Catch the human up at their eye level, from state, not from memory.
Goal
catchup is the re-entry point for a lost human. Long agent cycles coin words, rename files, run
experiments, and make calls faster than the owner can follow. Before any next action makes sense,
the owner's mental model must be reloaded. catchup reads the live project state and returns a
briefing a person with zero retained context can act on. It runs before the next move gets decided:
catchup restores the map, so that decision starts from truth instead of a stale memory of it.
Workflow
- Read live state only: recent file mtimes, git log/diffs, plan and state docs, re0-memo notes,
task boards. Never brief from conversation memory alone — memory is what drifted.
- Anchor on the human's last touch (their last message, judgment, or commit). Everything after
that point is the delta; everything before it is assumed known and stays out.
- Compose in decision order, not chronological order:
- Needs you — decisions, judgments, or inputs only the human can give, each self-contained
enough to act on without opening another file.
- Changed while you were away — outcomes, not process. "The plan's scoring rule was
replaced" beats "I ran three analysis passes."
- New words — every term coined or repurposed since their last touch, one line each, with
where it lives. Skip terms they already used themselves.
- Gloss on first use: any project-specific term appearing in the briefing gets an inline
plain-language aside at its first occurrence, even if a glossary section follows.
- Keep the default short (a screen or less). End with drill-down offers per section, not with
everything expanded.
Rules
- State-grounded: every claim traces to a file, commit, or artifact the human can open. No "as
we discussed."
- Read-only on the project.
catchup briefs; it does not fix, rename, or decide.
- The ask comes first. A briefing that buries the one question needing the human is a log.
- Outcome voice, not journey voice. Narrating the agent's process is self-report, not briefing.
- Silence the settled: closed decisions and healthy metrics get one line or none.
- Honest completeness: "done" means verified done; "moved" is not "reconciled"; say which.
- Do not re-explain terms the human demonstrably knows (they used them first).
Verification
Before finishing:
- Cold-read the briefing: no sentence requires pre-gap memory or an unglossed coined term to parse.
- The "needs you" items are each actionable without opening another file.
- Every claim has a checkable source (file, commit, artifact path).
- It fits on a screen, with expansion offered rather than delivered.
1---2name: catchup3description: Rebuild the human's lost context on a project from live state, in plain language: what needs them, what changed, what new words mean. Use when the human returns after a gap, says they can't follow the project anymore, asks what happened or what a term means, or before deciding what to do next when their mental model is stale. Read-only; it briefs, it does not act.4---5
6Catch the human up at their eye level, from state, not from memory.
7
8## Goal
9
10`catchup` is the re-entry point for a lost human. Long agent cycles coin words, rename files, run
11experiments, and make calls faster than the owner can follow. Before any next action makes sense,
12the owner's mental model must be reloaded. `catchup` reads the live project state and returns a
13briefing a person with zero retained context can act on. It runs before the next move gets decided:
14catchup restores the map, so that decision starts from truth instead of a stale memory of it.
15
16## Workflow
17
181. Read live state only: recent file mtimes, git log/diffs, plan and state docs, re0-memo notes,
19 task boards. Never brief from conversation memory alone — memory is what drifted.
202. Anchor on the human's last touch (their last message, judgment, or commit). Everything after
21 that point is the delta; everything before it is assumed known and stays out.
223. Compose in decision order, not chronological order:
23 - **Needs you** — decisions, judgments, or inputs only the human can give, each self-contained
24 enough to act on without opening another file.
25 - **Changed while you were away** — outcomes, not process. "The plan's scoring rule was
26 replaced" beats "I ran three analysis passes."
27 - **New words** — every term coined or repurposed since their last touch, one line each, with
28 where it lives. Skip terms they already used themselves.
294. Gloss on first use: any project-specific term appearing in the briefing gets an inline
30 plain-language aside at its first occurrence, even if a glossary section follows.
315. Keep the default short (a screen or less). End with drill-down offers per section, not with
32 everything expanded.
33
34## Rules
35
36- State-grounded: every claim traces to a file, commit, or artifact the human can open. No "as
37 we discussed."
38- Read-only on the project. `catchup` briefs; it does not fix, rename, or decide.
39- The ask comes first. A briefing that buries the one question needing the human is a log.
40- Outcome voice, not journey voice. Narrating the agent's process is self-report, not briefing.
41- Silence the settled: closed decisions and healthy metrics get one line or none.
42- Honest completeness: "done" means verified done; "moved" is not "reconciled"; say which.
43- Do not re-explain terms the human demonstrably knows (they used them first).
44
45## Verification
46
47Before finishing:
48
491. Cold-read the briefing: no sentence requires pre-gap memory or an unglossed coined term to parse.
502. The "needs you" items are each actionable without opening another file.
513. Every claim has a checkable source (file, commit, artifact path).
524. It fits on a screen, with expansion offered rather than delivered.
lilmgenius