Skip to main content
skill CX Codex

improve-system — Claude Code skill: what it does, how to install

Inspects past session logs and correction history to uncover recurring agent mistakes. Suggests targeted prompt edits and audits token consumption.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add gasserane/personal-skills --skill improve-system -a claude-code
Open on GitHub

What improve-system does

This skill reviews session histories, QA logs, and cost calibration tables to spot habitual missteps. It parses files such as qa-rejection-log.md, cost-calibration-log.md, and feedback notes inside .remember. Instead of acting on isolated slips, it requires two independent occurrences before bundling an issue into an action item.

Once patterns are clustered, the assistant compiles concrete changes for review. These include edits to CLAUDE.md, updates to individual skills, or custom PreToolUse hooks. Every proposal requires manual confirmation. The skill never updates files autonomously, presenting a numbered overview for approval instead.

The routine relies on structured tracking logs inside the agent-improvements directory. Without recorded session notes and feedback logs in the workspace, the analysis will find no data to evaluate.

What it helps with

  • identify recurring agent failure patterns across recent chats
  • monitor the QA rejection rate metric over time
  • pinpoint why specific sessions consumed too many tokens
  • turn repeated human corrections into standing rules for CLAUDE.md

Example prompts for the agent

  • /improve-system
  • Check recent sessions and show what the agent keeps getting wrong
  • Why was that session expensive? Run a token audit

How to install improve-system

1Skills CLI (Vercel)

npx skills add gasserane/personal-skills --skill improve-system -a claude-code

2Manually: put the files in place

~/.claude/skills/improve-system/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/improve-system/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about improve-system

How to install improve-system in Claude Code?

Run npx skills add gasserane/personal-skills --skill improve-system -a claude-code. The configuration is saved to ~/.claude/skills/improve-system/SKILL.md.

How does improve-system differ from /system-audit?

While /system-audit focuses on fixing one-off bugs and configuration drift, improve-system isolates recurring behavioural trends across runs. It translates frequent human corrections into permanent instructions.

Does the skill modify files automatically?

No, automatic file updates are barred. The skill drafts a numbered list of edits and pauses until you explicitly approve or reject each proposed change.

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set29

localise Adapts finished English deliverables into Romanian, French, Spanish, or Russian with glossary checks, back-translation QA, and deck reviews. CX Codex +1 620installs capture-desktop Parses pasted Claude Desktop transcripts, sorting generated artifacts, rules, and notes directly into local project files. CX Codex +1 604installs accreditation-desk-review Runs the desk-review stage of an IPPF Member Association accreditation: judge the self-assessment and evidence against each standard and check, record reviewer conclusions, and draft interview questions. CX Codex +1 557installs implementation-pack Builds a post-award project pack: applies the signed grant-agreement figures to a proposal baseline and produces files a project manager can run with. CX Codex +1 551installs learned Runs a four-question post-project review and logs verbatim takeaways into Obsidian. Useful for capturing practical lessons right after delivering work. CX Codex +1 529installs research-scoping Previews and maps research questions in one pass before running heavy workflows. Helps spot tensions and decide whether to escalate. CX Codex +1 414installs decision-debate Runs a structured debate between autonomous specialist agents on tough MEL and SRHR design choices. It helps review clashing perspectives before committing to a plan. CX Codex +1 356installs mel-discipline Makes the agent walk five stages when writing an analytical deliverable, from scoping to an honest report on what it checked. Useful when a deadline pushes you to cite a source you never opened. CX Codex +1 355installs

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor