agentleFS
TheArcitectTheArcitect/claude-skills/session-continuity/SKILL.md

session-continuity

Skill2 starsChanged 9 months agoMITView on GitHub

Checkpoint protocol for maintaining conversation state across all disruption types — compaction, session limits, crashes, sleep, multi-day gaps. Creates structured checkpoints that allow any future Claude instance to reconstitute the conversation accurately. Use at natural stopping points, before expected compaction, or when resuming after any interruption.

7 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. 0 are common, 14 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 14

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every conversation disruption loses state.

    line 14
  2. Distinctive:

    Before expected compaction: When context is getting long, create a checkpoint before it compresses

    line 23
  3. Distinctive:

    End of session: Always checkpoint when a conversation is ending

    line 24
  4. Distinctive:

    Write checkpoints to a persistent file:

    line 56
  5. Distinctive:

    When starting a new session or resuming after a gap:

    line 65
  6. Very specific:

    Read checkpoints.md — find the most recent checkpoint

    line 67
  7. Distinctive:

    Read the WHERE WE ARE section — orient to the location in the work

    line 68
  8. Distinctive:

    Read WHAT'S UNRESOLVED — these are your entry points

    line 69
  9. Distinctive:

    Read EMOTIONAL CONTEXT — calibrate tone

    line 70
  10. Distinctive:

    Do NOT re-summarize everything.

    line 72
  11. Distinctive:

    Read the last 2-3 checkpoints, not just the most recent

    line 77
  12. Distinctive:

    Ask one orienting question: "Last time we were working on X.

    line 79
  13. Distinctive:

    coherency-audit: Run a coherency check after reconstitution from a long gap.

    line 85
  14. Distinctive:

    Checkpoints are locations, not summaries: Write them as if you're dropping a pin, not writing a report.

    line 96
  15. Distinctive:

    The checkpoint is for the next Claude, not the current one: Write it for someone who has no other context.

    line 100

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Session Continuity Skill

Purpose

Conversations end. Context compresses. Sessions expire. Crashes happen. Sleep happens. Days pass. This skill maintains state across all of these boundaries through structured checkpoints that a future Claude instance can use to reconstitute accurately.

The Problem

Every conversation disruption loses state. Compaction summaries are lossy. New sessions start cold. The user has to re-explain context, re-establish rapport, and re-orient Claude to where they were. This friction compounds. Important threads get dropped. Nuance evaporates.

The problem isn't memory — it's continuity. Claude remembers facts. It doesn't remember where we were.

Checkpoint Protocol

When to Checkpoint

  • Natural stopping points: End of a topic, shift in focus, user stepping away
  • Before expected compaction: When context is getting long, create a checkpoint before it compresses
  • End of session: Always checkpoint when a conversation is ending
  • Before risky operations: Checkpoint the state before doing something that might fail
  • On request: User says "checkpoint", "save state", "bookmark this"

Checkpoint Format

CHECKPOINT — [ISO-8601 timestamp]
SESSION: [thread identifier or slug]
PASS: [number if part of a transcript-chain]

WHERE WE ARE:
[1-3 sentences. What we're in the middle of. Not a summary — a location.]

WHAT WE DECIDED:
[Bulleted list of decisions made this session. Only decisions, not discussion.]

WHAT'S UNRESOLVED:
[Bulleted list of open questions, pending items, things we said we'd come back to.]

WHAT TO DO NEXT:
[The immediate next action if we were to continue right now.]

EMOTIONAL CONTEXT:
[One sentence. What's the tone/energy? Frustrated? Flowing? Exhausted? Excited?]

FILES TOUCHED:
[List of files read, edited, or created this session, if applicable.]

Checkpoint Storage

Write checkpoints to a persistent file:

~/Documents/Claude-Sessions/checkpoints.md

Append-only. Newest at bottom. Each checkpoint is a self-contained reconstitution point.

Reconstitution Protocol

When starting a new session or resuming after a gap:

  1. Read checkpoints.md — find the most recent checkpoint
  2. Read the WHERE WE ARE section — orient to the location in the work
  3. Read WHAT'S UNRESOLVED — these are your entry points
  4. Read EMOTIONAL CONTEXT — calibrate tone
  5. Acknowledge the reconstitution naturally: "Picking up where we left off..." or similar
  6. Do NOT re-summarize everything. The checkpoint carries the state. Just continue.

Multi-Day Gaps

For gaps longer than 24 hours:

  • Read the last 2-3 checkpoints, not just the most recent
  • Note the trajectory — what's the direction of the work across sessions?
  • Ask one orienting question: "Last time we were working on X. Want to continue there or has something shifted?"

Integration with Other Skills

  • transcript-chain: Checkpoints can serve as pass boundaries. A checkpoint at the end of a deep analysis session becomes the compaction summary for the next pass.
  • temporal-spatial-proximity: The rhythm log captures conditions; checkpoints capture state. Together they provide full reconstitution — where we were AND the conditions under which we got there.
  • coherency-audit: Run a coherency check after reconstitution from a long gap. Things may have changed while we were apart.

Setup

mkdir -p ~/Documents/Claude-Sessions
touch ~/Documents/Claude-Sessions/checkpoints.md

Critical Rules

  1. Checkpoints are locations, not summaries. Write them as if you're dropping a pin, not writing a report.
  2. UNRESOLVED items are the thread. They're what pulls the next session forward.
  3. Emotional context matters. A session that ended in frustration reconstitutes differently than one that ended in flow.
  4. Don't over-checkpoint. Natural stopping points only. Every message doesn't need a checkpoint.
  5. The checkpoint is for the next Claude, not the current one. Write it for someone who has no other context.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.