| name | c1-agent-build-loop |
| description | Guide a coding agent through a complete spec-to-verify implementation loop. Use when a developer asks an agent to build, refactor, fix, or plan non-trivial software work and wants better results through an interview, tight spec, explicit verification, small checkpoints, and a reusable agent environment. |
Agent Build Loop
Overview
Run coding-agent work as a tight engineering loop: clarify the goal, write a small spec, define verification, get approval, implement in checkpoints, and verify before calling the work done.
Use this as the orchestrator. For deeper work, route to:
c1-agent-spec for turning the request into a decision-ready spec.
c1-agent-verify for the verification plan and completion checks.
c1-agent-environment when the repo lacks reusable agent instructions, context, or guardrails.
Workflow
-
Ground in the repo before asking questions.
- Inspect entrypoints, existing patterns, tests, package scripts, and relevant docs.
- Do not ask the user for facts that can be discovered from the workspace.
-
Interview for intent.
- Clarify the real goal, user or audience, success criteria, constraints, and non-goals.
- Prefer a few high-impact questions over a long survey.
- If the user already gave a decision-complete plan, summarize it and move on.
-
Produce a tight spec.
- Keep scope small enough to review.
- Include intended behavior, key interfaces, data flow, edge cases, and acceptance criteria.
- Name assumptions explicitly.
-
Define verification before implementation.
- List the commands, tests, manual checks, screenshots, logs, or external signals that will prove the work.
- If verification is impossible or incomplete, say so before starting.
-
Get explicit approval before mutating work.
- Do not edit files, install packages, run migrations, commit, deploy, delete data, or perform other mutating work until the user clearly approves.
- Read-only exploration is allowed unless the user says otherwise.
-
Implement in checkpoints.
- Work in the smallest coherent slice.
- After each slice, inspect the diff and run the relevant verification.
- Update the user when direction, scope, or risk changes.
-
Finish with evidence.
- Report what changed, what was verified, what could not be verified, and any follow-up risks.
- Do not claim completion without running the agreed checks or clearly naming the gap.
Useful Prompts
Use or adapt these prompts inside the session:
Interview me until the goal, constraints, success criteria, and non-goals are clear. Then write a tight implementation spec.
Before implementing, define the verification plan you will use to prove this works.
Break this into the smallest useful checkpoint, complete that slice, verify it, then continue.