Suggest Patterns
Mines tool failure history to suggest new patterns for memory.
Inputs
None. Runs analysis on existing failure data.
Workflow
1. Load Existing Patterns
- •
memory_read("learned/patterns")— existing error_patterns - •Count existing, list pattern names
2. Mine Patterns from Failures
- •Read
memory/learned/tool_failures.yaml - •Or use
scripts/pattern_miner.pyif available:mine_patterns_from_failures() - •Group similar errors, suggest when frequency ≥ 5
- •Output: pattern, frequency, recommended_category, tools, example errors
3. Build Output
For each suggestion (top 10):
- •Pattern name
- •Frequency
- •Recommended category
- •Tools affected
- •Example errors
- •To add:
skill_run("learn_pattern", '{"pattern": "...", "category": "...", "meaning": "...", "fix": "...", "commands": ["..."]}')
If no suggestions: "No new patterns to suggest! All common errors (5+) already captured."
4. Log
- •
memory_session_log("Ran pattern discovery", "Found X new patterns, existing: Y")
Key MCP Tools
- •
memory_read— learned patterns - •
memory_session_log— session logging - •File read:
memory/learned/tool_failures.yaml - •Optional:
scripts/pattern_miner.pyfor mining logic
Chaining
- •Chains to:
learn_pattern— save suggested patterns