Save For Later

Most brainstorms end the same way. A long chat. Maybe a design doc. Then the tab closes.

Next session, you re-explain the campaign from scratch. A sub-agent gets a vague brief and invents constraints you already rejected. The "don't do X" line that took twenty minutes to discover is gone.

That is not a model problem. It is a continuity problem.

SFL - Save For Later is a small skill that upgrades brainstorming with a living project pack: one place where locked decisions, rejected approaches, subtleties, and handoff notes accumulate for the life of a campaign.

Two densities, one truth

The full pack is skimmable human continuity: status, intent, decisions, open threads, subtleties, handoff notes, file index.

The thin slice is task-shaped: mission, MUST, NEVER, exact needles, paths to open. Nothing else.

Workers should never become a second author of campaign truth. They write receipts if you want them; they do not invent the spine.

The pack spine

Every campaign pack uses the same sections:

1. Status - phase, next action, blockers
2. Intent - why this exists
3. Decisions - dated locks and rejections
4. Open threads - unresolved forks
5. Subtleties - promoted nuance only (not raw chat)
6. Handoff notes - if you pick this up now
7. File index - load-bearing paths
8. Changelog - what changed in the pack

You can maintain that by hand in markdown. The shipped helper also keeps an append-only event log and re-renders the pack so you do not lose history while editing the view.

Brainstorm upgrade, not brainstorm replacement

If you already use a structured brainstorming discipline, SFL runs beside it:

1. Name the campaign early.
2. Birth a pack (or adopt mid-flight from existing design notes).
3. Every locked decision or rejected approach gets promoted into the pack.
4. When you park, the handoff lives inside the pack.
5. When you write the design and plan, point the file index at them.

The design doc and plan stay the build contract. The pack is continuity and nuance around them.

Hard gate

This matters enough to say twice.

A pack is not permission to build.

If your brainstorming method has a hard gate before implementation, keep it. SFL does not cross that line for you. Continuity without a gate just helps you ship the wrong thing faster.

What ships in the zip

• SKILL.md for agent harnesses that load skills as markdown
• sfl.py - Python 3 stdlib helper: init, promote, slice, render
• demo pack + example thin slice
• install notes

No cloud account. No proprietary runtime. Drop the skill in, point it at a project root, start saving for later.

Get SFL

Email unlocks the zip and the copyable skill text. Product landing also lives at /applications/sfl.

Install in thirty seconds

1. Unzip.
2. Copy the sfl folder into your agent skills directory.
3. In a project: python3 path/to/sfl/sfl.py init my-campaign
4. Promote as you lock decisions. Slice when you brief a worker.

Full steps are in INSTALL.md inside the package.

Why this is the product

I got tired of re-teaching the same campaign every morning and watching sub-agents re-decide problems I had already closed.

The fix is not a longer chat. The fix is a file that survives the chat, with two densities so humans and workers consume different projections of the same truth.

Save the decision when you make it. Save the rejection when you earn it. Save the handoff when you leave.

That is SFL.