Draft a decision-ready brief with ak:research-prompt

Use ak:research-prompt to prepare one self-contained research brief for a human or AI researcher. The Skill captures the project context, one decision, source standards, sub-questions, evidence format, and completion bar in exactly one paragraph. It writes the assignment; it does not search, cite findings, or make the decision.

Choose ak:research-prompt for a handoff

Use ak:research-prompt when

• You want a research prompt or brief that another person or tool can execute.
• The researcher will not have access to the current conversation.
• One decision needs several coordinated evidence questions.
• You want source hierarchy, contradiction handling, and per-finding citation requirements fixed before research begins.

Choose another workflow when

• You want the research performed and saved as a cited report. Use ak:research.
• You only need current library documentation. Use ak:docs-seeker.
• You need repository evidence gathered before the brief can be scoped. Use ak:scout first.
• You have multiple unrelated decisions. Create one brief per decision rather than combining them.

Prepare the decision context

Before you start:

• Complete Onboarding, and confirm the Skill is installed for the runtime and scope you use.
• State the project and current situation in language a researcher with no prior context can understand.
• Provide the decision, audience, deadline, known facts, constraints, intended use, and important non-goals.
• Identify primary sources the researcher should prefer and weak sources that may be used only as signals.
• Remove secrets, personal data, private URLs, internal identifiers, and any confidential detail the downstream researcher is not authorized to receive.

Run the Skill

/ak:research-prompt "Decision: choose an authorization model for our multi-tenant API. Audience: architecture review. Deadline: Friday. Compare database RLS and service-layer checks; require current primary sources, security failure modes, migration cost, and unresolved gaps."

/ak:research-prompt "Decision: choose an authorization model for our multi-tenant API. Audience: architecture review. Deadline: Friday. Compare database RLS and service-layer checks; require current primary sources, security failure modes, migration cost, and unresolved gaps."

$ak:research-prompt "Decision: choose an authorization model for our multi-tenant API. Audience: architecture review. Deadline: Friday. Compare database RLS and service-layer checks; require current primary sources, security failure modes, migration cost, and unresolved gaps."

The Skill accepts a research topic or decision as free-form input. It has no published mode flag. The source contract requires exactly one paragraph, with no heading, preface, explanation, or second deliverable.

What the paragraph must contain

Understand what happens during a run

1. The Skill gathers context. It uses the conversation and only the relevant, approved project files needed to explain the situation.
2. It isolates one decision. Adjacent questions are excluded or assigned to a separate brief.
3. It selects coverage. Three to six numbered sub-questions cover evidence, trade-offs, risks, compatibility, and implementation implications as needed.
4. It defines evidence quality. Primary sources lead; forums and social posts remain weak signals rather than proof.
5. It defines verification. The downstream researcher must revisit conflicts and single-source claims and label facts, inference, and uncertainty.
6. It returns one paragraph. The brief appears in the conversation by default and is ready to hand to ak:research or another research runner.

Keep the handoff private and bounded

• The source-defined workflow performs no web search, provider call, repository clone, package install, or research execution.
• It creates no file by default. Writing the paragraph to a path, sending it to a person, or submitting it to a research provider is a separate action.
• Reading relevant local project files can expose private context to the active runtime. Limit inspection to what is needed to make the brief self-contained.
• The Skill itself adds no external cost. The selected research runner may use paid models, search quotas, private repositories, or subscription sources; state those approvals in the brief when they are intended.
• Do not pre-fill expected conclusions or frame the question as marketing copy.

Verify the brief

A complete result is exactly one paragraph and can be understood without the original conversation. It names one decision, contains three to six inline numbered sub-questions, defines include and avoid boundaries, prioritizes primary sources, requires gap handling, specifies per-finding URLs and relevance, and asks for one detailed Markdown result.

The Skill bundles no scripts or tests. Verification is a contract check against the paragraph: one paragraph, one mission, complete context, no fabricated fact, no sensitive material, and no research claims presented as already verified.

Troubleshoot or continue

Know the current limits

• The Skill cannot make missing project facts true; unknowns should become sub-questions or explicit assumptions.
• One paragraph favors a focused decision over an exhaustive program of work.
• The brief enforces a source and output contract, but the downstream research runner determines actual coverage, cost, and evidence quality.
• Stable and beta package identical ak:research-prompt behavior and output rules.