create-plans_skill

This skill generates Claude-executable phase plans with verification criteria for solo development projects, including briefs, roadmaps, and handoffs.
  • Shell

1.2k

GitHub Stars

2

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill glittercowboy/taches-cc-resources --skill create-plans

  • README.md9.7 KB
  • SKILL.md17.4 KB

Overview

This skill creates hierarchical, Claude-executable project plans optimized for a single human (you) and a single implementer (Claude). Plans are compact prompts (PLAN.md) that contain objective, context, atomic tasks, verification criteria, and explicit outputs. The goal is fast, verifiable deliveries with repeatable handoffs and minimal coordination overhead.

How this skill works

On invocation it scans the repo for planning artifacts and git state, then offers context-aware options (start brief, create roadmap, plan phase, resume handoff). When planning a phase it generates PLAN.md as the actual execution prompt: objective, 2–3 atomic tasks, checkpoints, verification steps, success criteria, and SUMMARY.md output spec. It can load relevant domain expertise selectively to keep context use low and enforces checkpoint and deviation rules during execution.

When to use it

  • Planning a new project or converting a brief into an executable roadmap
  • Breaking a roadmap into atomic, Claude-runnable phase plans
  • Preparing a phase handoff for later execution or resumption
  • Creating verifiable tasks with automated CLI/API steps and human checkpoints
  • Splitting large work into many small, high-quality plans to avoid context degradation

Best practices

  • Keep each PLAN.md to 2–3 tasks and scope a phase to complete within ~50% context usage
  • Prefer automation: if Claude can run a CLI/API, include it in tasks (avoid human-action requests)
  • Include measurable verification criteria and a SUMMARY.md output spec for each phase
  • Use checkpoint:human-verify and checkpoint:decision sparingly and only for true human gates
  • Commit only meaningful milestones and phase completions; maintain git history for artifacts

Example use cases

  • Convert a product brief into a compact roadmap and domain-aware phase list
  • Create a phase plan to implement database schema, client, or API routes as separate PLAN.md files
  • Produce a handoff PLAN.md when pausing work so a subagent can resume later
  • Generate verifiable tasks that run CLI commands to deploy, test, or seed infrastructure
  • Load domain expertise (e.g., nextjs, macOS) selectively before creating domain-specific phases

FAQ

It works without a repo but recommends initializing git. If no repo is found it will ask to initialize and will explain when commits happen.

How big should a plan be?

Aim for 2–3 atomic tasks per PLAN.md so Claude stays under ~50% context pressure. Many small plans outperform a few large ones.

When does Claude stop and ask me?

Claude uses gates for architectural decisions, low-confidence research, verification failures, and before writing PLAN.md. Routine fixes and blocker resolutions are handled automatically and logged.

Built by
VeilStrat
AI signals for GTM teams
© 2026 VeilStrat. All rights reserved.All systems operational