Generate a clear PR title, summary, and test plan from a diff.
Used 7×
Success rate86%
1. bash(git diff main)
2. Identify changed modules
3. Infer intent from changes
4. Write title + summary + test plan
5. Format as markdown
Data
optimize_sql_query
Analyze a slow SQL query, add indexes, and verify with EXPLAIN.
Used 5×
Success rate80%
1. Run EXPLAIN ANALYZE on query
2. Identify seq scans
3. Propose composite index
4. CREATE INDEX
5. Re-run EXPLAIN → verify speedup
Engineering
refactor_to_classBELOW THRESHOLD
Extract repeated logic from multiple files into a shared base class.
Used 3×
Success rate67%
1. search(duplicated pattern)
2. Design base class interface
3. write_file(base_class.py)
4. Update each inheriting file
5. Run tests
Engineering
generate_test_suite
Read a module and generate comprehensive tests covering happy path + edge cases.
Used 6×
Success rate83%
1. read_file(module)
2. List all public functions
3. For each: happy path + error cases
4. write_file(test_module.py)
5. bash(pytest --cov)
Min Success Rate70%
50%99%
Skills below this rate are not promoted to the library.
Total skills6
Promoted5
Below threshold1
Procedural memory stores reusable skills - learned procedures extracted from successful tasks. New tasks search the skill library first, then learn a new skill if none matches.
Procedural memory stores reusable skills - step-by-step procedures extracted from tasks the agent has successfully completed. When a new task arrives, the agent first searches its skill library for a high-similarity match. If found, it executes the stored procedure directly rather than reasoning from scratch. If no match is found, the agent solves the task from first principles and, if successful, extracts a new skill to store for future reuse. Skills are promoted to the library only if their historical success rate exceeds a configurable threshold. This visualization shows a 6-skill library across three categories and demonstrates live skill invocation and learning.
6 skills across Research, Engineering, and Data categories - each with usage count, success rate, and pseudo-code
Skill Library tab shows all stored skills with success rate bars and promotion status
Skill Invocation tab: select a new task and watch the agent find the best-matching skill
Adjustable minimum success rate threshold - lower it to include more skills, raise it to be stricter
Simulate learning: assign a novel task to trigger the agent to extract and store a new skill
Toggle pseudo-code visibility to inspect the step-by-step procedure stored in each skill
Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.
