Skip to content
Back to skills

Checkpoint

ASecurity

Mid-session orientation, pulled manually when the user feels drift or overload. Sorts the live session into decided / open / blocked / dropped, surfaces the threads still in the air, then proposes what would go to the brain and how to proceed — with options. Read-only and lightweight by design; the point is a low threshold to invoke. It sorts and proposes, it does not write — consolidate persists. Invoke on "checkpoint", "where are we", "let's sort this", "wo stehen wir", "kurz sortieren", "i...

  • 5 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 14, 2026
ai-agentsgorails

Security analysis

A100/100

Scanned October 14, 2026

npx -y skills add EpistemicLabs/eml --skill checkpoint --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Checkpoint?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Checkpoint
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/epistemiclabs-checkpoint/badge)](https://www.skillsdirectory.com/skills/epistemiclabs-checkpoint)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: checkpoint
description: Mid-session orientation, pulled manually when the user feels drift or overload. Sorts the live session into decided / open / blocked / dropped, surfaces the threads still in the air, then proposes what would go to the brain and how to proceed — with options. Read-only and lightweight by design; the point is a low threshold to invoke. It sorts and proposes, it does not write — consolidate persists. Invoke on "checkpoint", "where are we", "let's sort this", "wo stehen wir", "kurz sortieren", "innehalten".
---
<!-- managed-by: engramprotocol-setup -->

# checkpoint

Pause mid-session and get the picture straight — before the context drifts far
enough that reconstructing it becomes archaeology. The deliverable is **clarity
now**, not persistence.

Stay cheap: read the session, sort it, propose. **No writes.** A low threshold to
invoke is the whole feature — if this ever feels like half a consolidate, it's
too heavy.

**Neighbours:** `consolidate` is the *save* half (end of session, writes the
brain). The `session-discipline` system rule is the *automatic* version of this
pause. This skill is the *manual* pull — sort the live state, hand the rest to
consolidate.

## Steps

1. **Sort the live session** into four buckets — decided ✅ / open 🔲 / blocked ⛔
   / dropped ❌ — and list the **threads still in the air** so the user sees if
   the session is spreading thin. Blocked means a task with blockers (its status
   stays planned), not a status of its own.

2. **Surface the picture** in chat and stop:

       ## Checkpoint — <date>

       THREADS IN THE AIR (n)
         • <thread> — <one line>

       ✅ decided (n)    <title> — <one line>
       🔲 open (n)       <title> — <one line>
       ⛔ blocked (n)    <title> — waiting on <what>
       ❌ dropped (n)    <title>

3. **Propose — don't write.** Two short proposals, both optional for the user:
   - *Brain candidates:* the facts that would actually be lost if compaction hit
     before consolidate (decision made / status changed / newly blocked /
     progress worth an entry on task X). As many
     as there genuinely are — no cap, but don't pad with soft stuff either. Just a
     list — consolidate writes them, not this skill.
   - *How to proceed*, as options, e.g.:
     `a) continue with <X>  ·  b) tighten scope to <Y>  ·  c) break → fresh session  ·  d) consolidate now`

That's it. On "break" → consolidate; it writes the log and opens the fresh
session itself (its step 7) — the work does not go on here. On "consolidate" → hand into that skill (which re-judges the
candidates properly before any write).

## Guardrails
- Read-only. The checkpoint never writes engrams — it proposes, consolidate persists.
- If it feels heavy, cut back to sorting + the one "how to proceed" prompt.
- Chat picture follows the conversation language; brain candidates are phrased in
  English (consolidate needs them that way).

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.