Case study 06 / Open source / Developer tooling
Interrogation Mode
An open-source skill that helps coding agents clarify decisions and agree on a testable specification before implementation.
- MY ROLE
- Skill author and maintainer.
- CONTEXT
- Independent open-source developer tool
- FOCUS
- Agent skills / Markdown / Requirements design
Overview
Interrogation Mode is an explicitly invoked requirements workflow for coding agents. It turns an underspecified request into a specification with acceptance criteria the user can review.
Workflow
The agent first reads relevant project context, then identifies decisions that must be made before implementation. Auto mode sizes the interview using scope, uncertainty, consequences and integrations. Light, Standard and Deep are also available explicitly.
The same approval boundary applies at every depth. Once blockers are resolved, the agent produces a specification and waits for the user to authorize implementation.
Engineering decisions
Keep the workflow in a shared skill with host-specific invocation and installation instructions. This makes the intended behavior inspectable across agents and keeps the user's decisions in the current conversation.
- Problem
- A short implementation request can hide decisions that affect interfaces, failure behavior or acceptance.
- Decision
- Separate read-only discovery, blocker resolution, a testable specification and explicit approval.
- Reason
- Adjust the depth of the interview to the task while treating consequential unknowns as blockers at every depth.
- Trade-off
- The workflow adds an approval step and relies on the coding agent following instructions. Host permissions provide any technical enforcement.
Validation and limits
Published acceptance scenarios describe what to check when the request is already clear, a blocker remains open or the user approves a specification. The workflow is instruction-based; adherence depends on the agent and its environment.
Links
Kochura Deploy
Backend / Product