AgentSkillsCN

core

混合任务系统——结合 prd.json、原生任务与解析学习机制。

SKILL.md
--- frontmatter
name: core
description: Hybrid task system - prd.json + native Tasks + resolution learning
allowed-tools: Read, Write, Edit, TaskCreate, TaskUpdate, TaskList, Grep, Glob
model: sonnet
user-invocable: false

Hybrid Task System

Two Layers

LayerToolPurpose
Long-termprd.jsonSprint history, resolutions
Short-termNative TasksActive work

Rule: Work with native Tasks during session, batch-update prd.json at end.

prd.json Story Schema

json
{
  "id": "S26-001",
  "title": "Fix tooltip clipping",
  "priority": 1,
  "passes": null,
  "type": "fix",
  "category": "components",
  "notes": "",
  "resolution": ""
}
FieldValues
passesnull (pending), true (done), false (failed), "deferred"
typefix, feature, refactor, qa, perf
priority0=critical, 1=high, 2=medium, 3=low
resolutionHOW it was fixed (learning)

Resolution Learning

When completing bug fixes, document HOW:

code
[PATTERN]: [SPECIFIC FIX]

Examples:

  • null-check: Added optional chaining at line 45
  • missing-import: Added import for DateRange
  • type-mismatch: Changed Record<string, T> to Partial<Record<K, T>>
  • overflow: Added max-h + overflow-auto

Mistake Categories

CategoryFix Pattern
null-checkobj?.prop
missing-importimport { X } from 'y'
type-mismatchCorrect type annotation
missing-keyAdd missing Record key
overflowoverflow-auto + max-h

Context Optimization

ActionDo This
Check statusRead prd.json header (30 lines)
Start taskGrep specific story
Track progressNative TaskUpdate
Complete workBatch edit prd.json at session end

Archive Trigger

When completedStories > 500 or prd.json > 100KB: → Archive to prd-archive-YYYY-MM.json