tidy-skill

How tidy-skill compares

Honest positioning against the skills people most often install next to (or instead of) tidy-skill. Stars and install counts are not quality claims — they are market context.

中文版：comparison.zh-CN.md

Project What it optimizes Default on-disk behavior Best when
tidy-skill (this repo) Local environment hygiene: repo litter, workspace caches, machine sprawl Chat-first; Class A–E placement; DryRun cleanup; score/gate; doctor→repair Agents litter roots, caches grow, you want a measurable gate
planning-with-files Long-task survival across /clear and compaction Writes task_plan.md / findings.md / progress.md (or .planning/<slug>/) Multi-step tasks that must resume after context loss
obra/superpowers Spec → plan → TDD methodology pack Skill library across many hosts You want an opinionated engineering workflow, not hygiene
addyosmani/agent-skills Define → plan → build → verify → review → ship 24 lifecycle skills You want process discipline for production code
anthropics/skills Agent Skills format / reference Spec + examples You are authoring or studying the skill standard
affaan-m/ECC Agent harness OS: skills, hooks, memory, security Large multi-component install; doctor/repair; session retention knobs You want a full harness, not just hygiene
garrytan/gstack Role slash-commands + safety guards for shipping /careful /freeze /guard; team auto-update You want CEO/QA/CSO-style productivity roles
mattpocock/skills Small composable engineering skills Plugin subscribe vs editable skills.sh copy You want failure-mode skills, not environment governance

One-liner

planning-with-files keeps long tasks alive across context loss.
tidy-skill keeps the machine and repo from drowning in agent litter.
Use both: put PWF working memory under .planning/ (preferred) or opt the root triple in via policy; let tidy score, audit, and gate everything else.

Philosophy clash (and how to resolve it)

Axis planning-with-files tidy-skill default
Context model Filesystem = durable disk Chat = default; files need intent
Root process Markdown Expected (task_plan.md, …) Forbidden patterns (*_plan.md, progress.md, …)
Parallel work .planning/YYYY-MM-DD-slug/ .agent_tmp/ / .agent_reports/
Lifecycle Gitignored working memory Score / audit / DryRun / retention
Safety Plan hooks, completion gate Read-only audits; no auto VHDX/config mutation

Resolution (pick one):

  1. Preferred: keep PWF under .planning/ — tidy-skill classifies .planning/** as intentional Class C working memory (allowed).
  2. Root triple: add a project policy that opts the PWF names in:
{
  "version": 1,
  "planning_root_globs": ["task_plan.md", "findings.md", "progress.md"],
  "min_score": 80,
  "require_agent_dirs": true
}

Copy from skills/tidy-skill/references/tidy-skill.policy.pwf.example.json as .tidy-skill.json. Those names stop counting as suspicious root litter and are not cleaned by default sweeps. Still gitignore them.

  1. Ignore only (weaker): ignore_root_globs hides names from forbidden matching without labeling them as planning memory.

Without an opt-in, a default PWF root triple still looks dirty to tidy-skill — by design. Hygiene and crash-proof planning are different products; coexistence is explicit, not silent.

What tidy-skill uniquely owns

  1. Three-layer hygiene — repository artifacts, multi-repo workspace caches, local machine (WSL2 / Docker / VHDX / package / model caches).
  2. Classes A–E + Artifact Intent Check — placement law before write.
  3. Measurable CI gate — score_repo_hygiene, hygiene_snapshot gate, tidy_doctor exit codes.
  4. Doctor → repair loop — diagnose then DryRun-first safe fixes (tidy_repair), without auto host/VHDX mutation.
  5. Project policy shared by Python + PowerShell — .tidy-skill.json.
  6. Offline stdlib Python — no runtime network dependency for core scripts.
  7. Safety verbs — dryrun / careful / guard (same meaning in docs and CLIs).

What peers do better (and what we borrowed)

Peer surface Who does it well What tidy-skill took
Doctor → repair / uninstall path ECC tidy_repair.py + uninstall section in installation.md
Safety as short verbs gstack (/careful /guard) dryrun · careful · guard product language
Failure-mode README, not methodology dump mattpocock/skills Four failure modes (root litter / cache sprawl / unsafe cleanup / no CI gate)
Subscribe vs editable install mattpocock, ECC Dual-path install matrix (skills CLI vs clone/local)
Session retention knobs ECC (ECC_SESSION_RETENTION_DAYS) Documented tmp 7d / reports 30d + clean script flags
Broad host mirrors / marketplace PWF, superpowers, addy Keep thin; expand only with real install paths
Recovery-after-wipe narrative PWF, superpowers Keep honest self-audit + fixture evals; no fake benchmarks
Side-by-side comparison page addy This document

Still not copying: 200+ skill harnesses, Memory Vaults, TDD methodology packs, auto host-config writers, star-count marketing.

Non-goals

This site is open source. Improve this page.