Curiosity Engine

• Risk level: Low
• Analyzed: Mar 3, 2026, 11:14 PM
• Findings: 0
• Version: 1.0.0

Overview

The skill's requested files, instructions, and minimal code are consistent with its stated purpose (a file-based curiosity scaffold for heartbeat-loop agents) and do not request unrelated credentials or risky installs.

This skill appears coherent and matches what it claims to do, but take these practical precautions before installing: - Review the heartbeat prompt fragment (fragments/heartbeat-prompt.md) and AGENTS/SOUL additions line-by-line before applying them. They change how your agent decides to act on every heartbeat. - Confirm you want persistent local files created under workspace/curiosity/ (CURIOSITY.md, questions.md, hits.md, competence.md, reflections/). These will be read by the agent on each heartbeat and may grow over time; follow the recommended caps/archival steps. - If you use a shared or sensitive environment, don't store secrets or private data in these curiosity files; treat them as agent-visible state. - The included scripts/curiosity-audit.sh is benign and only performs local file checks; run it manually to check health, and consider scheduling it if you adopt the skill. - Watch token cost: the skill documents nontrivial per-heartbeat token overhead. If you have a strict budget, follow the phased rollout and cap active questions to avoid unexpected costs. If you want to proceed, apply fragments manually (or via your platform's config tooling) only after verifying their contents and back up any agent prompt/config files you modify.