agentleFS
Sign inStart free
asmamestasmamest/Nexus-Skills/skills/agentic-engineering/human-in-the-loop/SKILL.md

human-in-the-loop

Skill1 starsChanged 17 days agoMITView on GitHub

Design human approval and review checkpoints for agent actions, including risk-based policies, review payloads, pause/resume, auditability, and rejection handling.

We split this skill into the rules it gives an agent

7 rules. 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Put human judgment at decisions where risk, ambiguity, or accountability warrants it, without asking for approval on every harmless step.

    line 9
  2. Distinctive:

    Test rejection, edit, expiry, duplicate submissions, reviewer outage, state loss, and changed action arguments.

    line 21
  3. Distinctive:

    Do not use vague approval text for irreversible operations.

    line 27
  4. Distinctive:

    Do not treat silence or timeout as approval.

    line 28
  5. Distinctive:

    Use idempotency keys for approved side effects.

    line 32
  6. Distinctive:

    Avoid approving an entire future plan when later actions can materially change.

    line 34
  7. Distinctive:

    Keep emergency stop and cancellation paths available.

    line 35

“Common” is measured: kept by at least 6 in 10 of the 4 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

Human-in-the-Loop Agent Design

Goal

Put human judgment at decisions where risk, ambiguity, or accountability warrants it, without asking for approval on every harmless step.

Workflow

  1. Classify actions by impact: read-only, reversible write, external communication, financial/legal/production impact, destructive operation, or permission change.
  2. Define approval rules outside the model prompt. Tool code or policy middleware must enforce them.
  3. Prepare a review payload containing proposed action, normalized arguments, target, expected effect, evidence, risk explanation, and relevant diff. Redact secrets.
  4. Pause before execution, not after, for actions requiring approval.
  5. Bind approval to the exact action and arguments. If arguments or target change, request approval again.
  6. Authenticate the reviewer and check authorization, separation of duties, and approval expiry.
  7. Support approve, edit, reject, request-more-information, and timeout outcomes where appropriate.
  8. Persist enough state to resume safely. Revalidate permissions and external state after a long pause.
  9. Record reviewer, decision, timestamp, action hash/version, and outcome in an access-controlled audit log.
  10. Test rejection, edit, expiry, duplicate submissions, reviewer outage, state loss, and changed action arguments.

UX principles

  • Explain what will happen and why it needs approval.
  • Show the actual target and material parameters.
  • Make reject/cancel as clear as approve.
  • Do not use vague approval text for irreversible operations.
  • Do not treat silence or timeout as approval.

Implementation safeguards

  • Require server-side authorization at execution time.
  • Use idempotency keys for approved side effects.
  • Separate policy decision from model recommendation.
  • Avoid approving an entire future plan when later actions can materially change.
  • Keep emergency stop and cancellation paths available.

Deliverable

Provide a risk/action matrix, approval UX contract, persistence/resume behavior, audit schema, authorization checks, and failure tests.

References

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.

Check your setup, free