From improve
Make a directed process change in the right place so it propagates to the running process. Handles locating the target file, making the edit, and verifying the change takes effect. Use when you know what needs to change — either from your own observation, a lint rule, or because the human pointed at the problem. Do NOT use for uncertain improvements — use improve:experiment for those.
How this skill is triggered — by the user, by Claude, or both
Slash command
/improve:fixThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You know what to change. This skill handles the mechanics: find the right file, make the edit, ensure propagation, verify effect.
You know what to change. This skill handles the mechanics: find the right file, make the edit, ensure propagation, verify effect.
Every process behavior lives in a file. Map the problem to its source:
| Behavior | Where it lives |
|---|---|
| Agent prompt/instructions | <plugin>/agents/<name>.md |
| Agent model tier | Frontmatter model: in agent definition |
| Agent tool access | Frontmatter tools: in agent definition |
| Skill instructions | <plugin>/skills/<name>/SKILL.md |
| Skill reference data | <plugin>/skills/<name>/references/<file> |
| Hook behavior | <plugin>/hooks/scripts/<script> |
| Hook registration | <plugin>/hooks/hooks.json |
| Cron schedule | Managed via /loop — not a file edit |
| Board/kanban rules | Skill instructions that reference bd commands |
| Plugin metadata | <plugin>/.claude-plugin/plugin.json |
| Project instructions | CLAUDE.md or .claude/ config files |
| Global instructions | ~/.claude/CLAUDE.md or ~/.claude/me.md |
If you have a topology map from improve:attach, use it. If not, use Grep and Glob to locate the behavior.
Read the target file first. Always. Then make the minimal change that addresses the problem.
Principles:
Not all changes take effect immediately. Know the propagation model:
| Change type | Propagation | Action needed |
|---|---|---|
| Agent definition edit | Next agent spawn picks it up | None — but running agents won't see it |
| Skill SKILL.md edit | Next skill invocation picks it up | None |
| Skill reference edit | Next skill invocation that reads the reference | None |
| Hook script edit | Next hook event fires it | Ensure script is executable (chmod +x) |
| hooks.json edit | Requires plugin reinstall | Run reinstall-plugin.sh <plugin> in separate terminal, then /reload-plugins |
| plugin.json edit | Requires plugin reinstall | Same as above |
| CLAUDE.md edit | Immediate in current session | None |
Important: If the change requires a plugin reinstall, tell the human. You cannot run the reinstall from within a Claude Code session (CLAUDECODE env var blocks nested CLI). The human runs it in a separate terminal.
After the change, confirm it will take effect:
After a successful fix, consider:
improve:lint to record it as a watch ruleOffers UI/UX design guidance for web and mobile with 50+ styles, 161 color palettes, 57 font pairings, and 99 UX guidelines across 10 stacks. Use for designing pages, components, color systems, or reviewing UI code.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
npx claudepluginhub cosmicdreams/claude-plugins --plugin improve