Reusable skills for Claude Code.

A skill is a self-contained directory with a SKILL.md (frontmatter + instructions) and optional references/, scripts/, assets/ subdirectories. Drop a skill into ~/.claude/skills/<name>/ (user-level) or <repo>/.claude/skills/<name>/ (project-level) to make it available to Claude Code, then invoke with /<name> or have the model trigger it via the skill's description heuristics.

tldr-plan is not a plan validator. It does not approve or reject the plan, emit severity findings, or perform the audit itself. It is also not a one-shot summarizer that runs once and is done.

What it does: distill a raw implementation plan into a compact-first, self-contained TLDR artifact — problem context, assumptions, scope and out-of-scope, hard constraints, D0–D6 decision trace, acceptance criteria, evidence requirements, stop conditions, and optional diagrams.

Iterative by design. Run it after each plan revision; the human audits the TLDR, fixes gaps in the plan, re-runs the skill, audits again. Handoff to a coding agent happens only after the human audit passes — tldr-plan is the audit surface, not the handoff event.

plan v1 → tldr-plan v1 → human audit finds gaps
       → revise plan v2 → tldr-plan v2 → human audit
       → … → human audit passes → THEN hand the plan to a coding agent

Single reader: the human auditor. The implementation agent does NOT read this artifact; the agent reads the original plan file directly.

If you want findings/severity, use a validator skill (see "vs related skills" below). If you want forward-looking ADR/design-spec generation, use a preflight skill. This skill sits upstream of all those: it translates a long or ambiguous plan into a compact artifact a human can audit before handing the plan to an agent.

claude-skills/
  README.md
  LICENSE
  skills/
    <skill-name>/
      SKILL.md
      references/    # optional supporting docs
      scripts/       # optional helper scripts
      assets/        # optional binary assets
  docs/              # cross-skill authoring guides
  scripts/           # repo-level tooling (lint, package)

Recommended (via the open skills CLI from Vercel Labs — works against any repo with this layout):

# user-level (available across all projects)
npx skills add https://github.com/taoluo/claude-skills --skill tldr-plan -g -y

# project-level (scoped to one repo, run from project root)
npx skills add https://github.com/taoluo/claude-skills --skill tldr-plan -y

Manual fallback:

git clone --depth 1 https://github.com/taoluo/claude-skills.git /tmp/cs
cp -R /tmp/cs/skills/tldr-plan ~/.claude/skills/   # or <repo>/.claude/skills/
rm -rf /tmp/cs

Then in Claude Code: /tldr-plan <plan-file-path> (or let the model trigger it based on the skill's description).

See docs/authoring-guide.md for SKILL.md structure, frontmatter gotchas (especially the colon-in-description trap), description tuning for the CLI's trigger heuristic, and a checklist for verifying before you publish.

MIT. See LICENSE.