Keep a long-running coding task on track by saving the plan, findings and evidence in project files that a new session can read.
Choose a project that needs more than one session. Define what belongs in the phase plan, source notes, and verification log so another agent can identify the next action.
List the project, key phases, current blocker, and approval boundaries.
Copy the request into a new Floatboat conversation and provide the required material.
Download FloatboatPreview and copy the request. It is not sent automatically from this page.
Reference: Planning with Files https://github.com/OthmanAdi/planning-with-files/blob/master/skills/planning-with-files/SKILL.md Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding. My task: Use Planning with Files for a multi-step database migration. Record the phases and next action in task_plan.md, source findings in findings.md, and checks in progress.md. Start with a read-only inventory, and wait for approval before changing the database. Expected deliverable: A three-file planning outline and a fresh-session recovery check.
The workflow uses three project files. task_plan.md holds the goal, phase status and next action. findings.md keeps research and source notes. progress.md records what ran, what failed and which checks passed. The Skill asks the agent to reuse an existing plan and keep state current as work moves between phases. A new session can read these files and resume from the documented state. Automatic injection and lifecycle recovery depend on enabled hooks and the host. The upstream workflow treats reading local session records as a separate, explicit catch-up operation.
File-based planning is useful when work spans multiple sessions or reviewers. Preserve the next decision, what was tried, and which evidence belongs to the current revision.
Split a database migration into inventory, dry run and rollout phases. Record the last verified snapshot and the first unchecked step so a fresh session can resume from the dry run without assuming the rollout was approved.
Give a research session a short decision question, then store source links, dates and conflicting claims in findings.md. Keep the chosen criteria and next step in the plan so material copied from the web does not become a project instruction.
Record the revision, exact checks that passed and the unresolved blocker in progress.md. A new reviewer can reproduce the missing check and see which conclusions belong to the previous run.
Use named plan directories for independent tasks and pin each host to the intended plan where supported. Confirm which plan is active before editing phase status so one thread does not advance the other task.
Start with one project and a goal you can verify. Carry the next action forward after each phase, and check that a fresh session can resume from the files before you rely on automatic recovery.
Use the upstream install command for your compatible host. Start in the intended project directory and ask the agent to use Planning with Files for the specific task.
Read an existing plan before making changes. Record the goal, phases and one next action in the plan, and put source notes and executed checks in their respective files.
Ask a new session to read the project files and identify the pending action. If you need automatic recovery, verify the host’s hook setup using its upstream guide before depending on it.
Read another Skill guide and check its original source before using it.
