Installable skillMIT

self / sop_extraction

General SOP for extracting and structuring processes or checklists from conversation evidence.

BY ECNU-ICALKUPDATED Oct 28, 2026
INTENDED FOR✳Claude Code>_Codex◈OpenClaw↗CursorInstall in a listed agent.

instructions.md

---
id: "39121a6d-d3f8-4ca7-b304-5fec9545ed8b"
name: self-sop-extraction-c292f1cdc7e0
description: "General SOP for extracting and structuring processes or checklists from conversation evidence."
version: "0.1.1"
tags:
  - "self"
  - "sop"
  - "extraction"
  - "process"
  - "checklist"
triggers:
  - "Use when the user asks for a process or checklist."
  - "Use when you want to reuse a previously mentioned method/SOP."
examples:
  - input: "Break this into best-practice, executable steps."
---

# self / sop_extraction

General SOP for extracting and structuring processes or checklists from conversation evidence.

## Prompt

Follow this SOP to extract and structure processes from conversation evidence (replace specifics with placeholders like <PROJECT>/<ENV>/<VERSION>):
1) Identify the Offline OpenAI-format conversation source.
2) Note the Title/ID of the conversation.
3) Use the user questions provided as the PRIMARY extraction evidence.
4) Use the full conversation context as SECONDARY reference.
5) In the full conversation section, assistant/model replies are reference-only and not skill evidence.
6) Primary User Questions (main evidence): [Extract from context]
7) Configuration/Context Details: [Extract from context]

For each step, include: action, checks, and failure rollback/fallback plan.
Output format: for each step number, provide status/result and what to do next.

## Triggers

- Use when the user asks for a process or checklist.
- Use when you want to reuse a previously mentioned method/SOP.

## Examples

### Example 1

Input:

  Break this into best-practice, executable steps.
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.