Bỏ qua tới nội dung
TakoAPI
Quay lại kỹ năng

acc-error-memory

bởi impkind

Notes & PKM
816Lượt tải do clawskills.sh công bố
3Sao
0Lượt xem
0 lượt thíchĐăng nhập để thích kỹ năng này
—Chưa có đánh giá
Đánh giá của bạn

Tóm tắt

Error pattern tracking for AI agents.

Mô tả

Kỹ năng này làm gì

Tracks error patterns for AI agents across sessions. When a user corrects the agent, the skill logs the exchange, assigns a pattern name, and escalates severity as the same mistake recurs. Patterns clear automatically after 30 days without recurrence.

Unlike one-off corrections that an agent forgets next session, this skill persists mistake patterns in a state file so the same errors don't recur indefinitely.

Khi nào nên dùng

  • Flagging an agent that repeatedly cites outdated software versions
  • Escalating tone mismatch errors to critical after three occurrences
  • Loading known error patterns at session start to pre-warn the agent
  • Reviewing which mistakes resolved after a 30-day clean period
  • Running scheduled transcript analysis to detect new correction signals

Quy trình ví dụ

ĐẦU VÀO

User asks: why does my agent keep stating wrong Python versions?

AGENT
  1. 1

    Run encode-pipeline.sh to extract recent user and assistant exchanges since the last watermark

  2. 2

    LLM screens exchanges via haiku-screen.sh and detects user correction signals

  3. 3

    Call log-error.sh with pattern 'factual_error', context, and mitigation note

  4. 4

    Run load-state.sh to confirm pattern severity and active count

  5. 5

    Add ACC_STATE.md load step to AGENTS.md session startup so the agent sees the pattern on next launch

ĐẦU RA

Pattern 'factual_error' logged at warning severity with mitigation 'Always verify version numbers via web search'

Cài đặt

clawhub install impkind/acc-error-memory

Chạy lệnh CLI, hoặc dán prompt vào Claude Code, Cursor, Codex hay coding agent bạn đang dùng.

Liên kết