<< All versions

Skill v1.0.2

currentAutomated scan100/100
pageel/para-workspace/vibecode

~2 modified

──Details
PublishedAugust 23, 2026 at 04:55 PM
Content Hashsha256:25869ed4b0583503...
Git SHA9e7967ac61d1
Bump Typepatch
Compare with v1.0.1
──Files
Files (1 file, 6.2 KB)
SKILL.md6.2 KBactive
SKILL.md · 116 lines · 6.2 KB

version: "1.0.2" 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: 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

ModeFile to loadWhen
reviewreferences/review-report.mdStep 7 — Generate review report
loopreferences/loop-config.mdStep 1 — Define exit criteria
loopreferences/loop-report.mdStep 5 — Generate final report
autoreferences/auto-queue.mdStep 1 — Build execution queue
--sandboxreferences/sandbox-report.mdAfter sandbox loop/auto — Generate diff report
sessionreferences/session-quality-gate.mdPhase start — Quality Gate proposal (TDD/graph/QA/hardened)
session§ Phase Context Reload (below)Phase start — Reload project rules/skills + tool-specific skills
allreferences/verification.mdVerify step — project-type commands

References

FilePurpose
references/review-report.mdPlan audit scorecard with Agent Learning Notes
references/sandbox-report.mdSandbox diff report + companion check
references/loop-config.mdExit criteria presets by project type
references/loop-report.mdIteration log + lessons learned
references/auto-queue.mdExecution queue + progress tracker
references/session-quality-gate.mdSensitive keywords registry + Quality Gate template for session phases
references/verification.mdBuild/lint commands by tech stack
scripts/session-manager.shDynamic session KI (vibecode_session) management helper script
.agents/skills/plan/scripts/check-phase-tasks.jsUniversal pre-git task completion checker (exit 0 = all done, exit 1 = unchecked tasks remain)

Mode Chaining Guide

text
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)

Scope Expansion Guard (Dynamic Task Addition)

During a vibecode session (especially in loop or auto mode), if the Agent discovers a missing requirement or edge case and creates a new ad-hoc task, it MUST:

  1. Check if the current session or active plan enforces CSA or TDD (by reading .para-workspace.yml, project.md, or active memory).
  2. If CSA/TDD is active, the Agent MUST HALT and ask the user: "This new task requires a Spec update (CSA) and/or Test Case (TDD). Shall I propose the spec/test changes first?"
  3. Only after the Spec/Test is updated and approved (or if the user explicitly overrides), may the Agent proceed to implement the production 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 reloadPathCondition
1Project rules indexProjects/[project]/.agents/rules.mdIf exists
2Project skills indexProjects/[project]/.agents/skills.mdIf exists
3Project domain skillProjects/[project]/.agents/skills/[project]/SKILL.mdIf exists
4Project developer guidelinesProjects/[project]/.agents/AGENTS.md (or agents.md in root)If exists

Conditional (based on activated Quality Tools)

Activated ToolSkill to loadPath
🧪 TDDTDD Guidelines.agents/skills/tdd/SKILL.md
🔍 --graphGraph Intelligence Router.agents/skills/para-graph/SKILL.md
🔒 --hardenedTDD Guidelines (co-required).agents/skills/tdd/SKILL.md
⚗️ /brainstormBrainstorm Templates.agents/skills/brainstorm/SKILL.md
📋 /qaQA 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)
  • [ ] Phase Start Baseline Gate presented at EVERY Phase ≥ 1 (proposing /brainstorm, /qa, /spec, /research)
  • [ ] Post-Baseline Approval Gate enforced: Agent STOPPED turn after brainstorm/spec/qa output and waited for user confirmation before coding (Guard #17)
  • [ ] 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
  • [ ] Pre-git Task Done Check (check-phase-tasks.js) executed with exit code 0 before every git commit
  • [ ] Mandatory User confirmation obtained (suggesting /plan [project-name] end) before performing session teardown or archiving/cleaning plan files
← v1.0.1All versions