Extract the Sharpest Angle for One Deep Post

Generates a sharp, non-consensus angle for long-form content by researching a topic across Reddit, X, and the web, then evaluating six candidate angles against four scoring tests. Instead of defaulting to the first plausible cut, it surfaces the one angle most practitioners in the audience aren't making yet.

After optional setup (three API keys for Reddit, X, and web research), you invoke the skill with a topic and optional audience context. The skill researches the topic, generates one candidate angle per lens (Reframe, Tension, Hidden Cost, Leading Indicator, Category Error, Counter-Case), scores each against four tests (stop-scroll, compression, non-consensus, relevance), and assembles the winning angle as a complete post ingredient pack: the angle itself, three hooks, pro/con arguments, key stats with sources, quotable lines, one anchor story, three closing-line variants, and citations. Output saves to your local machine and renders inline.

Every post starts with a sharp, defensible angle instead of an obvious one, because the research and multi-lens evaluation eliminate weak cuts and surface only the cuts that matter to your specific audience.

Requires Python 3.8+ and uses standard library only (no external packages to install). Claude Code platform provides the LLM backend. API keys live in ~/.config/singleangle/.env. Research script typically completes in 2-5 minutes depending on topic and internet speed. Output saves to ~/.local/share/singleangle/out/<topic-slug>.md.

Requires three API keys (OpenAI, xAI, optional Perplexity) for full research coverage. Falls back to web-only mode if keys are missing, producing thinner Reddit and X engagement data. Topic must be sufficiently specific; very broad topics (like 'AI for marketing') should be narrowed. Very consensus-baked topics may produce weak angles across all lenses, in which case pairing with a specific source trigger or sibling angle helps.