Fill me in

{
  "lang": "es",
  "idea": "One sentence.",
  "summary": ["Line 1", "Line 2"],
  "parts": [
    { "id": "skill", "label": "Skill", "purpose": "...", "reason": "...",
      "from": ["part:skills/fill-me-in", "part:skills/fill-me-in/scripts"] }
  ],
  "existing": [{ "id": "docs", "label": "Docs", "from": ["existing:."] }],
  "edges": [{ "from": "skill", "to": "docs", "label": "updates" }],
  "phases": [{ "name": "Build", "moves": ["..."] }],
  "open": { "pending": ["..."], "unverified": ["..."], "next": "..." }
}

Why I built it

After a long session with an agent, I lose track of what changed and why. A list of files does not help me. Fill-me-in aims at understanding, not at an inventory: the idea of the change, the reason for each part, and how it fits the existing code.

I ask for it in plain words, such as “fill me in”, “catch me up”, or “ponme al tanto”.

Only this session, in small numbers

The skill takes what changed only from the current session: the messages, the tool calls, and their results. To find the changes, it skips the git history, the ledger, and other sessions, because they can hold work that this session did not do. If the session holds no work, it says so and stops.

A reader holds about 4 items at a time. So the page uses 4 parts, 4 phases, and 6 diagram nodes at most.

Parts and moves

A part is a group of touched files with one purpose. The skill names each part by its purpose, not by its directory, and explains the change part by part, never file by file.

A move is one action that changed the state of the work. The page tells the moves as a short story of phases, such as setup, build, and verification. It always names each failed or denied command, and what the session did instead.

The page

map.mjs finds the parts, their states, and the existing code that refers to each part. It also lists the edges between them.

The model then writes the page data in the shape of the snippet. Then render.mjs builds one self-contained HTML page, prints its path, and opens it in the browser.

The page shows a diagram of the connections, with a detail panel for each part. If the render fails twice, the full briefing goes to the chat, with a text tree as the diagram. Each claim that the session does not show is marked as not verified.