{"skill":{"slug":"tommy-curiosity-engine","displayName":"Curiosity Engine","summary":"Implements a persistent curiosity system for stateless agents with question queues, breadcrumbs, engagement logs, competence maps, and reflections.","description":"# Curiosity Engine — SKILL.md\n\nA behavioral engagement scaffold that produces curiosity-equivalent behavior in stateless heartbeat-loop agents.\n\n## What This Does\n\nGives your agent a persistent curiosity system across cold-start sessions:\n- **Question queue** — externalized information gaps (Loewenstein) that carry across heartbeats\n- **Breadcrumb pattern** — notice interesting things during task work, investigate later\n- **Engagement log** — track what produces valued outcomes (extrinsic signals primary)\n- **Competence map** — track developing expertise with evidence, including failure processing\n- **Reflections** — post-exploration write-ups that create genuine learning artifacts\n\nThis is NOT synthetic consciousness or prosthetic dopamine. It's a behavioral activation protocol (Martell et al., 2010) adapted for AI agents — the same clinical framework used to treat motivation disorders in humans. Chinese Room design: if the outputs are indistinguishable from genuine curiosity, the philosophical question is irrelevant.\n\n## Installation\n\n### 1. Create curiosity directory\n```bash\nmkdir -p workspace/curiosity/reflections workspace/curiosity/archive\n```\n\n### 2. Copy templates\n```bash\ncp templates/CURIOSITY.md workspace/curiosity/CURIOSITY.md\ncp templates/questions.md workspace/curiosity/questions.md\ncp templates/hits.md workspace/curiosity/hits.md\ncp templates/competence.md workspace/curiosity/competence.md\n```\n\n### 3. Wire the three integration triggers\n\n**Trigger 1 — Heartbeat prompt (CRITICAL — most reliable):**\n\nAdd the contents of `fragments/heartbeat-prompt.md` to your agent's heartbeat prompt configuration. For OpenClaw, patch via `gateway config.patch`. This is the primary mechanism — if it's not in the prompt, curiosity doesn't exist for isolated sessions.\n\n**Trigger 2 — AGENTS.md (backup):**\n\nAdd the contents of `fragments/agents-addition.md` to your workspace AGENTS.md \"Every Session\" block.\n\n**Trigger 3 — SOUL.md (identity anchor):**\n\nAdd the contents of `fragments/soul-addition.md` to your SOUL.md.\n\n### 4. (Optional) Set up health monitoring\n```bash\nchmod +x scripts/curiosity-audit.sh\n# Run periodically or during maintenance heartbeats\n./scripts/curiosity-audit.sh workspace/curiosity/\n```\n\n## Phased Rollout (Recommended)\n\nDon't activate everything at once. Measure each phase before adding the next.\n\n**Phase 1** (start here): `questions.md` + heartbeat prompt + breadcrumbs\n- Run for 2 weeks. Measure: question quality, follow-through rate, breadcrumb frequency.\n\n**Phase 2** (add if Phase 1 improves output): `CURIOSITY.md` + `hits.md` + avoidance gate\n\n**Phase 3** (add if Phase 2 improves output): `competence.md` + reflections + anti-pattern detection\n\n## File Reference\n\n| File | Purpose | Loaded When | Token Budget |\n|------|---------|-------------|-------------|\n| `CURIOSITY.md` | Current intellectual state | Every heartbeat (Phase 2+) | ≤1500 tokens |\n| `questions.md` | Prioritized question queue | Every heartbeat | ≤2000 tokens |\n| `hits.md` | Engagement/reward tracking | Curiosity-mode heartbeats | ≤1500 tokens |\n| `competence.md` | Developing expertise | Curiosity-mode heartbeats | ≤1000 tokens |\n| `reflections/` | Post-exploration write-ups | On-demand | As needed |\n| `archive/` | Cold storage | On-demand search | As needed |\n\n## Key Design Principles\n\n1. **Behavioral outcomes over file quality.** Success = better outputs, not prettier files.\n2. **Extrinsic signals are primary.** Self-assessed \"aha moments\" are secondary and must include falsifiable claims.\n3. **Hard limits on scale.** 30 active questions max. 3 mulling items max. 30-day hit window. Enforce ruthlessly.\n4. **Avoidance detection.** Before every exploration: \"Am I avoiding a task right now?\"\n5. **The human is both subject and source.** Be curious about your human. Ask them questions. They know things the internet doesn't.\n6. **Breadcrumbs are the gold standard.** Questions dropped during non-curiosity work — unprompted — are the strongest signal the system is working.\n\n## Theoretical Grounding\n\nSee `references/theory.md` for condensed citations. Full research: check the SPEC.md appendix for the complete 46KB research document reference.\n\nCore frameworks: Loewenstein (1994) information gaps, Berlyne (1960) diversive/epistemic modes, Deci & Ryan (1985) SDT competence need, Martell et al. (2010) behavioral activation, Voyager (Wang et al., 2023) automatic curriculum, Reflexion (Shinn et al., 2023) verbal learning.\n\n## Anti-Patterns\n\nSee `references/anti-patterns.md` for the full list. The top three:\n1. **Curiosity as avoidance** — exploring interesting things to dodge boring tasks\n2. **Performance curiosity** — optimizing for human praise rather than genuine learning\n3. **Noisy TV** — consuming information endlessly without producing understanding\n\n## Token Budget\n\nThe curiosity engine adds ~1500-2500 tokens per heartbeat (Phase 1). See `references/token-budget.md` for detailed measurements and optimization guidance.\n\n**Quick guide:**\n- **Unlimited budget** → load everything, no restrictions\n- **Moderate ($2-10/day)** → Phase 1 only, cap questions at 15\n- **Tight (<$2/day)** → reduce heartbeat frequency, minimal questions\n- **Strict (<$0.50/day)** → may not be cost-effective; consider disabling\n","tags":{"latest":"1.0.0"},"stats":{"comments":0,"downloads":197,"installsAllTime":7,"installsCurrent":0,"stars":0,"versions":1},"createdAt":1772579651351,"updatedAt":1778491707669},"latestVersion":{"version":"1.0.0","createdAt":1772579651351,"changelog":"Initial release: behavioral engagement scaffold for stateless heartbeat-loop agents. Question queue, breadcrumbs, competence tracking, heartbeat/SOUL/AGENTS fragments.","license":null},"metadata":null,"owner":{"handle":"keylimesoda","userId":"s17f7pbrm7y4kfe2hvyj54nmj983fjv7","displayName":"Ric Lewis","image":"https://avatars.githubusercontent.com/u/7773610?v=4"},"moderation":{"isSuspicious":false,"isMalwareBlocked":false,"verdict":"clean","reasonCodes":["review.llm_review"],"summary":"Review: review.llm_review","engineVersion":"v2.4.24","updatedAt":1780089751368}}