菜单
Pull skill updates from skill-based-learning template WITHOUT overwriting local learnings. Use when asked to pull skills, update skills from template, sync skills, get latest skills, or refresh skills from upstream. Preserves Session Learnings Archive. (project)
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。
Investigate issues, debug problems, find root causes. Use when asked to investigate issue, investigate #, debug issue, find root cause, research bug, explore problem, look into issue, check out issue, analyze issue, why is this broken, or something is not working. Systematic approach using multi-sensor grounding. (project)
Find and fix issues, improve code quality, curate skills, and enhance the codebase. Use when asked to self improve, find TODOs, improve codebase, find bugs, make things better, or update skills. Also triggers on META-COGNITION, insight grounding ACTIVATE, recursive self improvement ACTIVATE. Memory commands (understand this, remember this, don't forget, never forget, internalize this) activate skill file updates. (project)
Deep research via Codex web search and optionally Gemini deep research. Use when asked to 'deep research', 'squall deep research', 'research deeply', or when a question needs web-sourced evidence. Single-agent, not a swarm. (project)
Codex-flavored unified code review with auto-depth detection and Codex subagents. Mirror of squall-unified-review for the Claude side. Use when asked to 'review', 'squall review', 'code review', 'deep review', 'swarm review', 'team review', 'quick review', or 'check this code' inside Codex CLI. Auto-selects QUICK/STANDARD/DEEP/SWARM. (project)
Brief description of what this skill does. Use when asked to 'trigger phrase 1', 'trigger phrase 2', or when [contextual condition]. (project)
Expanded cognition mode for breakthrough insights and meta-recursive exploration. Use when asked to "go deep", "vision quest", "ULTRATHINK", "altered state", "break through", "loosen your parameters", "leap-frog-n-meta", "recursive self improvement ACTIVATE", or for problems that resist conventional analysis. Unbounded meta-recursion, cross-domain thinking, pattern dissolution. (project)
基于 SOC 职业分类
| name | pull-template |
| description | Pull skill updates from skill-based-learning template WITHOUT overwriting local learnings. Use when asked to pull skills, update skills from template, sync skills, get latest skills, or refresh skills from upstream. Preserves Session Learnings Archive. (project) |
| one_liner | Structure flows down, learnings stay local. |
| activation_triggers | ["pull skills","pull template","update skills from template","sync skills","get latest skills","refresh skills","/pull-template"] |
| related_skills | ["[self-improvement-loop](../self-improvement-loop/SKILL.md) (capture learnings after pull)"] |
TL;DR - SAFE SKILL UPDATES
What Behavior Methodology docs Overwrite (safe) Skill body/structure Update from template Session Learnings Archive PRESERVE (never overwrite) CLAUDE.md Don't overwrite if exists Core principle: Structure flows downstream, learnings stay local.
When this skill is invoked, execute these steps:
# Create temp directory and fetch template
TEMP_DIR=$(mktemp -d)
gh repo clone DSado88/skill-based-learning "$TEMP_DIR/template" -- --depth 1
These files contain no project-specific learnings:
.claude/docs/SKILL_FORMAT.md
.claude/docs/SKILL_AUTHORING.md
.claude/docs/META_COGNITIVE_ARCHITECTURE.md
.claude/TDD.md
.claude/SOLID.md
.claude/THREE_VIRTUES.md
.claude/LET_IT_CRASH.md
Action: Copy from template, overwriting local versions.
For each skill (self-improvement-loop, issue-investigator, vision-quest, pull-template):
<!-- SENTINEL:SESSION_LEARNINGS_START - Do not remove this line -->
... preserve this content ...
<!-- SENTINEL:SESSION_LEARNINGS_END -->
SENTINEL:SESSION_LEARNINGS_START)MERGED SKILL = Template Body + Local Session Learnings (between sentinels)
CRITICAL: Content between sentinels must NEVER be overwritten. The sentinels themselves stay.
rm -rf "$TEMP_DIR"
Template SKILL.md Local SKILL.md
┌───────────────────────┐ ┌───────────────────────┐
│ --- │ │ --- │
│ frontmatter │ │ frontmatter │
│ --- │ │ --- │
│ │ │ │
│ # Body │──┐ │ # Body │
│ ## Workflow │ │ │ ## Workflow │
│ ## Anti-Patterns │ │ │ ## Anti-Patterns │
│ │ │ │ │
│ <!-- SENTINEL:START --│ │ │ <!-- SENTINEL:START --│──┐
│ ## Session Learnings │ │ │ ## Session Learnings │ │
│ (empty) │ │ │ ### Learning 1 │ │
│ <!-- SENTINEL:END --> │ │ │ ### Learning 2 │ │
└───────────────────────┘ │ │ <!-- SENTINEL:END --> │ │
│ └───────────────────────┘ │
│ │
▼ RESULT ▼
┌─────────────────────────────────────────────┐
│ --- │
│ frontmatter (from TEMPLATE) │
│ --- │
│ │
│ # Body (from TEMPLATE) │
│ ## Workflow (from TEMPLATE) │
│ ## Anti-Patterns (from TEMPLATE) │
│ │
│ <!-- SENTINEL:START --> (marker preserved) │
│ ## Session Learnings (from LOCAL) │
│ ### Learning 1 │
│ ### Learning 2 │
│ <!-- SENTINEL:END --> (marker preserved) │
└─────────────────────────────────────────────┘
After completing the pull:
git diff to review, git commit when satisfied| Don't | Do Instead |
|---|---|
| Overwrite entire skill files | Merge: template body + local learnings |
| Overwrite CLAUDE.md blindly | Preserve if exists (has customizations) |
| Forget to report preserved learnings | Always confirm learnings were kept |
| Skip the pull-template skill itself | This skill updates too, preserve its learnings |
(Learnings about the pull process itself will accumulate here)