agentleFS
hwl668hwl668/Agentcheckpoint/skills/agent-checkpoint/SKILL.md

agent-checkpoint

Skill2 starsChanged 42 days agoMITView on GitHub

Save verifiable, restorable checkpoints of coding-agent work with the agentck CLI. Use when completing a unit of work, before a risky change, when a milestone is reached, or when handing work over to another agent (Claude, Codex, GLM, ...). Also use when resuming work that a previous agent checkpointed.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. ainullm/agent-checkpoint★ 3same name
    2. NexaiGuy/nex-ai-skills★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After completing a coherent unit of work (a fix, a refactor step).

    line 17
  2. Distinctive:

    When you are about to hand work to another agent or end your session.

    line 19
  3. Distinctive:

    When the user asks to "checkpoint", "save progress", or "hand off".

    line 20
  4. Very specific:

    Git state is collected for you: Never describe the repo state in claims — agentck records HEAD, diffs, and untracked files itself.

    line 42
  5. Very specific:

    Evidence beats claims: If you say work is complete, back it with --test CMD so the exit code is recorded as machine-observed fact.

    line 44
  6. Very specific:

    Treat [OBSERVED] and [VERIFIED] as ground truth.

    line 62
  7. Very specific:

    Treat [AGENT-REPORTED] as claims: re-verify anything you build on.

    line 63
  8. Very specific:

    If verify reports drift or stale tests, tell the user before continuing.

    line 65

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

AgentCheckpoint skill

AgentCheckpoint (agentck) freezes the real working state of an agent session and proves to the next agent whether that state is still valid. Its core rule binds YOU:

Machine-observed facts and agent-reported claims are stored separately. Never present a claim as a verified fact.

When to checkpoint

  • After completing a coherent unit of work (a fix, a refactor step).
  • Immediately BEFORE a risky change, so it can be restored.
  • When you are about to hand work to another agent or end your session.
  • When the user asks to "checkpoint", "save progress", or "hand off".

How to save

Requirements and decisions go through flags — they are stored as agent-reported claims, never as facts:

agentck save <short-label> \
    --agent glm \
    --goal "What this work is trying to achieve" \
    --completed "One concrete claim per flag" \
    --decision "Why you chose this approach" \
    --constraint "Any constraint you must respect" \
    --blocker "What is currently blocking you, if anything" \
    --next "The single most useful next action" \
    --do-not "Files or areas that must not be touched" \
    --test "pytest tests/test_cache.py"

Rules:

  1. Git state is collected for you. Never describe the repo state in claims — agentck records HEAD, diffs, and untracked files itself.
  2. Evidence beats claims. If you say work is complete, back it with --test CMD so the exit code is recorded as machine-observed fact. Do not write "tests pass" as a claim unless you ran them via --test or in the session.
  3. --completed items must be specific and falsifiable ("switched eviction policy to LRU in src/cache.py"), not vague ("made some changes").
  4. For longer context, write the semantic JSON (fields: agent, goal, status, completed, decisions, constraints, blockers, next_steps, do_not, tests — unknown keys are rejected) and run agentck save <label> --semantic file.json.
  5. Forking two lines of work from one point: agentck save <label> --parent <ref>.

How to resume work another agent checkpointed

agentck verify          # drift, staleness, integrity, resume confidence
agentck resume codex    # or claude / glm / plain — evidence-tagged handoff
  • Treat [OBSERVED] and [VERIFIED] as ground truth.
  • Treat [AGENT-REPORTED] as claims: re-verify anything you build on.
  • Respect the "Do not touch" list.
  • If verify reports drift or stale tests, tell the user before continuing.

Restore (ask first)

agentck restore <ref> restores into a NEW git worktree — safe, non-destructive. agentck restore <ref> --in-place --yes rewrites the current tree (a safety checkpoint is saved first). Only run in-place restore when the user explicitly asked to reset their working tree.

Useful reads

agentck list      # all checkpoints
agentck log       # checkpoint DAG as a tree (forks visible)
agentck show latest
agentck diff <ref-a> <ref-b>
agentck replay    # re-run the recorded --test commands, compare results

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.