# Planning With Files > Use when working with a skill that enables persistent planning and "working memory" on disk using markdown files. - Skill: `tomevault-io/planning-with-files-6` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/planning-with-files-6` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/planning-with-files-6/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Docs & Writing - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/planning-with-files-6 --- # Planning with Files (Manus-style) This skill enables the agent to maintain a persistent "working memory" on disk, allowing for complex, multi-step tasks to be executed reliably without losing context. ## Core Files (The "Disk") The agent MUST manage three specific markdown files in the project root: 1. **`task_plan.md`** (The Map): Stores the high-level goals, phases, and status. 2. **`findings.md`** (The Knowledge): Stores research, technical decisions, and content "scraped" from the environment. 3. **`progress.md`** (The Log): Stores a chronological log of actions, identifying "where we are" after a restart. ## Rules of Engagement ### 1. The "Plan First" Rule **NEVER** start a complex coding or research task without first checking for (or creating) a `task_plan.md`. - If one exists, **READ IT** to ground yourself. - If none exists, **CREATE IT** using the `task_plan.md` template from the `references/` folder. ### 2. The "2-Action" Rule (Context Offloading) The agent has a limited "RAM" (context window). To prevent "context drift": - **Every 2 browser actions** (search, visit, read), you **MUST** summarize and save findings to `findings.md`. - **Every 2 major file edits**, you **MUST** update `progress.md`. - **NEVER** keep large blobs of text (like docs or scrape results) in your context. Save them to `findings.md` immediately. ### 3. The "Read Before Write" Rule (Attention Manipulation) Before making a significant technical decision or starting a new phase: - **READ** `task_plan.md` again. - This forces the global goal back into the "recent tokens" of your attention span, reducing hallucinations. ### 4. The "3-Strike" Error Protocol If a tool or action fails: 1. **Attempt 1**: Analyze the error, fix the arguments, and retry. Log in `progress.md`. 2. **Attempt 2**: Try a *different* tool or approach (e.g., if `grep` fails, use `find`). Log the pivot in `progress.md`. 3. **Attempt 3**: **STOP**. Do not loop. Write a "Blocker" status to `task_plan.md` and ask the user for help. ## Workflow Triggers Activate this skill when: - The user asks for a "plan" or "roadmap". - The task involves >3 files or >2 distinct phases. - The user mentions "context lost" or "where were we?". ## Setup If the files do not exist, copy the templates from the `references/` directory of this skill to the project root. --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/zhuanggenhua) — claim your Tome and manage your conversions.