beam-ai-team/deep-thinking

Deep thinking

One entry point for rigorous thinking: frame the problem, then choose stress-test, framework, or expert feedback, and synthesize.

Safety Contract

This skill is analytical by default. Before saving analysis, decision records, or expert-review outputs, show the destination path, source materials, and whether confidential strategy or personal context is included. Require explicit user approval in the current turn before writing or sharing files. In-chat framing, critique, and synthesis do not require approval.

Purpose

Hard problems benefit from (1) a clear frame — one-sentence problem, binding constraint, eigenquestion — and (2) a structured lens: multi-perspective stress-test, a mental model, or expert personas. This skill does both: it frames first, then runs the chosen path and synthesizes.

Use when: Messy context, important decision, or content you want to validate before shipping.

Time: 3–8 minutes depending on path.

---

Workflow

Step 0: Frame (always)

Before any analysis, produce:

1. One-sentence problem — The core issue in a single sentence.
2. Binding constraint — The limiting factor that shapes what’s possible.
3. Eigenquestion — The question whose answer determines the answers to the other questions.

Output these clearly so the user (and the rest of the workflow) can stay aligned.

---

Step 1: Choose path

Offer the user (or infer from their request):

• Stress-test — Multi-perspective critique (recipient POV, critic, devil’s advocate, first principles, risk). Use the ultrathink workflow: run the same five-perspective dialogue and synthesis as in the ultrathink skill. Load skills/general/ultrathink/SKILL.md and follow its Step 2 and Step 3.
• Framework — Apply a structured model (pre-mortem, decision matrix, inversion, etc.). Run mental-models: uv run beam-next-mental-models --format brief, pick 1–2 models that fit the frame, load the model file from 00-system/mental-models/models/{category}/{slug}.md, and apply it.
• Expert feedback — Get persona-based review. Run expert-review: detect work type, pick 2–3 personas, load persona files from 00-system/expert-personas/, and produce verdict + strengths + gaps + one thing to change per persona.

If the user said "ultrathink" or "stress test", default to Stress-test. If they said "apply a framework" or "mental model", default to Framework. If they said "expert feedback" or "what would X think", default to Expert feedback. Otherwise offer the three options and let them pick.

---

Step 2: Run the path

• Stress-test: Execute the ultrathink multi-perspective dialogue and synthesis (see ultrathink SKILL.md).
• Framework: Execute the chosen mental model’s process (questions, steps) and summarize.
• Expert feedback: Execute expert-review (personas, verdicts, recommendations).

---

Step 3: Synthesize

After the path, produce a short synthesis:

• How the frame (eigenquestion, constraint) connects to the findings.
• Top 2–3 actionable takeaways.
• Optional: "Want to run another path? (e.g. stress-test after framework)"

---

Triggers

• "deep think", "think deeply", "frame this", "what’s the real question", "eigenquestion"
• "ultrathink", "stress test this", "challenge this" → default to Stress-test
• "apply a framework", "mental model", "pre-mortem" → default to Framework
• "expert feedback", "what would [persona] think" → default to Expert feedback

---

Relationship to other skills

• ultrathink — Implements the stress-test path; can be loaded directly for "ultrathink this" or invoked from deep-thinking.
• mental-models — Provides the framework path; use its scanner and model files.
• expert-review — Provides the expert path; use its persona selection and review format.

Deep-thinking is the umbrella; those three remain available for direct use when the user wants only one path.