AutoHarness vs planning-with-files
Which should you use? A side-by-side look at what each one does, who it suits, what it costs, how hard it is to set up, and where each falls short.
Open source · MIT
Claude Code plugin that learns reusable skills from your sessions and prunes unused ones.
Open source · MIT
A file-based planning skill that keeps AI coding agents on track across resets and crashes.
The short answer
Choose AutoHarness if you need…
- Developers who use Claude Code heavily and repeat the same kinds of tasks
- Teams that want project-level skills to build up without manual upkeep
- People who want to see and control what gets learned, since skills are plain files with a ledger marker
Choose planning-with-files if you need…
- Developers running long multi-step tasks in Claude Code or Codex CLI
- People who lose agent progress when the context window fills or is cleared
- Teams wanting a plan that lives in the project as plain markdown files
- Users of agents that support the Agent Skills standard
Side-by-side
| AutoHarness | planning-with-files | |
|---|---|---|
| Type | Dev tool | Dev tool |
| Price | Free and open source | Free and open source |
| Licence | MIT | MIT |
| GitHub stars | 10.6k | 27.8k |
| Setup difficulty | Beginner | Beginner |
| How you get it | See the README | Install with a package manager |
| Main language | Python | Shell |
| Last code update | 6 days ago | 6 days ago |
What each one is
AutoHarness
AutoHarness is a plugin for Claude Code, Anthropic's coding assistant. While you work, it turns useful episodes from your sessions into reusable skill files under .claude/skills/, merges similar skills, updates them as they are used, and removes ones that stop being used. It only manages skills it generated itself and leaves your own or installed skills alone.
planning-with-files
This is a skill for AI coding agents that stores a task plan, research notes, and progress log as markdown files in your project. Lifecycle hooks re-inject that planning context so the agent can pick up where it left off after /clear, context compaction, or a crash. The README says it works across 60+ agents via the Agent Skills standard, with native plugins for several hosts including Claude Code and Codex CLI.
Best for
AutoHarness
- Developers who use Claude Code heavily and repeat the same kinds of tasks
- Teams that want project-level skills to build up without manual upkeep
- People who want to see and control what gets learned, since skills are plain files with a ledger marker
planning-with-files
- Developers running long multi-step tasks in Claude Code or Codex CLI
- People who lose agent progress when the context window fills or is cleared
- Teams wanting a plan that lives in the project as plain markdown files
- Users of agents that support the Agent Skills standard
Strengths
AutoHarness
- Installs as a Claude Code plugin with zero third-party Python dependencies, according to the README
- Only touches skills it wrote itself, so your own skills are not modified
- Tunable cadence and scope through AUTOHARNESS_* environment variables
planning-with-files
- Plain markdown files on disk that you can read, edit, and version-control or gitignore
- Works through the Agent Skills standard across many agents, with native plugins for several hosts
- MIT licensed with a published benchmark and install options via npm, the Claude Code marketplace, or npx
Not ideal for
AutoHarness
- Users who do not use Claude Code, since the plugin only runs inside it
- Users on Python older than 3.11 or without python3 on PATH, because hooks silently stay off
- People who want a measured benchmark score for skill quality, since the README says validation comes from use rather than a benchmark
planning-with-files
- Short one-off questions where a plan file adds overhead
- People who want a hosted service rather than local files in their repo
- Users of hosts not listed in the README, since native support depends on the host and results may vary
Trade-offs
AutoHarness
- Requires Python 3.11+ as python3 on PATH; the README warns that older interpreters silently disable the hooks
- The README lists Linux and macOS as the platforms; Windows is not mentioned
- Effectiveness is described as validated by use, not a benchmark, so quality claims are hard to verify independently
planning-with-files
- Benchmark figures are from the project's own measurements and are reported with stated limits; independent verification is not shown in the README
- Commands and tools vary by host, so the experience differs between integrations
- Session catchup modes can read local agent session records, which the README says requires an explicit opt-in
Getting started
AutoHarness
Add the marketplace (type inside Claude Code)
/plugin marketplace add tigerless-labs/autoharnessInstall the plugin (type inside Claude Code)
/plugin install autoharness@autoharnessplanning-with-files
Install the skill globally via npx skills
npx skills add OthmanAdi/planning-with-files --skill planning-with-files -gFrequently asked questions
What is the difference between AutoHarness and planning-with-files?
AutoHarness: Claude Code plugin that learns reusable skills from your sessions and prunes unused ones. planning-with-files: A file-based planning skill that keeps AI coding agents on track across resets and crashes. AutoHarness suits developers who use claude code heavily and repeat the same kinds of tasks, while planning-with-files suits developers running long multi-step tasks in claude code or codex cli.
Are AutoHarness and planning-with-files free?
AutoHarness: Free and open source. planning-with-files: Free and open source.
Is AutoHarness or planning-with-files more popular?
planning-with-files has more GitHub stars (27.8k) than AutoHarness (10.6k). Popularity isn't the same as fit — check which one matches your use case above.
What are other alternatives to AutoHarness and planning-with-files?
Other options people compare them with include Agentic Awesome Skills, wshobson/agents, Distilly, claude-obsidian, OpenHands.
Keep comparing
Built from our catalogue entries for both products: open-source details come from each project's README (AI-assisted, refreshed weekly) and GitHub data; AI-tool details are editorial. Spotted something wrong? Let us know from either product's page.