Guide · Agent skill
How to use the Continuous Learning Skill for Claude Code
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Continuous Learning Skill (skill name: continuous-learning) is an agent skill for Claude Code that reviews each finished coding session and saves reusable patterns, such as error fixes and project conventions, as new learned skills. It is aimed at developers who use Claude Code daily and want recurring fixes captured automatically. After this guide you will know how to install the skill, configure it, trigger it, and review what it saves.
Continuous Learning Skillaffaan-m/ECC · MIT Download the skill Quick start
- 1Check requirements. Confirm you have Claude Code with hooks support and a bash-capable environment to run evaluate-session.sh.
- 2Download and unzip. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 3Add the Stop hook. Add the Stop hook entry to ~/.claude/settings.json so it runs evaluate-session.sh when a session ends.
- 4Review config.json. Adjust min_session_length, extraction_threshold and the pattern lists, keeping auto_approve set to false.
- 5Work normally. Use Claude Code as usual. The hook evaluates each session when it ends.
- 6Review saved patterns. Read the files in ~/.claude/skills/learned/ and delete any that are incorrect or too narrow.
What is an agent skill, and what does this one do?
An agent skill is a folder of instructions and, in some cases, scripts that an AI agent can load to handle a specific kind of task. This skill is delivered as a SKILL.md file and is used with Claude Code.
When it is set up, the skill runs as a Stop hook, which means it runs once when a Claude Code session ends. It checks whether the session had enough messages (the default is 10 or more), looks for patterns worth keeping, and saves useful ones to ~/.claude/skills/learned/.
- error_resolution: how a specific error was fixed
- user_corrections: patterns taken from corrections you made
- workarounds: solutions to quirks of a framework or library
- debugging_techniques: approaches that worked when debugging
- project_specific: conventions particular to your project
Install the Continuous Learning Skill
Download the ZIP for this skill from this page and unzip it. Place the resulting folder in ~/.claude/skills/ to make it available for all your projects, or in a project's .claude/skills/ folder to limit it to that project.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Turn on the Stop hook in settings.json
The skill only runs automatically once a Stop hook is added. Open ~/.claude/settings.json and add the following entry. It runs evaluate-session.sh from the skill folder whenever a session ends.
This requires Claude Code with hooks support and a bash-capable environment to run evaluate-session.sh.
{
"hooks": {
"Stop": [{
"matcher": "*",
"hooks": [{
"type": "command",
"command": "~/.claude/skills/continuous-learning/evaluate-session.sh"
}]
}]
}
}Customise thresholds and pattern types
Edit config.json to change how the skill behaves. The main settings are the minimum session length, the extraction threshold, which pattern types to detect, and which patterns to ignore. Setting auto_approve to false, as the default example does, is the cautious choice because you can review what gets saved.
The default example is shown below. Keep it as a starting point and adjust one setting at a time so you can see what changes.
- min_session_length: default 10 messages
- extraction_threshold: default medium
- auto_approve: default false
- learned_skills_path: default ~/.claude/skills/learned/
- ignore_patterns: simple_typos, one_time_fixes, external_api_issues
{
"min_session_length": 10,
"extraction_threshold": "medium",
"auto_approve": false,
"learned_skills_path": "~/.claude/skills/learned/",
"patterns_to_detect": ["error_resolution", "user_corrections", "workarounds", "debugging_techniques", "project_specific"],
"ignore_patterns": ["simple_typos", "one_time_fixes", "external_api_issues"]
}How to trigger the skill with natural requests
The automatic run happens at the end of each session through the Stop hook, so you do not need to phrase anything special for it. The source also describes a /learn command that extracts patterns manually during a session.
The source does not list other trigger phrases, so do not rely on specific wording. Use /learn when you want to capture a pattern immediately, and let the Stop hook handle the rest.
- Automatic: finish a session normally and the hook evaluates it
- Manual: run /learn during a session to extract patterns now
Review and manage what the skill saves
Saved patterns go into the learned folder as skills you can read. Because auto-extraction can save low-quality or incorrect patterns, read each new file before relying on it, and delete any that are wrong or too specific.
The source notes that this approach is less reliable than a hook-based design for observation. Treat the output as suggestions to check, not as verified guidance.
- Open ~/.claude/skills/learned/ regularly and review new files
- Delete patterns that are one-off fixes or incorrect
- Keep auto_approve set to false while you are learning how it behaves
Safety notes: scripts, network and private data
This skill runs scripts, specifically evaluate-session.sh, so you should read the script before enabling the hook. The source states that it does not use the network.
Session transcripts are processed and saved. If secrets, credentials or private code appear in a conversation, they may end up in a saved skill. Do not use this skill on sensitive or confidential projects unless you have reviewed the output carefully.
- Runs scripts: yes (evaluate-session.sh)
- Uses network: no
- Writes new files into your ~/.claude folder
What people use it for
- Capturing a fix for a specific error you hit repeatedly
- Recording a debugging approach that worked on a tricky bug
- Saving project conventions so future sessions follow them
- Noting workarounds for quirks in a framework or library
- Building up project-specific guidance across many sessions
Frequently asked questions
How do I install the continuous-learning skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills. Then add the Stop hook to ~/.claude/settings.json.
What do I need to run it?
The source lists Claude Code with hooks support, a bash-capable environment to run evaluate-session.sh, and a ~/.claude/settings.json file where you add the Stop hook.
Is the continuous-learning skill free?
The source identifies the licence as MIT. It does not mention pricing, so check the original repository for any terms beyond the licence.
Does it send my code over the internet?
The source says the skill does not use the network. It does process and save session transcripts locally, so sensitive content can end up in saved files.
What are the alternatives to this skill?
The source compares it with Homunculus v2, which uses PreToolUse and PostToolUse hooks for observation, background analysis, and confidence scoring. It describes that design as more refined, but this skill is simpler to set up.
Why are some patterns not saved, or saved incorrectly?
The skill only saves patterns it judges useful, and it can save low-quality or incorrect ones. Short sessions below min_session_length are skipped, and ignore_patterns filters out typos, one-time fixes and external API issues. Review the learned folder and adjust config.json.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers who use Claude Code every day and want recurring fixes and project conventions captured with little effort, provided they review what gets saved. Skip it for confidential projects, for anyone unwilling to let an agent write files into ~/.claude, or if you need reliable, guaranteed triggering.
Download the skill