-
Notifications
You must be signed in to change notification settings - Fork 12
FAQ
Q: Does this require a specific LLM provider?
No. Edit config/agent-config.json to use any OpenAI-compatible or Anthropic API. The default uses 9router/opencode-free, but every agent's model field can be changed independently.
Q: Can I use my own API keys?
Yes. Copy .env.example to .env and add your Semantic Scholar and CrossRef keys. The Python tools and the Literature Scout agent use these for API calls.
Q: Do I need OpenCode to run this?
Yes. The agents are designed for the OpenCode agent platform. The install.sh script symlinks agent files into OpenCode's agents/ directory. The Python CLI tools (literature_client.py, citation_verifier.py) work standalone.
Q: How long does a paper take?
The full pipeline takes 30 minutes to 4 hours depending on LLM speed, literature volume, and number of revision rounds.
Q: Can I add my own template?
Yes. See the Template-Guide for detailed instructions.
Q: The generated text sounds too AI-like. What can I do?
Provide a stronger voice sample (2-3 paragraphs of your published writing) in data/voice-profile/. The 41 Humanizer patterns are enforced as generation-time constraints, but the voice profile anchors the style. Without a voice sample, the Writer uses a generic academic voice.
Q: What happens if no novelty engine produces a useful hypothesis?
The system reports "no novel angle found" and the paper is not written. This is by design — the system refuses to produce incremental or derivative work.
Q: Are the novelty engines actually novel?
The engines are designed to think from angles no human researcher would naturally occupy. The Heretic generates 50 hypotheses from title alone. The Cross-Pollinator imports solutions from 15 fields the target field has never touched. The Counterfactual Generator rewrites history. These are cognitive frames, not prompt tricks.
Q: How reliable is the citation verification?
Every citation is checked against 2+ independent sources (Semantic Scholar AND CrossRef). If both sources agree on the paper's existence and title, it's verified. If only one source finds it, it's flagged as "weak verification." If neither source finds it, the paper is blocked — the pipeline will not produce a paper with hallucinated citations.
Q: How strict is the AI-pattern detection?
Very strict. The Style Auditor scans every sentence of the completed paper against all 41 Humanizer patterns. Pattern density must be < 1 per 2000 words. Em dashes count must be ZERO. One em dash = automatic failure. The pipeline will not proceed past Style Audit until the density threshold is met.
Q: What does "all 10 reviewers must pass" mean?
Each of the 10 reviewer personas (Theorist, Empiricist, Pragmatist, Skeptic, Historian, Methodologist, Ethicist, Competitor, Student, Dreamer) must recommend acceptance. If even one recommends rejection or major revision, the paper goes back to the Writer with that reviewer's annotations. The cycle repeats until all 10 accept.
Q: How do I run the tests?
python -m pytest tests/ -vQ: How do I lint the Python tools?
flake8 tools/ --max-line-length=100Q: How do I add a new agent?
See the Adding-a-New-Agent guide.
Q: How do I contribute?
See CONTRIBUTING.md for the full contribution guide. Look for issues labeled good-first-issue for beginner-friendly tasks.
Sisyphus Academica — MIT License