vibecode-execution-templates

name	Vibecode Execution Templates
description	Sidecar data for /vibecode workflow — verification checklists, exit criteria templates, and report formats loaded just-in-time.
trigger	/vibecode workflow execution
glob	null
source	user

Skill: Vibecode Execution Templates

Sidecar Skill for the /vibecode workflow. Contains execution templates, verification checklists, and report formats that the Agent loads only when executing a specific mode.

Pattern: Workflow = Logic → Sidecar Skill = Data Router. The /vibecode workflow instructs the Agent to read this skill at mode-start time.

When to Load

Mode	File to load	When
review	references/review-report.md	Step 7 — Generate review report
loop	references/loop-config.md	Step 1 — Define exit criteria
loop	references/loop-report.md	Step 5 — Generate final report
auto	references/auto-queue.md	Step 1 — Build execution queue
--sandbox	references/sandbox-report.md	After sandbox loop/auto — Generate diff report
session	references/session-quality-gate.md	Phase start — Quality Gate proposal (TDD/graph/QA/hardened)
session	§ Phase Context Reload (below)	Phase start — Reload project rules/skills + tool-specific skills
all	references/verification.md	Verify step — project-type commands

References

File	Purpose
references/review-report.md	Plan audit scorecard with Agent Learning Notes
references/sandbox-report.md	Sandbox diff report + companion check
references/loop-config.md	Exit criteria presets by project type
references/loop-report.md	Iteration log + lessons learned
references/auto-queue.md	Execution queue + progress tracker
references/session-quality-gate.md	Sensitive keywords registry + Quality Gate template for session phases
references/verification.md	Build/lint commands by tech stack
scripts/session-manager.sh	Dynamic session KI (vibecode_session) management helper script

Mode Chaining Guide

Recommended chains:

  New plan:     review → loop --sandbox → apply → auto
  Quick fix:    loop (standalone, real project)
  Risky phase:  review → loop --sandbox [Phase N] → apply → auto --resume
  Quick peek:   loop --sandbox --max 1 (one-shot preview)
  Full dry-run: auto --sandbox (entire plan in sandbox)
  Free session: session (DSP draft → JIT phases → Quality Gate → commit)
  Targeted:     session refactor-auth (DSP with topic → Phase 1 ready)
  Hardened:     session --hardened (auto-detect + forced TDD on sensitive code)

Phase Context Reload (Session Mode)

At the start of every Phase in session mode, Agent MUST reload context to prevent drift in long sessions. This table defines WHAT to reload.

Mandatory (EVERY Phase)

#	What to reload	Path	Condition
1	Project rules index	Projects/[project]/.agents/rules.md	If exists
2	Project skills index	Projects/[project]/.agents/skills.md	If exists
3	Project domain skill	Projects/[project]/.agents/skills/[project]/SKILL.md	If exists
4	Project developer guidelines	Projects/[project]/.agents/AGENTS.md (or agents.md in root)	If exists

Conditional (based on activated Quality Tools)

Activated Tool	Skill to load	Path
🧪 TDD	TDD Guidelines	.agents/skills/tdd/SKILL.md
🔍 --graph	Graph Intelligence Router	.agents/skills/para-graph/SKILL.md
🔒 --hardened	TDD Guidelines (co-required)	.agents/skills/tdd/SKILL.md
⚗️ /brainstorm	Brainstorm Templates	.agents/skills/brainstorm/SKILL.md
📋 /qa	QA Review Templates	.agents/skills/qa/SKILL.md

Pattern: The /vibecode workflow references this table at Step 2a (Context Reload). Agent reads the sidecar skill → follows the routing table → loads the right skills. This keeps the workflow file lean and the routing logic centralized here.

Output Checklist

After execution, verify:

• Step 0 Pre-flight executed (rules + skills reloaded)
• Correct mode was used for the target task complexity
• All exit criteria explicitly defined (loop mode)
• User confirmed before real writes
• Phase gates respected (auto mode)
• Quality Gate and Pre-Code Checkpoint presented at EVERY Phase start (session mode — TDD/graph/QA/hardened)
• User approval obtained at Pre-Code Checkpoint and confirmed to run /plan [project-name] dev before making any code modifications
• Rollback mechanism available (loop: git stash, sandbox: delete)
• Path substitution correct (--sandbox: sandbox prefix)
• Bash scripts syntax-checked with bash -n (--sandbox)
• Sandbox cleaned up after completion
• Final report generated with file inventory + lessons learned
• Mandatory User confirmation obtained (suggesting /plan [project-name] end) before performing session teardown or archiving/cleaning plan files