跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
安全

plan

Epic decomposition into trackable, right-sized tasks. Three modes — audit-aware (codebase-audit reports), workflow-audit-aware (handoff.yaml with pre-rated findings), standalone (from scratch). Light convention scanning for projects without CLAUDE.md.

Terry Nyberg · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add Terryc21/workflow-audit --skill plan

技能说明


name: plan description: Epic decomposition into trackable, right-sized tasks. Three modes — audit-aware (codebase-audit reports), workflow-audit-aware (handoff.yaml with pre-rated findings), standalone (from scratch). Light convention scanning for projects without CLAUDE.md. version: 1.3.0 author: Terry Nyberg license: Apache-2.0 allowed-tools: [Glob, Grep, Read, Write, AskUserQuestion] metadata: tier: analysis category: planning

Implementation Plan Generator

Quick Ref: Mode detect → ingest or analyze → interactive input → understand → size → impact → phased plan → risk → test → rollback → write → proceed.

YOU MUST EXECUTE THIS WORKFLOW. Do not just describe it.

Required output: Every task MUST include Size, Urgency, Risk, ROI, Blast Radius, and LOE ratings. Do not omit these ratings.

Rating format: See radar-suite-core.md for column definitions, indicator scale, and table formatting rules.

Pre-flight: Git Safety Check

git status --short

If uncommitted changes exist:

AskUserQuestion with questions:
[
  {
    "question": "You have uncommitted changes. Commit before proceeding?",
    "header": "Git",
    "options": [
      {"label": "Commit first (Recommended)", "description": "Save current work so you can revert if this skill modifies files"},
      {"label": "Continue without committing", "description": "Proceed — I accept the risk"}
    ],
    "multiSelect": false
  }
]

If "Commit first": Ask for a commit message, stage changed files, and commit. Then proceed.


Golden Rule

Plans describe WHAT to change and WHERE — never HOW.

A plan task says: "Add keyboard dismissal to AddItemView.swift:45-80 and EditItemView.swift:32-60. Acceptance: tapping outside any text field dismisses the keyboard."

A plan task does NOT say: "Add .onTapGesture { UIApplication.shared.sendAction(#selector(UIResponder.resignFirstResponder)...) }"

Exception: Include code (≤10 lines) ONLY when referencing an existing pattern in the repo, showing a required API signature, or documenting a known framework gotcha. The two worked examples at the end of this file show the calibration.

Step 1: Mode Detection

Detect which mode to run in: audit-aware, workflow-audit-aware, or standalone.

# Check for codebase audit reports
Glob pattern=".agents/research/*-codebase-audit.md"
Glob pattern=".agents/research/*-tech-reportcard.md"

# Check for workflow audit handoff
Glob pattern=".workflow-audit/handoff.yaml"

Decision logic:

ConditionMode
User passes --workflow-auditWorkflow-audit-aware (error if no handoff found)
User specifies audit modeAudit-aware (error if no report found)
User specifies standalone modeStandalone (ignore all reports)
.workflow-audit/handoff.yaml found AND audit_date <14 daysWorkflow-audit-aware
Codebase audit report found AND <14 days oldAudit-aware
Report found AND ≥14 days oldStandalone (warn: "Stale audit report found — run /codebase-audit for fresh data")
Both workflow-audit and codebase-audit foundWorkflow-audit-aware takes priority (more specific)
No report foundStandalone

Parse the report filename date: YYYY-MM-DD-codebase-audit.md or YYYY-MM-DD-tech-reportcard.md. Parse audit_date from handoff.yaml for workflow-audit mode.

If multiple reports exist, use the most recent one.

Mode summary:

ModeSourceRatings Pre-computed?Convention Scan?
Audit-aware.agents/research/*-codebase-audit.mdNo — extract and rateNo — uses audit context
Workflow-audit-aware.workflow-audit/handoff.yamlYes — use as-isYes — if no CLAUDE.md
StandaloneUser-provided task descriptionNo — rate from scratchYes — if no CLAUDE.md

Output:

## Mode: [Audit-Aware / Workflow-Audit-Aware / Standalone]
Report: [filename or "none"]
Report age: [N days or "n/a"]
Staleness warnings: [count or "none"]

Step 2A: Audit Ingest (audit-aware mode only)

Skip this step in standalone mode — go to Step 2B.

Read the audit report and extract structured data.

Read file_path=".agents/research/[report-filename]"

Parse grades

Look for the GRADES_YAML HTML comment block first:

<!-- GRADES_YAML
categories:
  - name: Architecture
    score: 91
    grade: A-
    ...
-->

If no YAML block, fall back to parsing the grade summary line:

**Overall: B+ (84)** (Arch A- [91] | Security C+ [77] | ...)

Extract Top 10 Issues

Scan the report for the prioritized findings list. Each finding should have:

  • Category
  • Severity (CRITICAL / HIGH / MEDIUM / LOW)
  • Description
  • File:line reference
  • LOE estimate

Detect special conditions

ConditionWhat to extract
Incomplete (I) gradeCategory name + trigger description + file:line
DO NOT SHIP recommendationAll blocker items listed under the recommendation
CRITICAL findingsFull finding details

Produce Report Digest

## Report Digest

| Source | Date | Overall Grade |
|--------|------|---------------|
| [codebase-audit / tech-reportcard] | [date] | [grade] |

### Incomplete Triggers (if any)
| Category | Trigger | File |
|----------|---------|------|
| [category] | [description] | [file:line] |

### Ship Recommendation
[SHIP / CONDITIONAL SHIP / DO NOT SHIP] — [reason]

### Top Issues
| # | Category | Severity | Issue | File | LOE |
|---|----------|----------|-------|------|-----|
| 1 | [cat] | CRITICAL | [desc] | [file:line] | [Xh] |
| 2 | [cat] | HIGH | [desc] | [file:line] | [Xh] |
| ... | | | | | |

Step 2B: Codebase Analysis (standalone mode only)

Skip this step in audit-aware mode — audit already did this work.

Freshness

Base all analysis on current source code only. Do not read or reference files in `.agents

阅读完整来源 ↗