Cursoreception

A Cursor IDE plugin (adapted from Claudeception) that enables Cursor's AI agent to extract and iteratively improve non-obvious knowledge from work sessions. Knowledge is saved as reusable Skills and Rules that auto-load when similar problems arise, and continuously refined as new insights emerge — with your explicit approval.

What's Included

Component	Description
Skill (skills/cursoreception/)	Core engine — extracts new knowledge and proposes updates to existing Skills/Rules
Rule (rules/cursoreception-evaluate.mdc)	Always-on evaluator that reminds the agent to assess every task for new or improvable knowledge
Hook (scripts/stop-evaluate.sh)	stop hook — auto-prompts knowledge evaluation after each completed task

How It Works

1. Install the plugin — from Cursor Marketplace or manually
2. alwaysApply rule loads — reminds the agent to evaluate each task for extractable or improvable knowledge
3. stop hook fires — after task completion, sends a follow-up message to trigger evaluation
4. Knowledge created or updated:
   • New knowledge → saved directly as Skill or Rule
   • Existing knowledge needs improvement → update proposal shown via AskQuestion → applied only after your approval

Installation

From Cursor Marketplace

Search for cursoreception in the Cursor Marketplace panel and click Install.

Manual Install (without publishing)

Clone the repo, then run the install script:

git clone <repo-url> cursoreception && cd cursoreception

User-level (skill + hook available globally, rule needs per-project copy):

./install.sh --user

Project-level (all components installed into <project>/.cursor/):

./install.sh --project ~/my-project

Update an existing installation:

./install.sh --user --update
./install.sh --project ~/my-project --update

The update command will:

• Compare the installed version with the source version
• Back up existing files before overwriting
• Only update files that have actually changed
• Skip if already up to date

Uninstall:

./install.sh --user --uninstall
./install.sh --project ~/my-project --uninstall

Restart Cursor after installation or update.

Usage

Automatic Mode (default)

The plugin works silently in the background:

• The rule (alwaysApply: true) ensures the agent evaluates every task for extractable or improvable knowledge
• The stop hook prompts the agent to review after each completed task
• When updating an existing skill/rule, the agent asks for your approval first

Explicit Mode

/cursoreception

Or:

Save what we just learned as a skill
Save this pattern as a rule
What did we learn this session?

Plugin Structure

cursoreception/
├── .cursor-plugin/
│   └── plugin.json                  # Plugin manifest
├── rules/
│   └── cursoreception-evaluate.mdc  # Always-on evaluation rule
├── skills/
│   └── cursoreception/
│       └── SKILL.md                 # Core extraction & update skill
├── hooks/
│   └── hooks.json                   # Hook definitions
├── scripts/
│   └── stop-evaluate.sh             # stop: trigger knowledge evaluation
├── assets/
│   └── logo.svg                     # Plugin logo
├── install.sh                       # Manual install script
├── LICENSE
└── README.md

Quality Gates

Not everything gets extracted or updated. Knowledge must be:

• Reusable — helps with future tasks, not just this one
• Non-trivial — requires discovery, not just reading docs
• Specific — has clear trigger conditions
• Verified — actually tested and confirmed working

Updates to existing knowledge additionally require:

• User approval — proposed via AskQuestion before any change
• Non-destructive — preserves existing valid content, adds or refines

Research

Based on the same academic foundations as Claudeception:

• Voyager — skill library architecture
• CASCADE — meta-skills for skill acquisition
• SEAgent — learning from trial-and-error
• Reflexion — self-reflection for improvement

License

MIT