From ai-conversation-intelligence
Analyze AI conversation history to discover corrections, repeated workflows, validated approaches, and recurring instructions that should become skills, agents, hooks, or memory entries
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-conversation-intelligence:mine-patternsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Analyze conversation history from the ai-chat-browser MCP to discover automatable patterns.
Analyze conversation history from the ai-chat-browser MCP to discover automatable patterns.
Process specific sessions queued by the SessionEnd hook.
Scan all indexed sessions or a filtered subset.
Query the ai-chat-browser MCP search_messages tool for high-signal messages.
Correction signals (user correcting the assistant):
"no don't" OR "wrong" OR "stop" OR "that's not" OR "I said"Confirmation signals (user validating an approach):
"perfect" OR "exactly" OR "yes that's right" OR "keep doing"Instruction signals (user giving persistent directives):
"always" OR "never" OR "make sure" OR "from now on" OR "remember to"For each query, use limit: 50 and context_messages: 3.
For each hit from Step 1, use get_conversation with the search parameter to retrieve the matching message with ±3 messages of surrounding context.
For each high-signal hit with context, classify the pattern:
| Pattern Type | Signal | Proposed Output |
|---|---|---|
| correction | User says "no/don't/wrong" then provides right way | feedback memory or hook rule |
| repeated_workflow | Same tool call sequence across 3+ sessions | Skill candidate |
| repeated_delegation | Same autonomous task requested 3+ times | Agent candidate |
| validated_approach | User confirms with positive signals | feedback memory |
| recurring_instruction | Same instruction in 3+ sessions | Rule or CLAUDE.md update |
Assign 0-100 confidence based on:
Before creating a nomination:
Keyword pre-filter: Use Grep to search existing files:
~/.claude/projects/<project>/memory/knowledge/ (existing memories)~/.agents/rules/ (existing rules)~/.agents/skills/ (existing skills)~/Projects/patryks-treadmill-claude-plugins/plugins/*/skills/ (treadmill skills)LLM similarity check: If keyword matches found, evaluate:
Auto-merge (confidence 80+, memories and rules only):
knowledge/ or ~/.agents/rules//review-nominationsNominate (confidence 50+, or any skill/agent/hook):
~/.claude/projects/<project>/memory/nominations/---
id: candidate_<8-char-hash>
confidence: <score>
type: feedback | skill | agent | hook | rule | claude_md
target: new | update:<existing-file-path>
source_sessions:
- <session-id-1>
- <session-id-2>
discovered_at: <ISO-8601>
pattern_type: correction | repeated_workflow | repeated_delegation | validated_approach | recurring_instruction
auto_merged: false
---
# <Title>
## Evidence
<Relevant message excerpts from source sessions>
## Proposed Content
<The actual memory/skill/hook content to create or diff to apply>
Discard (confidence < 50): Skip silently.
After processing, summarize:
Suggest running /review-nominations if there are pending items.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub patrykkopycinski/patryks-treadmill-claude-plugins --plugin ai-conversation-intelligence