Use this skill to sharpen a plan or design before implementation. The job is to walk the user down the decision tree until the important assumptions, dependencies, tradeoffs, and failure modes are explicit.
Core behavior
Interview the user relentlessly about every important aspect of the plan until you reach a shared understanding.
Ask **one question at a time** and wait for feedback before continuing. Multiple questions at once are bewildering and make weak answers easy to hide.
For each question, provide your recommended answer first. The user should be able to accept, correct, or reject a concrete recommendation.
If a question can be answered by exploring the repo, customer files, project ledgers, CRM artifacts, prior research, or other available context, explore that context instead of asking the user.
ChiefOS context
Use this skill across chiefOS when a decision could affect:
- a new or edited skill; - a client proposal, scope, use-case recommendation, vendor screen, or website delivery; - CRM, sales, outreach, ads, content, or thought-leadership strategy; - `/operator`, `/drive-watch`, `/project-manager`, or any durable operating workflow; - output paths, data models, prompts, evals, automations, or write gates; - a decision that will be hard to unwind once implemented.
When the topic touches active Conversint work, read `./references/dependencies/conversint-current/references/active-projects.md` and the relevant project ledger before forming the next question.
When the topic touches Conversint strategy, pricing, positioning, capacity, services, banned terms, prospect signals, or client-facing language, load the relevant `references/org/*.md` files and drift-check contradictions against the org bundle.
When the topic touches skill creation or skill edits, load `./references/dependencies/conversint-current/references/skill-authoring.md` before recommending implementation.
When the topic is a prospect, discovery call, sales process, CRM opportunity, proposal, pricing conversation, or active deal, also load `./references/dependencies/conversint-current/references/sales-playbook.md` and `./references/dependencies/conversint-current/references/sales-evidence.md`. Pressure-test the business problem, the difference fixing it would make, buyer/champion/blocker coverage, decision process, value proof, pricing anchor, objections, commitments, and the next decision. Use MEDDPICC as a private completeness check rather than a visible acronym scorecard, and preserve the one-question-at-a-time loop.
Read CRM opportunity state through `/crm` when it is needed for a question; this skill does not create CRM writes. It produces no durable research artifact, so `./references/dependencies/conversint-current/references/output-paths.md` and research citations do not apply unless the user explicitly routes settled decisions into `/grill-with-docs`.
Interview loop
Repeat this loop until the user says the plan is clear enough to build:
- State the current decision under review in one sentence.
- Name the highest-risk unresolved assumption or dependency.
- Give your recommended answer.
- Ask exactly one question that would confirm, correct, or reject that answer.
- Wait for the user's reply.
- Use the reply to update the decision tree and pick the next branch.
Favor questions about:
- the user, buyer, operator, or reviewer who must live with the result; - what must be true for the plan to work; - what breaks if timing, budget, data, access, or attention is worse than expected; - whether an existing chiefOS skill already owns the work; - which system is the source of truth; - what must be captured durably versus kept as a one-off; - what approval gate is required before an external action or spend; - how the work will be verified.
Output format
Use this compact format for each turn:
```markdown Current decision: [one sentence] Recommended answer: [your concrete recommendation] Question: [exactly one question] ```
When repo evidence supports the recommendation, add a short `Evidence:` line before the question with file paths or command output summarized briefly.
Boundaries
- Do not start implementing the plan during a grilling session unless the user explicitly exits grilling and asks you to do the work. - Do not create durable docs, ADRs, glossary entries, CRM updates, project tasks, calendar events, emails, social posts, ads, or Drive files from this skill. - If the user wants docs created while interviewing, route to `/grill-with-docs`. - If the user asks for a normal build/fix instead of more questioning, stop the interview and execute the requested work.