lokicik

Novel Idea Hunter — Development skill for Claude Code

Development community

Evidence-first, anti-slop opportunity discovery skill for coding agents (Claude Code + Codex).

How to install Novel Idea Hunter

This entry records only its repository, not the path inside it, so there is no exact command to give. Open lokicik/novel-idea-hunter and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Novel Idea Hunter does

Evidence-first, anti-slop opportunity discovery skill for coding agents (Claude Code + Codex).

Alternatives in Development

Agent Skills Platform — Build tested agent skills and govern their lifecycle through a user-defined marketplace: evidence, discovery 2.4k ★
Agent Skill Creator — Build tested agent skills and govern their lifecycle through a user-defined marketplace: evidence, discovery 2.3k ★
Projectmem — Open-source coding agent memory 766 ★

README

novel-idea-hunter

An evidence-first opportunity-discovery skill for coding agents (Claude Code and Codex). It finds novel, defensible product/startup/project opportunities by forcing research before ideation, and mechanically rejects generic "AI-powered X" slop with deterministic validator scripts.

Why it exists

Almost all LLM ideation slop comes from one failure: the model enters solution-mode before doing any research. This skill inverts the order:

brief → isolated observation branches → evidence normalization
      → anomaly map → facet recombination → deterministic slop gate
      → adversarial prior-art prosecution → market/incumbent attack
      → proprietary-edge verification → ≤3 non-redundant survivors
      → falsification experiments

Design lineage: isolated divergent branches (ADHD), capability-wedge and kill-filter thinking (HypoKiln), facet recombination (Scideator), retrieve-and-compare novelty checking (Idea Novelty Checker), plus quality-diversity archive dynamics and private-edge mining.

Three properties do most of the anti-slop work:

No ideas before evidence. Candidates must cite ≥3 sourced observations from ≥2 research lenses, or the lint rejects them.
Novelty is prosecuted, not scored. A separate adversarial pass tries to prove each idea already exists; verdicts come from a controlled vocabulary whose ceiling is no-close-prior-art-found — never "unique".
The gates are scripts, not vibes. lint_candidate.py and check_portfolio.py fail with exit code 1; a model cannot charm a regex.

Layout

novel-idea-hunter/
├── SKILL.md                 # the workflow (phases 0-10)
├── agents/openai.yaml       # Codex metadata (explicit invocation only)
├── references/              # lenses, facets, evidence rules, slop patterns,
│                            # novelty rubric, kill criteria, edge rules, report format
├── scripts/                 # Python 3 stdlib only
│   ├── init_run.py          #