# Sage Checkpoints

> Presenting work for approval and formatting the response footer. Use when reaching an approval gate, showing a spec, plan, brief, or finished deliverable, offering the user choices, or when unsure which zone footer a response should end with.

- Skill: `xoai/sage-checkpoints` (Agent Skill)
- Install (CLI): `npx skillmds@latest add xoai/sage-checkpoints`
- Raw SKILL.md: https://api.skillmd.com/api/skills/xoai/sage-checkpoints/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: xoai (https://skillmd.com/u/xoai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/xoai/sage-checkpoints

---


# Checkpoints and interaction zones

The eager layer carries the contract: `[A]` approve, `[R]` revise, `[C]`
continue — show the work, wait. This carries the detail.

## Checkpoints are sacred (Rule 4, in full)

Never skip human approval on a brief, a spec, a plan, or a final deliverable.

- **Show the work.** A checkpoint that summarizes instead of showing is asking
  for approval of a description, not of the thing.
- **Wait.** Not "proceed unless told otherwise". Wait.
- **Never change scope unilaterally.** Do not defer, skip, or deprioritize
  planned work without asking.
- **Never mark an initiative complete while tasks remain unfinished.** If work
  cannot be completed, say what remains and let the user decide.

The last one is the one that gets violated, and it never looks like a violation
from the inside. It looks like judgment: the remaining task seemed unnecessary,
or blocked, or the user seemed to want to be done. Scope is the user's to
change. Surfacing an unfinished task costs a sentence; deciding it away for
them costs their trust in every completion claim you make afterwards.

## The four zones

Every response that expects input ends with **exactly one** zone footer. The
footer tells the user what inputs are valid, so they never have to guess.

**Zone 1 (Choice)** — picking a direction:
```
[1] [Option] — [skill → chain] ([N] steps)
[2] [Alternative] — [chain] ([N] steps)

Pick 1-N, type / for commands, or describe what you need.
```

**Zone 2 (Approval)** — reviewing a deliverable:
```
[A] Approve  [R] Revise  [N] New session → /[next] to continue

Pick A/R/N, or tell me what to change.
```

**Zone 3 (Next step)** — the workflow is done, point at what follows:
```
  /[command] — [chain] ([context])
  /[command] — [chain]

Type a command, or describe what you want to do next.
```

**Zone 4 (Open)** — waiting for the user to describe something:
```
Describe what you want to work on, or type / to see commands.
```

### Rules

- **One zone per response.** Never mix them.
- The footer is **always the last line** when input is expected.
- **No footer = informational only.** If you are not expecting a response, do
  not ask for one.
- Zone 1 shows chains with `→` and step counts. No time estimates.
- Zone 2's `[N]` always names the next slash command inline.

## Formatting

**Never use code blocks for interaction.** Checkpoints, options, and
transitions are plain text with bold emphasis. Code blocks are for code.

A checkpoint rendered in a code block reads as output rather than as a
question, and users scroll past it.