菜单
Design the complete skill set for the target OS — each skill's purpose, inputs, process, output format, auto-triggers, and quality checks.
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。
基于 SOC 职业分类
Audit a generated OS repository for completeness, consistency, and architectural integrity against the 15 design principles and 23-item checklist.
Transplant a skill, command, or capability from one agentic-system repo into another via a 5-phase gated protocol (inventory → import → localize → wire → verify). Use when grafting a feature from an upstream repo that has its own dependencies, platform assumptions, or upstream-specific file references that need adapting before the feature will work in the destination repo. Not for copying a self-contained file — use only when the feature has a dependency footprint and the destination repo has its own conventions the import must respect.
Design multi-dimensional evaluation rubrics with calibrated scales, thresholds, and function-adaptive weights.
Identify who evaluates the user's output, what each audience prioritizes, and where their priorities conflict.
Define the user context model — identity axes, context files, persona variants, and privacy rules for the target OS.
Design compounding feedback loops — what data accumulates, what reads it, and how the system gets smarter with use.
| name | design-skills |
| description | Design the complete skill set for the target OS — each skill's purpose, inputs, process, output format, auto-triggers, and quality checks. |
domain-input/ filesreference/skill-anatomy.md (structural template for skills)reference/architecture-patterns.md (design principles)Map the user's workflow to skills. Every recurring action and every event-driven response from domain-input/domain-workflow.md is a skill candidate.
Use these universal skill categories as a starting framework:
| Category | Purpose | Job-Search-OS Examples |
|---|---|---|
| Triage | Quick assessment — should the user invest time? | /quick-start, /job-fit-scorer |
| Research | Deep-dive on a target or topic | /company-research, /salary-research |
| Generate (artifact) | Create a deliverable from context + input | /resume-tailor, /cover-letter, /work-product |
| Generate (communication) | Create outreach or messages | /connection-request, /referral-request, /hiring-manager-msg |
| Prepare | Get ready for an event | /interview-prep |
| Practice | Interactive rehearsal | /mock-interview |
| Debrief | Post-event analysis → update history | /interview-debrief |
| Follow-up | Post-event communication | /thank-you-note |
| Track | Pipeline / status management | /app-tracker |
| Audit | Review existing assets for quality | /linkedin-audit |
| Negotiate | Analyze and respond to an offer | /negotiate |
| Retro | Periodic performance review | /weekly-retro |
| Artifact | Create a non-text deliverable (PDF, CSV) | /pdf, /generate-report |
| Batch | Process items at scale in parallel | /batch, /scan |
For each category, ask: "Does this domain have an equivalent? What would it be called? What would it do?"
Not every skill needs to exist in v1. Prioritize:
Target: 8-15 skills for v1. More than 18 means the scope is too broad. Fewer than 6 means critical workflow gaps exist.
For each v1 skill, design the full specification using this template:
## Skill: /[name]
### Purpose
[one sentence — what problem does this solve for the user]
### When to Use
- [trigger 1]
- [trigger 2]
### Inputs
- **Required:** [what the user provides]
- **Auto-loaded:** [which context-library files]
- **Optional:** [what's nice to have but not required]
### Process (High-Level)
1. [Step 1 — what to do]
2. [Step 2 — what to do]
3. [Step 3 — what to do]
### Output Format
[describe or sketch the output structure]
### Auto-Triggers
- After this skill runs: [what runs automatically — sub-agents, other skills]
- After this skill runs: [what state files get updated]
### Quality Checks
**Good output:**
- [specific criterion 1]
- [specific criterion 2]
**Bad output:**
- [specific anti-pattern 1]
- [specific anti-pattern 2]
### Empty Context Behavior
- If [file] is missing: [STOP / WARN / proceed]
- Message: "[exact text]"
Design the trigger chains — what automatically happens after each skill:
## Auto-Behavior Map
| After This Skill | Auto-Trigger | Purpose |
|-----------------|-------------|---------|
| /[create skill] | /review-as-[audience] | Quality gate |
| /[debrief skill] | Update [history file] | Accumulate data |
| /[any application action] | Prompt /[tracker] | Pipeline tracking |
Confirm which skill works with zero context (from quality-contract.md). Design it to:
Chart how skills relate to each other:
[triage] ──→ [generate artifact] ──→ [review sub-agent]
──→ [track]
[research] ──→ [prepare] ──→ [practice] ──→ [debrief] ──→ [history update]
──→ [follow-up]
[retro] reads ──→ [history], [tracker], [briefings]
Write the complete skill set design to output/designs/skill-designs.md.
For each skill, include the full specification from Step 3.
Tell the designer:
Skill set designed:
- [N] skills for v1: [list with one-line descriptions]
- [N] skills deferred to v2: [list]
- Auto-behavior chains: [N] auto-triggers
- Zero-setup entry point: /[X]
- Sub-agent triggers: [list which skills trigger which reviewers]
Skill flow: [one sentence describing the main user journey through skills]
Next: Run /design-sub-agents to design the reviewer personas.
Good skill design:
Bad skill design: