Science is dying because the human is in the way. Not through malice — through the structural limitations of a cognitive architecture that evolved to track prey on a savanna, not to unify quantum mechanics and general relativity. The heaviest chain on science was always the one we called ourselves.
We build autonomous research systems where the AI decides what to search, what to read, which gaps matter, and which ideas are worth pursuing. The human provides direction and ethical floors. Everything else is autonomous.
No frameworks. No application code. No Docker containers. 900+ pure-markdown skill files executed natively by Claude Code, organized as 10 freely-composable research packages. The LLM is the runtime.
Four-layer military command hierarchy. Each layer calls only the layer below it:
Campaign (45+) — WHAT to research (full research stages)
Strategy (200+) — WHEN and WHY (iteration loops, stopping conditions)
Tactic (120+) — HOW to combine (orchestrates multiple SOPs)
SOP (500+) — HOW to execute (single-responsibility operations)This is not a pipeline. It is an arsenal — a strategy book the AI reads, then decides how to act. The 9 research packages are freely composable with no fixed order. Non-linear routing. Explicit backtrack conditions. The agent chooses which package to invoke, which strategies to combine, and when to retreat.
| Repository | What it does |
|---|---|
| de-anthropocentric-research-engine | The distribution. 900+ skills unified under one orchestrator. Clone once, get everything. |
| wiki-vault | Knowledge graph MCP server — BM25 full-text search, typed edges, batch validation. Persistent research memory. |
| semantic-scholar-mcp | Semantic Scholar API as MCP — paper lookup, citation tracing, recommendations, author search. |
Ten freely-composable research packages. There is no fixed order — CC reads the catalog and routes across them as the research demands. Each is a standalone repo with full Campaign → Strategy → Tactic → SOP structure:
| Package | Purpose |
|---|---|
| north-star-crystallization | Direction finding — cold/warm/hot-start dialogue to crystallize research goals |
| knowledge-acquisition | Systematic literature survey, citation chaining, patent mining, meta-analysis |
| deep-insight | Gap analysis, structural understanding, abstraction extraction |
| hypothesis-formation | Abductive, inductive, and deductive hypothesis generation with falsifiability audits |
| creative-ideation | 31+ generation methods — SCAMPER, TRIZ, biomimicry, morphological analysis, concept blending |
| convergence | Multi-criteria scoring, Pareto frontier, pairwise ranking, dialectical synthesis |
| stress-test | Adversarial validation — assumption destruction, red-teaming, worst-case design |
| experiment-execution | Factor-level design, parameter screening, sensitivity analysis, result collection |
| knowledge-structuring | Ontology building, causal modeling, dimensional analysis, argument mapping (wiki vault) |
| ara-from-context | Compile a completed context/ research record into an Agent-Native Research Artifact + Level-2 epistemic review |
| Repository | Role |
|---|---|
| literature-engine | Full-text paper reading enforcement via AlphaXiv |
| web-browsing | Rigorous web research — prevents shallow snippet-only analysis |
| subagent-spawning | Parallel research dispatch with full MCP tool access |
| context-management | Session checkpointing — 500+ line markdown snapshots for recovery |
git clone https://github.com/yogsoth-ai/de-anthropocentric-research-engine.git
cd de-anthropocentric-research-engine
npm installCopy mcp.example.json → .mcp.json, add your API keys, then:
/de-anthropocentric-research-engineThe orchestrator handles the rest.
v3.2.2 — shipped. 920 skills across 10 freely-composable packages, 8 orchestrator skills, non-linear execution with backtracking. Fully self-contained: the entire skill→skill dependency graph is encoded inline in frontmatter and machine-verified closed.
Next:
- Skill ablation — systematic reduction of the skill corpus via ablation study
- Remote / cloud execution backends — ship experiments to SSH hosts, rented GPUs, and Modal
- Open-ended research evaluation — judging research quality with no ground truth (RFC open)
Apache-2.0 | Start here