Mandu Agent Workflow

SkillAI & models

Canonical Mandu workflow for safe agent-driven application changes.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Mandu Agent Workflow skill

What this skill tells your AI

The instructions your AI receives, as published by konamgil/mandu in skills/official/mandu-agent-workflow/SKILL.md and read by ahel’s review.

Use this skill before changing a Mandu project. The canonical loop is:

context -> plan -> apply -> verify -> repair

Prefer the official MCP tools when available:

  • mandu.agent.context
  • mandu.agent.plan
  • mandu.agent.apply
  • mandu.agent.verify
  • mandu.agent.repair
  • mandu.agent.sync

The CLI equivalents use mandu agent <step>. Domain skills are addenda and must not replace this workflow. Do not edit .mandu/generated artifacts, write outside the planned scope, or invoke deployment providers.

For writes, pass explicit typed operations to mandu.agent.plan. Preview is the default. Execute only the resulting bound plan with mandu.agent.apply using dryRun=false. CLI fallback:

mandu agent plan "<task>" --operations typed-operations.json --write --json
mandu agent apply --from .mandu/agent-plan.json --json
mandu agent apply --from .mandu/agent-plan.json --execute --write --json

Never weaken or remove baseRevision, content-hash, scope, permission, idempotency, verification, or rollback fields. An intent-only plan is a compatibility preview and must not be represented as an executed change. If a completed receipt must be reverted, use mandu agent repair --rollback <rollbackId> --apply.

End code-changing work with:

mandu agent verify --changed --json --write

Signals

GitHub stars
46
Forks
3
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
mandu-agent-workflow
Source
github.com/konamgil/mandu