# Decant Orchestrate

> Route a multi-stage repository task through the smallest useful Decant workflow. Use for work that mixes research, planning, implementation, review, or release; do not use for a single obvious read or edit.

- **Type:** Skill
- **Install:** `agentstack add skill-minwoo19930301-decant-decant-orchestrate`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [minwoo19930301](https://agentstack.voostack.com/s/minwoo19930301)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [minwoo19930301](https://github.com/minwoo19930301)
- **Source:** https://github.com/minwoo19930301/decant/tree/main/plugins/decant/skills/decant-orchestrate

## Install

```sh
agentstack add skill-minwoo19930301-decant-decant-orchestrate
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Decant Orchestrate

Coordinate the other Decant skills without turning every request into a large process.

## Preflight

1. Restate the requested outcome, constraints, and forbidden actions.
2. Inspect the repository and current git state before proposing edits.
3. If the CLI is available, run `decant doctor` and preview routing with `decant route "" --json`.
4. Treat `decant run "" --dry-run` as a preview. Run a mutating pipeline only when the user authorized implementation.

This skill helps choose a workflow. It does not change the model of the host agent's current task, whether that host is Codex or Claude Code, and does not make unsupported providers available.

## Hard-deadline progress gate

Use this gate whenever the user gives an explicit deadline and the work is low risk. For a budget of 15 minutes or less, treat it as a fast lane and recommend medium-or-lower host effort before starting. This skill cannot change the host model or effort and cannot enforce a wall-clock timer.

1. Deliver a runnable primary implementation slice by 30% of the budget, capped at five minutes.
2. Run a narrow smoke check, then core tests.
3. Expand behavior and documentation only with the remaining time.
4. Stop expansion when the final 20% begins; use that time to verify and report honestly.

Tests, plans, or documentation alone do not satisfy the first progress gate. Security-sensitive, deployment, destructive, irreversible, or otherwise high-risk work never qualifies for this shortcut; use the normal risk-proportionate workflow without weakening its safety gates.

## Choose the smallest path

- Read-only evidence gathering: use `decant-research` only.
- Ambiguous or high-impact change: use `decant-spec`, then `decant-build`.
- Approved, well-specified edit: use `decant-build` directly.
- Failure or regression: use `decant-debug`; edit only when a fix is authorized.
- Review request: use `decant-review` without editing.
- Publication or handoff: use `decant-release` after implementation checks pass.
- Skill creation or trigger tuning: use `decant-skill-lab`.

Skip stages whose output cannot change the decision. Add review depth as risk, blast radius, irreversibility, or verification difficulty rises.

## Handoff contract

For every active stage, record:

- goal and non-goals;
- inputs and evidence locations;
- allowed mutations;
- result and unresolved risks;
- deterministic checks and semantic review;
- next owner or next action.

Separate observed facts, inferences, and proposals. Never call a stage complete from intention alone.

## Safety boundaries

- Preserve unrelated work in a dirty worktree.
- Do not broaden authorization from diagnosis to repair, or from local changes to publication.
- Do not hide destructive actions behind a recipe, subagent, hook, or retry loop.
- Stop repeated repair attempts after three materially similar failures and reassess the hypothesis.
- Keep databases, daemons, worktrees, swarms, and GUIs optional rather than core requirements.

## Source & license

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

- **Author:** [minwoo19930301](https://github.com/minwoo19930301)
- **Source:** [minwoo19930301/decant](https://github.com/minwoo19930301/decant)
- **License:** MIT

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

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-minwoo19930301-decant-decant-orchestrate
- Seller: https://agentstack.voostack.com/s/minwoo19930301
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.