generate-execution-plan
Creates a versioned, machine-readable YAML execution plan for multi-phase tasks. Use when: task spans 3+ files, multiple sessions, requires staged delivery, or needs a recoverable plan artifact. Do NOT use when: simple single-turn tasks that complete in one response. Signals: execution plan, multi-phase task, project plan, phased delivery, create plan, plan this work, staged execution, PLAN.yaml
- License 许可
- MIT
- 最近更新
- 2026-11-06
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- generate-execution-plan
- license
- MIT
- metadata
- { "tags": [ "planning", "execution", "agent", "multi-session" ], "author": "Igor Beylin", "version": "1.0.0", "eval-status": "untested", "last-tested": 1781654400000, "model-target": "claude-sonnet-4" }
- description
- Creates a versioned, machine-readable YAML execution plan for multi-phase tasks. Use when: task spans 3+ files, multiple sessions, requires staged delivery, or needs a recoverable plan artifact. Do NOT use when: simple single-turn tasks that complete in one response. Signals: execution plan, multi-phase task, project plan, phased delivery, create plan, plan this work, staged execution, PLAN.yaml
- compatibility
- Tested: Claude Code, Claude Sonnet 4+ Kernel alignment: - execution_plan_kernel.v1
03 SKILL.md 正文
generate-execution-plan
Overview
Produces a PLAN.yaml with phased steps, per-step success criteria, recovery paths, and a completion checklist. Each step declares its input artifacts and output artifacts to enable stateless recovery.
Workflow
- Load
execution_plan_kernel.v1Tier 1 - Decompose the task into phases (max 7 per plan, break into sub-plans if larger)
- For each phase: define WHAT (output artifact), HOW (key decisions), DONE WHEN (acceptance criteria)
- Identify recovery path for each phase (what to do if step fails)
- Output: complete PLAN.yaml + human-readable summary table
Output Format
plan-id: <id>
version: 1.0.0
created: <date>
model: <model>
task: <one line>
phases:
- id: PHASE-00
name: <name>
outputs: [<artifact>]
done-when: <criterion>
recovery: <recovery instruction>
04 同仓库更多 Skills
optimize-kernel
Transforms any l9-ops kernel YAML into a doctrine-compliant three-tier progressive disclosure structure. Use when: auditing an existing kernel, running the optimize-kernel CI step, or onboarding a kernel from another source. Do NOT use when: authoring a brand-new kernel from scratch (use templates/kernel.yaml.template instead). Signals: optimize kernel, audit kernel compliance, apply doctrine, kernel drift, tier structure missing, trigger triad missing, rewrite kernel
generate-skill-md
Scaffolds a production-ready SKILL.md from a natural language description of the skill's purpose. Use when: authoring a new skill from scratch, onboarding an external skill into the l9-ops standard, generating skill structure from a requirements brief. Do NOT use when: modifying an existing skill (edit directly). Do NOT use when the description is too vague to produce a Trigger Triad. Signals: create skill, new skill, scaffold SKILL.md, write SKILL.md, add skill to library, skill template
review-code-l9
Performs an 8-dimension L9-compliant code review on a diff, file, or PR. Use when: reviewing a PR, auditing new code, validating an implementation against spec, checking a diff before merge. Do NOT use when: non-code review tasks (use a different skill). Signals: review code, review PR, code review, diff review, check implementation, audit code, validate against spec, L9 review