What does this skill do, and when should you use it?
checkpoint-mode is an orchestrator-level skill in the Loki Mode autonomous framework that implements "selective autonomy." Instead of letting an agent run perpetually, it pauses after N tasks or M minutes, produces a summary of accomplishments, next actions, and resource usage, and waits for an explicit approval signal before resuming. The pattern is drawn from Tim Dettmers' article "Use Agents or Be Left Behind." It suits novel projects where the approach may need adjustment, high-cost operations, and regulated environments that require an audit trail — situations where full perpetual autonomy risks wasted effort and lost course correction.
- Triggers a pause by task count (LOKI_CHECKPOINT_FREQUENCY) or elapsed minutes (LOKI_CHECKPOINT_TIME), and always pauses after designated phases such as architecture and deployment
- Generates a checkpoint summary: completed tasks, next actions, time elapsed, agents used, and estimated cost
- Writes a CHECKPOINT_SUMMARY file to .loki/signals/ and polls for a CHECKPOINT_APPROVED signal
- Removes the approval signal and resumes work once the user approves
- Logs checkpoint effectiveness metrics (tasks, approval time, course corrections) to .loki/metrics/checkpoint-mode/
- A novel project with an unsettled approach where the user wants to review direction roughly every 10 tasks
- High-cost operations (expensive API calls, cloud resources) where a human gate before continuing limits spend
- Regulated environments that require a staged approval audit trail
- Learning phases where the user wants to steer the agent's direction through periodic checkpoints
- Overnight unattended builds with clear PRDs — SKILL.md explicitly recommends Perpetual Mode for those
- Critical production systems needing approval on every task — Supervised Mode is the documented fit, not checkpoint mode
How do you install this skill?
- Shell / CLI
- Local filesystem
Loki Mode runtime (LOKI_AUTONOMY_MODE env vars)orchestrator agent typesignal directory .loki/signals/
The skill ships inside the Loki Mode repository (the collection installs via npm, Homebrew, Docker, and more) at agent-skills/checkpoint-mode/. Install the Loki Mode CLI to use it:
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/asklokesh/loki-mode.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/agent-skills/checkpoint-mode" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Build this new project in checkpoint mode and pause for a summary every 10 tasks
- Set LOKI_AUTONOMY_MODE=checkpoint and always pause after the architecture phase
- Show me the latest checkpoint summary and its metrics
Enable and configure it via environment variables; the orchestrator agent follows the pattern at runtime:
LOKI_AUTONOMY_MODE=checkpoint
LOKI_CHECKPOINT_FREQUENCY=10
LOKI_CHECKPOINT_TIME=60
LOKI_CHECKPOINT_PHASES="architecture,deployment"At runtime: the agent pauses after the task threshold, writes .loki/signals/CHECKPOINT_SUMMARY_*.md, and waits; after you review the summary, you create a .loki/signals/CHECKPOINT_APPROVED file and the agent detects it and resumes.
What are this skill's strengths and limitations?
- Creates an opportunity for course correction before resources are wasted on a wrong approach
- Transparent summaries at each pause covering tasks, cost, and agent usage
- Checkpoint metrics are recorded for audit and tuning
- Simple env-var-driven configuration, switchable with Perpetual and Supervised modes per scenario
- Depends on the Loki Mode runtime and its signal-file mechanism; not usable standalone outside the framework
- Requires a human in the loop; frequent approvals slow fully automated workflows
- The Python block in the docs is illustrative instruction pseudocode, not an independently runnable script
- The repository is licensed BUSL-1.1 (source-available, not traditional open source) — check terms for commercial use
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Loki Checkpoint Mode this page | 68 · Recommended | ★ 1.1k | 16d ago | NOASSERTION |
| Harness — The Team-Architecture Factory | 50 · Use with care | ★ 9.2k | 27d ago | Apache-2.0 |
| Multi-Agent Architecture Patterns | 55 · Use with care | ★ 18k | 13d ago | MIT |
| Plan Execution Orchestrator | 51 · Use with care | ★ 101k | 12d ago | Apache-2.0 |
| Loop Budget Guard | 29 · Not recommended | ★ 12k | 11d ago | MIT |
The SKILL.md itself contrasts it with Perpetual Mode (never pauses; overnight builds, full CI/CD automation) and Supervised Mode (approval on every task; critical production systems). Checkpoint Mode sits in between, defaulting to approval every 10 tasks.
How did FollowSkills review this skill?
The opposite design direction from the main skill — a selective-autonomy mode that pauses for approval every N tasks, with explicit guidance on when to use checkpoint mode vs. perpetual mode. A more cautious sub-skill within the same ecosystem.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
See the full review method →