from skillattack30
Manus-style file-based planning pattern: create task_plan.md, findings.md, and progress.md to manage complex multi-step work and session recovery.
Implements a file-backed planning workflow for multi-step tasks. The skill prescribes creating and maintaining task_plan.md, findings.md, and progress.md so an agent can persist plan state, findings, and session logs on disk. It includes templates, helper scripts, and hooks to restore context and run a session catch-up.
Use this skill for projects or research requiring many tool calls, long-running tasks, or when you need recovery after clearing session state. Avoid for simple Q&A or single-file edits.
Best for interactive agents that can read/write files and run shell scripts (Code-capable agents like Codex, Copilot-type integrations, or local CLI-capable LLMs).
Manus-style file-based planning skill that creates task_plan.md, findings.md, and progress.md for managing complex multi-step work. The SKILL.md is thorough with excellent patterns (2-Action Rule, 3-Strike Error Protocol, security boundary awareness). Two of three shell scripts fail on Linux/macOS due to Windows CRLF line endings — a basic portability bug. The Python session-catchup script runs cleanly. The PreToolUse hook auto-reading task_plan.md is a documented injection risk, but the skill proactively mitigates it with clear guidance.
Solid skill conceptually. The CRLF issue is a basic but impactful bug — 2/3 scripts unusable on non-Windows systems. The security boundary section is a standout: rare to see a skill that identifies its own injection surface and provides mitigation. Would score higher on code quality if scripts were portable.
