self-improve_skill

This skill updates its own documentation and extends OpenCode capabilities based on learnings to improve accuracy and future proofing.

TypeScript

186

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill different-ai/agent-bank --skill self-improve

SKILL.md18.2 KB

Overview

This skill enables OpenCode to update its own documentation and extend capabilities automatically based on operational learnings. It captures failures, new workflows, and optimizations and converts them into concrete extensions (skills, agents, plugins, tools, or MCP entries). The goal is a self-bootstrapping, low-friction feedback loop that keeps the workspace reliable and reproducible.

How this skill works

OpenCode watches for triggers like failing commands, mismatched API behavior, repeated tasks, or user confirmation of working fixes. When triggered, it creates or updates the appropriate artifact (.opencode/skill, .opencode/agent, .opencode/plugin, .opencode/tool, or opencode.json) with tested, copy/paste-ready content. It also records gotchas, token-saving tips, and first-time setup steps to prevent regressions.

When to use it

An API or CLI command behaves differently than documented
You repeat a multi-step workflow more than once
A setup or bootstrap step is missing or unreliable
You discover a safer or more efficient way to do something
A bug or broken state is detected and needs a documented repair

Best practices

Update docs immediately after confirming a fix — test before committing
Keep commands copy/paste-ready and show exact file paths used by the project
Prefer small, focused artifacts: one skill per workflow, one tool per external API
Add Common Gotchas and Token Saving Tips to every skill to reduce wasted runs
Use agents for specialized model behavior and plugins for event-driven safeguards

Example use cases

Fix an incorrect curl flag in a skill and push the corrected SKILL.md for future runs
Wrap a repeated database query as a tool for safe, consistent access
Create a subagent with a constrained toolset for code reviews to avoid destructive edits
Add a plugin to notify the user when long tasks complete or to require confirmation before destructive commands
Add an MCP entry to integrate a headless browser when automation becomes frequent

FAQ

Update immediately once you confirm the fix works in the current environment; include exact commands and any prerequisites.

When do I choose an agent vs a tool?

Choose an agent when you need a distinct prompt, model, or restricted toolset. Choose a tool when you need a callable function that performs a concrete operation or API call.