Peer Review

Multi-model peer review layer using local LLMs via Ollama to catch errors in cloud model output. Fan-out critiques to 2-3 local models, aggregate flags, synthesize consensus. Use when: validating trade analyses, reviewing agent output quality, testing local model accuracy, checking any high-stakes Claude output before publishing or acting on it. Don't use when: simple fact-checking (just search the web), tasks that don't benefit from multi-model consensus, time-critical decisions where 60s latency is unacceptable, reviewing trivial or low-stakes content. Negative examples: - "Check if this date is correct" → No. Just web search it. - "Review my grocery list" → No. Not worth multi-model inference. - "I need this answer in 5 seconds" → No. Peer review adds 30-60s latency. Edge cases: - Short text (<50 words) → Models may not find meaningful issues. Consider skipping. - Highly technical domain → Local models may lack domain knowledge. Weight flags lower. - Creative writing → Factual review doesn't apply well. Use only for logical consistency.

Key things to check before installing or using this skill: - Local model & resource requirements: It expects Ollama running locally with specific models (mistral:7b, tinyllama:1.1b, llama3.1:8b). Those models consume disk, memory, and will increase latency — confirm you have hardware and are comfortable hosting them locally. - Missing scripts and automation: The SKILL.md refers to bash scripts in workspace/scripts/, but no scripts are bundled. Ask the publisher for the scripts or inspect the intended implementation before running ad-hoc commands. - External posting and credentials: The workflow mentions posting results to Discord channels and exposing a Reef API endpoint, but the skill declares no environment variables or tokens. Clarify where reviews/logs are sent, what credentials are required, and ensure any outbound webhooks or bot tokens are provided and stored securely (not hard-coded). - Data exposure: This skill is designed to send cloud-model outputs to local models and potentially to external logs or chats. Do not run it on sensitive data until you confirm where the outputs and derived critiques are stored or transmitted and who has access. - Test with non-sensitive data: Validate behavior on harmless samples first (check where files are written, what network calls are made, whether Discord/HTTP endpoints are invoked). - Operational tuning: Understand consensus thresholds, storage location (experiments/peer-review-results), and retention policies for logged reviews and TPR metrics. If the publisher provides the missing scripts and a clear configuration for external endpoints (with explicit environment variables for tokens and endpoints), the skill can be much less ambiguous. Without that, the mismatch between described external behaviors and declared requirements is the main reason for caution.

Type: OpenClaw Skill Name: peer-review Version: 1.0.0 The skill instructs the OpenClaw agent to execute external shell scripts (`scripts/peer-review.sh`, `scripts/peer-review-batch.sh`, `scripts/seed-test-corpus.sh`) which are explicitly stated as 'not bundled in skill to avoid duplication' and therefore not provided for analysis. While the stated purpose of the skill (multi-model peer review using local LLMs via Ollama) is benign and plausible, the reliance on and instruction to execute unprovided external scripts introduces a significant supply chain risk and an unverified execution path. The `SKILL.md` also lists `curl` as a dependency, which, combined with shell execution, could facilitate network exfiltration if the unprovided scripts were malicious. There is no direct evidence of malicious intent or prompt injection within the provided `_meta.json` or `SKILL.md` files themselves, but the critical components of the skill's logic are external and unverified.