bootstrapping, graceful degradation, and self-fixing.

Key features include:

• Automated updates of .opencode/skill/ documentation when API behaviors or CLI outputs deviate from expected norms.
• Reinforcement of successful workflows into permanent, reusable skills and tools.
• Identification of recurring tasks that can be abstracted into specialized agents or event-driven plugins.
• Continuous repair and optimization of the workspace state following debug commands or bootstrap sessions.
• Documentation of credentials, gotchas, and environment-specific configuration to improve future bootstrapping and setup.

Usage and operational details:

• The skill is invoked when an agent detects a discrepancy, a failed task, or a new successful workflow pattern in the 0 Finance CLI and development environment.
• Expected inputs include the context of the recent execution, error logs, or user confirmation of a successful process. The agent will output updated Markdown files (SKILL.md), suggest new tool implementations (.ts), or propose agent prompt refinements.
• Practical constraints include ensuring that automated updates do not override critical manual security configurations or sensitive credential files. It is strictly scoped to the .opencode directory to prevent accidental modification of core financial logic or smart contracts within the 0 Finance monorepo.
• It is highly recommended to use this skill in conjunction with @debug-workspace and @bootstrap commands to maintain the repository's self-aware state.