This is a file-based planning system that creates three tracking files (task_plan.md, findings.md, progress.md) to help Claude manage complex work that'll take more than a handful of tool calls. It includes tamper detection using SHA256 hashes to prevent prompt injection attacks through plan files, which is a real concern when Claude is reading structured data that might contain instruction-like text. The hooks inject plan context before each prompt and remind Claude to update progress after file edits. It supports multiple concurrent plans in a .planning directory and can recover your session state after /clear. If you're doing research, breaking down a big refactor, or managing any multi-step project where you want persistent structure across a conversation, this gives you that scaffolding without reinventing it each time.
npx -y skills add guanyang/antigravity-skills --skill planning-with-files --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
mattpocock/skills