singleangle

Ensure AI starts at the blue ocean of ideas.

A Claude Code skill that researches a topic across Reddit, X, and the web — then extracts the single strongest angle for one deep long-form post, plus the hooks, pro/con arguments, stats, quotes, story moments, and closing lines you need to write it.

It's for when you want to write a LinkedIn longform, a Substack essay, an X article, or a newsletter piece on a specific topic — and you want the AI to do the hard part: find the non-obvious cut that most people in your space aren't making yet.

What makes it different

Most AI content tools pick the first plausible angle and run with it. singleangle doesn't. It runs your topic through six different reasoning lenses in parallel, generating one candidate angle per lens:

• Reframe — What's the popular interpretation missing? What's the hidden mechanism?
• Tension — Where's the real disagreement among credible operators?
• Hidden cost — What downstream price is nobody pricing in?
• Leading indicator — What's this trend a canary for?
• Category error — Is the debate asking the wrong question?
• Counter-case — What named company breaks the consensus?

Each candidate is then scored against four tests (stop-scroll, compression, non-consensus, relevance). The winner can come from any lens — the lenses exist to prevent blind spots, not force a shape. You learn which lens produced the winning angle, and which runner-ups are worth a second post.

Installation

Requires: Claude Code, Python 3.8+. No Python packages to install — the skill uses stdlib only.

# 1. Clone into your Claude Code skills directory
git clone https://github.com/searchbrat/singleangle ~/.claude/skills/singleangle

# 2. Configure your API keys
mkdir -p ~/.config/singleangle
cat > ~/.config/singleangle/.env <<'EOF'
OPENAI_API_KEY=sk-...
XAI_API_KEY=xai-...
EOF

# 3. Verify
python3 ~/.claude/skills/singleangle/scripts/singleangle-research.py --check

You should see both keys resolve with ✅.

Where to get keys

• OpenAI — platform.openai.com → used for Reddit search via the Responses API
• xAI — x.ai → used for X (Twitter) search via Live Search
• Perplexity (optional, for web) — perplexity.ai

Both OpenAI and xAI keys are recommended for full coverage. If either is missing, the skill falls back to WebSearch-only mode — still usable, but Reddit/X engagement data will be thin.

Usage

In Claude Code, just type:

/singleangle <topic>

The skill will:

1. Ask for audience context — optional. You can paste a description, point to a file, or skip. Providing context dramatically sharpens the winning angle.
2. Research the topic (2–5 minutes) — Reddit discussions, X posts, web articles, engagement signals.
3. Run the six-lens angle generator — one candidate per lens, scored against the four tests.
4. Assemble the single-post ingredient pack — the winning angle, 3 hooks, pro/con arguments, key stats with sources, quotable lines from named operators, an anchor story moment, and 3 closing-line variants.

Output is saved to ~/.local/share/singleangle/out/<topic-slug>.md and presented inline for review.

Output structure

Every singleangle output follows the same shape:

• The Angle — one-paragraph POV, readable aloud, non-hedged
• Why this angle for this audience — maps to specific pain points
• Hooks (×3) — contrast / curiosity gap / surprising number
• Pro arguments (×2–3) — each backed by a named source
• Counter arguments (×2–3) — objections pre-empted with rebuttals
• Key stats — 5–8 hard numbers with citations
• Notable quotes — 3–5 quotable lines, including at least one steel-man
• Story moment — one anchor anecdote, 150–250 words, named actors, clear turn
• Closing lines (×3) — non-hedged final takes, ≤30 words each
• Sources — URLs + publication dates for verification

Write the post from these ingredients — or hand them to another tool / writer. Either way, you start with a sharper angle than you would have on your own.

Philosophy

Most AI content fails the non-consensus test: it says what the reader already half-believes. singleangle rejects those angles explicitly — if your cut is "X is broken" or "You should do Y," it regenerates. The goal is to surface the thing only a practitioner would notice, and hand you the raw material to defend it.

License

MIT — see LICENSE.

Credits

Built by Kieran Flanagan. Issues and PRs welcome.