think

What it does

The skill provides a multi-mode protocol for strategic decision making, combining five modes: strategic analysis, validation council, executive council, deep analysis orchestration, and discovery mode. It is user-invocable and aims to help users analyze strategies, stress-test decisions, launch a debate council, or think through problems. It explicitly guides the agent through structured steps, including choosing 1-2 decision frameworks, presenting an initial hypothesis, running three parallel branches (support, challenge, lateral), performing constraint analysis across regulatory, technical, timeline, political, and budget axes, and synthesizing a final recommendation with a stated confidence level. It also prescribes a parallel validation council and executive council process, requiring sub-agent tasks to read a saved strategic analysis file and return JSON results with markdown outputs. Finally, it defines an orchestration flow for deep analysis with temporary journaling and a discovery mode that can constrain output to mirroring and questions only until the user says proceed.

How it works

• Before analysis, the agent must select 1-2 frameworks from the decision frameworks skill and declare the chosen framework and rationale.
• Layer 1 drafts an initial hypothesis by stating the intuitive answer and what a reasonable person would assume.
• Layer 2 (MANDATORY) runs three parallel branches: Support (evidence for hypothesis), Challenge (attack the hypothesis), and Lateral (alternative approaches). Each branch must have content or an explanation if nothing is found.
• Layer 3 (MANDATORY) performs constraint analysis across Regulatory/Compliance, Technical/Feasibility, Team Capacity/Timeline, Political/Organizational, and Budget/ROI, with a question for each.
• Layer 4: Synthesis combines all layers into a hardened recommendation, including a confidence level, key assumptions, and conditions under which the recommendation fails.
• The output format for the strategic analysis is a markdown block with sections for topic, BLUF, framework applied, initial hypothesis, supporting evidence, challenges and risks, alternative approaches, constraint analysis, and final recommendation with confidence and assumptions.
• Mode B (Validation council) and Mode C (Executive council) provide parallel, structured outputs focusing on weaknesses, objections from multiple personas (CFO, CTO, Competitor, Customer), and a synthesized verdict with next steps.
• Mode D (Deep analysis) orchestrates an end-to-end workflow, saving the strategic analysis to a journal file, then spawning two parallel sub-agents (Validation council and Executive council) to return JSON results, followed by a final synthesis.
• Mode E (Discovery mode) can lock the output to a no-solution state, offering four sections: Mirroring, Context mapping, Unknowns, and Probing questions, until the user enables proceeding.

When to use it

• Use when you need strategic analysis, adversarial validation, executive debate, or discovery for complex decisions. Trigger phrases include: "Analyze this strategy", "Stress-test this decision", "Launch the council", and "Help me think through [problem]".

What it can touch

• Tools declared: "claude-code" (used for code- or task-oriented actions within the agent’s workflows). It references file paths, markdown outputs, and JSON structures in its internal prompts for sub-agent orchestration.

Caveats

• The skill requires explicit framework selection and complete layering (1-4) for formal analysis. It relies on parallel sub-agents that read a saved analysis file rather than the main context. Outputs are formatted as markdown and JSON per mode; results may depend on the quality of input prompts and available frameworks.