Vibe Loop

Purpose

Drive the Vibe workflow deterministically for a target repository.

This skill uses .vibe/STATE.md and .vibe/PLAN.md to recommend the next loop prompt and to validate invariants.

Inputs

• Repo root path (default: current working directory)
• Optional: --strict validation for CI-grade checks

Scripts

• scripts/agentctl.py
• scripts/vibe_next_and_print.py

How to use

• Show current status:
• Run: python3 scripts/agentctl.py --repo-root . --format json status
• Recommend next loop:
• Run: python3 scripts/agentctl.py --repo-root . --format json next
• Recommend next loop using a named workflow (for example workflows/continuous-refactor.yaml):
• Run: python3 scripts/agentctl.py --repo-root . --format json next --workflow continuous-refactor
• Pass the workflow file stem (continuous-refactor), not the .yaml filename.
• Validate invariants:
• Run: python3 scripts/agentctl.py --repo-root . --format json validate
• Strict mode (recommended in CI): add --strict
• Record loop completion from emitted LOOP_RESULT:
• Run: python3 scripts/agentctl.py --repo-root . --format json loop-result --line 'LOOP_RESULT: {...}'
• Print the next loop prompt body directly:
• Run: python3 scripts/vibe_next_and_print.py --repo-root . --show-decision
• The helper uses the same CODEX_HOME-aware layout detection as tools/bootstrap.py so it works from both repo trees and global installs.

Intended usage pattern

1) Run next to obtain recommended_prompt_id. 2) Use the vibe-prompts skill to print the corresponding prompt body. 3) Execute that prompt loop in the agent session. 4) Record LOOP_RESULT via agentctl loop-result. 5) Select next loop from next. Repeat until dispatcher returns stop or blocking.

The vibe_next_and_print.py helper wraps agentctl.py and prompt_catalog.py, prints the decision JSON (when --show-decision is set), and then emits the prompt body from the catalog. It locates the installed vibe-prompts skill via the same CODEX_HOME-aware logic as tools/bootstrap.py so it works from both repo sources and global installs.

This skill intentionally does not implement product code by itself; it selects and gates the next action.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: brifl
• Source: brifl/coding-agent-orchestration
• License: MIT

Install and usage instructions live in the source repository linked above.