Guide · Agent skill
How to use the Continuous Learning v1 (obsoleto) agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Continuous Learning v1 is a legacy agent skill for Claude Code. It was designed to review a finished coding session and save reusable patterns, such as error fixes or debugging techniques, as learned skills. The source file marks it as deprecated as of 2026-04-28, so this guide explains what it does, how it was configured, and why most people should use continuous-learning-v2 instead.
Continuous Learning v1 (obsoleto)affaan-m/ECC · MIT Download the skill Quick start
- 1Check your setup. Confirm you have Claude Code with hooks support. If you are new to this, consider continuous-learning-v2 instead, since it is the maintained path.
- 2Download and unzip. Download the ZIP from this page and unzip it into ~/.claude/skills/ or your project's .claude/skills/ folder.
- 3Check the script. Open evaluate-session.sh and confirm it exists, since it is not included in the provided files.
- 4Add the Stop hook. Add the hook entry to ~/.claude/settings.json pointing to ~/.claude/skills/continuous-learning/evaluate-session.sh.
- 5Review config.json. Set min_session_length, extraction_threshold, auto_approve and learned_skills_path to match how you want extraction to work.
- 6Review learned skills. After sessions end, check ~/.claude/skills/learned/ and keep only the patterns that are accurate and useful.
What this skill is supposed to make the agent do
The skill runs as a Stop hook at the end of each Claude Code session. It checks whether the session is long enough (the default minimum is 10 messages), identifies patterns that could be extracted, and saves useful ones to ~/.claude/skills/learned/.
The patterns it looks for are error_resolution, user_corrections, workarounds, debugging_techniques and project_specific.
- Session evaluation: checks whether the session has enough messages.
- Pattern detection: identifies extractable patterns in the session.
- Skill extraction: saves useful patterns to ~/.claude/skills/learned/.
Understand the v1 status before installing
The source states that v1 is obsolete and that continuous-learning-v2 is the preferred path for new installations. v1 is kept for backward compatibility with existing setups.
The source also says v1 relies on a skill to observe sessions, and its evaluation is probabilistic. v2 uses hooks for observation, which the source describes as more reliable. If you are starting fresh, look for continuous-learning-v2 rather than this skill.
Install the skill files
Download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder to scope it to one project.
You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills (upload the ZIP)
Connect the Stop hook in settings.json
The skill only runs if you add a hook entry to your ~/.claude/settings.json file. The entry points to evaluate-session.sh inside the skill folder. The source notes that this script is not included in the provided files, so confirm it exists in your installed folder before adding the entry.
Config values such as min_session_length, extraction_threshold and learned_skills_path are set in config.json, according to the source.
{
"hooks": {
"Stop": [{
"matcher": "*",
"hooks": [{
"type": "command",
"command": "~/.claude/skills/continuous-learning/evaluate-session.sh"
}]
}]
}
}Trigger it with natural requests
The source lists these requests as suitable for this skill: configuring automatic pattern extraction, configuring the Stop hook, reviewing or curating learned skills in ~/.claude/skills/learned/, adjusting extraction thresholds or pattern categories, and comparing v1 with v2.
Because the skill is deprecated, the source says not to invoke v1 for new learning requests. Requests about continuous learning, session learning and pattern extraction should be directed to continuous-learning-v2.
- Set up the Stop hook so my Claude Code sessions are evaluated when they end.
- Review the learned skills in ~/.claude/skills/learned/ and tell me which ones to keep.
- Compare the v1 Stop hook approach with v2 for my team.
Adjust the configuration
The source gives a sample config.json. The main settings are the minimum session length, the extraction threshold, whether to auto-approve extractions, the output folder, and which pattern types to detect or ignore.
The sample sets auto_approve to false, meaning the source's default leaves approval off.
- min_session_length: 10 (minimum messages before evaluation)
- extraction_threshold: medium
- auto_approve: false
- 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"]
}Safety notes: scripts and network
This skill runs scripts: the Stop hook executes a shell command on every session end. Read evaluate-session.sh before you add the hook, since the script is not included in the provided files and you should know exactly what runs on your machine.
The source says the skill does not use the network. It also says the hook evaluation is probabilistic, so do not rely on it to capture every useful pattern. Review learned skills before reusing them, because they are saved automatically from past sessions.
- Runs scripts: yes (evaluate-session.sh via the Stop hook).
- Uses network: no, according to the source.
- Requires Claude Code with hooks support.
What people use it for
- Keeping an existing v1 setup working while you plan a migration to v2.
- Saving recurring error fixes from a long debugging session as learned skills.
- Capturing project-specific conventions that come up repeatedly in your sessions.
- Comparing the v1 hook-based approach with the instinct-based v2 before choosing one for a team.
Frequently asked questions
Is the continuous-learning v1 skill deprecated?
Yes. The source file states that it was obsolete as of 2026-04-28 and that continuous-learning-v2 should be used instead. v1 is kept for backward compatibility with existing installations.
How do I install the continuous-learning skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills. Then add the Stop hook to settings.json.
What does the Stop hook do?
It runs at the end of each Claude Code session. It checks the session length, identifies reusable patterns, and saves useful ones to ~/.claude/skills/learned/.
Is continuous-learning v1 free and what is its licence?
The source lists the licence as MIT. The source does not mention pricing, so check the repository for any cost details.
What is the difference between v1 and v2?
According to the source, v2 moves observation from a Stop hook to PreToolUse/PostToolUse hooks, turns whole skills into atomic instincts with confidence scores, and uses project-scoped storage with global promotion. The source describes v2 as a strict superset of v1.
Why is my continuous-learning hook not capturing patterns?
The source says v1 relies on a skill to observe sessions, and its evaluation is probabilistic, so it may not catch every pattern. Also confirm that evaluate-session.sh exists and that the path in settings.json is correct, since the script is not included in the provided files.
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
Use this skill only if you already run v1 and need it to keep working, or if you are comparing the old hook-based design with v2. New users should choose continuous-learning-v2, which the source describes as the preferred and maintained path.
Download the skill