agentsclimarketplace

Review gate

Skill prasadmogulothu/agent-skills/review-gate

Reusable, token-optimized agent skills for autonomous development with Hermes, Claude Code & OpenCode. Multi-model orchestration patterns — context-checkpointing, human review gates, RLS security, GDPR consent, AI cost control, and test-fix loops. Framework-agnostic, agentskills.io compatible.

Install
npx -y skills add prasadmogulothu/agent-skills --skill review-gate

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when a task is marked as needing human review. Pause, present the artifact to a human (summary + link), and wait for explicit approval before continuing. Loop on change requests; never proceed on silence.

SKILL.md

1.7 KB, as published. Nobody here has run it

review-gate

Human-in-the-loop approval as a reusable step. Mark certain tasks as gated; at those, a human must approve before dependent work continues.

Steps

  1. Confirm the task passes your done-checks.
  2. Set the task Status: BLOCKED_REVIEW in tasks.md and add it to a Review Queue.
  3. Notify the human (chat / messaging gateway / PR comment) with:
    • What — a 3-bullet summary of what was built or changed.
    • Where — a clickable artifact: a git diff/PR, a preview URL, a component catalogue, or a report.
    • Ask — "Approve, or request changes?"
  4. Wait. Do not start any task that depends on this one. You may pick up an independent, non-dependent task meanwhile to keep momentum.
  5. Never proceed on silence. Only an explicit approval advances the gate.

Change loop

On a change request: set Status: CHANGES_REQUESTED, record the request, apply the fix with the same-tier model, re-run done-checks, re-present, and ask again. Loop until approved, then set DONE and clear it from the Review Queue.

Recommended always-gated work

Anything risky or irreversible: data-model/security changes, auth, payments, data deletion, and public deploys — gate these even when automated checks pass.

Token tactics

Near-zero model cost — runs in the agent's control flow. Because it pauses the workers while a human reviews, models sit idle (no spend) until approval. Oversight that also saves tokens.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.