| name | learn |
| description | Research a topic relevant to SWARM AI safety and grow the vault with findings. Queries external sources, synthesizes into claim candidates, follows the extraction pipeline. Triggers on "/learn", "/learn [topic]", "research this", "find papers on". |
| user-invocable | true |
| allowed-tools | Read, Write, Glob, Bash, mcp__qmd__vector_search |
| context | fork |
Runtime Configuration (Step 0)
Read vault/ops/derivation-manifest.md for research settings.
Default research depth: deep (AI safety domain, from config).
EXECUTE NOW
Target: $ARGUMENTS
- If target is a topic: research that topic using available tools
- If target is empty: ask "What topic should I research?"
Learn
Research a topic and grow the SWARM knowledge graph with findings.
Research Priority Areas
Topics most likely to yield governance claims:
- Multi-agent mechanism design
- Distributional AI safety
- Collusion detection algorithms
- Adversarial robustness in cooperative systems
- Welfare economics in AI systems
- Network topology effects on agent behavior
Workflow
- Formulate search queries — what specific aspect, what evidence would be decisive
- Research using available tools — web search, arXiv, semantic scholar
- Create inbox item — save research findings to
runs/research-[topic]-[date].md
- Run /extract on the file — process research into claim candidates
- Run /cross-link — connect new claims to existing vault
Research File Format (in runs/)
---
description: Research findings on [topic]
source_type: research
research_prompt: "[the query used]"
generated: YYYY-MM-DD
---
# [Topic] — research findings
[Synthesized findings from sources]
## Sources
- [Author et al. Year — title](url)
- ...
After creating the research file, run /extract [path-to-file] to process it into claims.
Quality Gates for Research-Sourced Claims
Research-sourced claims have lower initial confidence until validated by SWARM experiments:
- Research finding →
confidence: low until replicated in SWARM runs
- SWARM experiment confirms research →
confidence: medium or high based on criteria
- Document the chain: external source → research note → claim → SWARM validation
