菜单
Extract durable learnings and preferences from the current conversation for memory capture. Use this after a conversation to capture what should be remembered.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Consolidate and clean up existing brain memory. Run this periodically (like a sleep cycle) to dedupe, decay, merge, and relocate memories to vault.
Remove old pi session files to reclaim disk space
Testing-focused coding workflow. Extends code-assist with enhanced test generation, coverage analysis, and test quality gates.
Systematic debugging workflow for diagnosing and fixing issues. Follows a structured approach to isolate problems, form hypotheses, and verify fixes.
Generate documentation from code or scratch - READMEs, API docs, changelogs, and user guides. Complements codebase-summary which analyzes existing code.
Track skill usage with correctness scoring, generate reports, and identify unused or misused skills for pruning.
| name | extract-memory |
| description | Extract durable learnings and preferences from the current conversation for memory capture. Use this after a conversation to capture what should be remembered. |
| kind | sop |
Extract durable learnings and user preferences from a conversation that will remain useful across future sessions. This is the extraction phase of memory management - run this after conversations to capture what matters.
When to use:
cleanup-memory (which organizes existing memories)Difference from cleanup-memory:
extract-memory = "What should I remember from this conversation?" (captures new)cleanup-memory = "Clean up my existing memories" (organizes old)Scan the conversation and classify each substantive exchange:
Extractable:
Not extractable:
Compare candidates against existing memories to avoid duplicates.
Write concise, actionable memory entries (under 200 characters).
Good: "Use printf '%s' instead of echo for piping to jq — echo adds trailing newline" Bad: "User prefers better approaches"
Assign type and category:
{
"learnings": [
{"text": "concise, actionable learning statement"}
],
"preferences": [
{"category": "Communication|Code|Tools|Workflow|General", "text": "concise preference statement"}
]
}
Good extraction (decision):
{"learnings": [{"text": "Rho tmux config at ~/.rho/tmux.conf is now the active ~/.tmux.conf"}]}
Good extraction (preference):
{"preferences": [{"category": "Communication", "text": "X post voice: understated, self-aware, dry"}]}
Correct output (nothing to extract):
{"learnings": [], "preferences": []}
After extracting memories, consider running cleanup-memory periodically to organize and prune your brain storage.