How do I install the Autonomous Research Loop skill?

Run npx skillmds@latest add knuckles-team/autonomous-research-loop in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Autonomous Research Loop safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Autonomous Research Loop?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Autonomous Research Loop free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.

Who published Autonomous Research Loop?

Knuckles-Team (@knuckles-team) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Autonomous Research Loop by knuckles-team

Name: Autonomous Research Loop
Author: Knuckles-Team

Autonomous Research Loop

Runs one propose-only self-evolution "golden loop" cycle over the Knowledge Graph. Use when the user wants to "run the research loop", "evolve the knowledge graph", "find and address open topics", "propose specs/teams from the KG", or "self-improve the codebase". Intake of unresolved topics → acquire related sources → ADDRESSES-resolve (converge) → optional distil spec DRAFTS into .specify/ → synthesize a team PROPOSAL. Never auto-merges or executes anything — every artifact is a draft/proposal a human reviews.

File contents

Autonomous Research Loop (propose-only)

Drives the KG's self-evolution golden loop — one cycle of:

Intake — unresolved Concept topics (no ADDRESSED_BY edge).
Acquire — semantically related sources per topic (vector search over the KG; optional external X/SearXNG/scholarx when KG_RESEARCH_EXTERNAL=1).
Resolve — write ADDRESSES / ADDRESSED_BY edges so the loop converges (addressed topics stop re-surfacing; new open topics rise).
Distil (optional, KG_GOLDEN_DISTILL=1) — SpecDraft markdown written to .specify/specs/kg-distilled/ (DRAFTS only).
Synthesize — a TeamSpec/AgentSpec proposal addressing the topics, persisted to the KG (not executed).

Propose-only guarantee: no code execution, no PR merge, no edits outside .specify/ drafts and KG proposal nodes.

Run it

On-demand via the graph-os MCP orchestration tool:

graph_orchestrate(action="golden_loop", max_topics=5)

…or directly:

KG_DAEMON_ROLE=client python -c \
  "from agent_utilities.knowledge_graph.research.golden_loop import run_golden_loop_cycle; \
   import json; print(json.dumps(run_golden_loop_cycle(max_topics=5), default=str, indent=2))"

Always-on daemon (optional)

Set KG_GOLDEN_LOOP=1 (interval KG_GOLDEN_LOOP_INTERVAL, default 3600s) on the host daemon (gateway / graph-os-daemon) to run the cycle continuously and throttled behind the foreground gate. Off by default (autonomous LLM work is opt-in). Still propose-only.

Output

A JSON report: topics_intake, topics_resolved, sources_linked, spec_drafts (paths), team (lead + members + persisted node/edge counts), errors.

Prerequisites

Embeddings backfilled (the host daemon's embed_backfill thread) so acquire's vector search has substrate.
A reachable graph backend (tiered: epistemic L1 + pggraph L3).

---name: autonomous-research-loopdescription: Runs one propose-only self-evolution "golden loop" cycle over the Knowledge Graph. Use when the user wants to "run the research loop", "evolve the knowledge graph", "find and address open topics", "propose specs/teams from the KG", or "self-improve the codebase". Intake of unresolved topics → acquire related sources → ADDRESSES-resolve (converge) → optional distil spec DRAFTS into .specify/ → synthesize a team PROPOSAL. Never auto-merges or executes anything — every artifact is a draft/proposal a human reviews.license: MIT---# Autonomous Research Loop (propose-only)Drives the KG's self-evolution **golden loop** — one cycle of:1. **Intake** — unresolved `Concept` topics (no `ADDRESSED_BY` edge).2. **Acquire** — semantically related sources per topic (vector search over the KG; optional external X/SearXNG/scholarx when `KG_RESEARCH_EXTERNAL=1`).3. **Resolve** — write `ADDRESSES` / `ADDRESSED_BY` edges so the loop **converges** (addressed topics stop re-surfacing; new open topics rise).4. **Distil** *(optional, `KG_GOLDEN_DISTILL=1`)* — `SpecDraft` markdown written to `.specify/specs/kg-distilled/` (DRAFTS only).5. **Synthesize** — a `TeamSpec`/`AgentSpec` **proposal** addressing the topics, persisted to the KG (not executed).**Propose-only guarantee:** no code execution, no PR merge, no edits outside`.specify/` drafts and KG proposal nodes.## Run itOn-demand via the graph-os MCP orchestration tool:```textgraph_orchestrate(action="golden_loop", max_topics=5)```…or directly:```bashKG_DAEMON_ROLE=client python -c \ "from agent_utilities.knowledge_graph.research.golden_loop import run_golden_loop_cycle; \ import json; print(json.dumps(run_golden_loop_cycle(max_topics=5), default=str, indent=2))"```## Always-on daemon (optional)Set `KG_GOLDEN_LOOP=1` (interval `KG_GOLDEN_LOOP_INTERVAL`, default 3600s) on the**host daemon** (gateway / `graph-os-daemon`) to run the cycle continuously andthrottled behind the foreground gate. Off by default (autonomous LLM work isopt-in). Still propose-only.## OutputA JSON report: `topics_intake`, `topics_resolved`, `sources_linked`,`spec_drafts` (paths), `team` (lead + members + persisted node/edge counts),`errors`.## Prerequisites- Embeddings backfilled (the host daemon's `embed_backfill` thread) so acquire's vector search has substrate.- A reachable graph backend (tiered: epistemic L1 + pggraph L3).

Frequently asked questions