orchestrate-workflow

A setup guide for a governed team of AI agents: one writes the change, independent reviewers inspect it, and a fixer applies accepted feedback. Agentstack provisions the agents while another executor runs the workflow.

In plain words
What is it for?
Use it to run implementation, adversarial code review, and fixing stages through Sandcastle, Claude Code workflows, or a custom Docker-based process.
Why use it?
It separates agent setup from workflow execution and keeps each role limited to the tools and instructions it needs. This makes review responsibilities clearer.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/tarekkharsa/agentstack/orchestrate-workflow
Any agent
npx skills add Tarekkharsa/agentstack --skill orchestrate-workflow
Clone the repo
git clone --depth 1 https://github.com/Tarekkharsa/agentstack

Made for: Claude Code, Codex.

Per session 80 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,118 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00080 $0.01118
Opus 5 $0.00040 $0.00559
Sonnet 5 $0.00016 $0.00224
Haiku 4.5 $0.00008 $0.00112

Measured 2d ago against content hash 4dc405376b95, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

orchestrate-workflow scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

crates/cli/catalog/skills/orchestrate-workflow/SKILL.md · 108 lines

How it starts

The opening of the file, as written. The whole thing — 108 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Orchestrate a governed multi-agent workflow

Use when you want the Bun-in-Rust shape — an implementer writes, independent adversarial reviewers attack the diff, a fixer applies feedback — with each agent's capabilities, model, and secrets managed by agentstack instead of hand-assembled per run.

The division of labor is fixed: an executor runs the loop (sandcastle, Claude Code workflows, your own script); agentstack defines and provisions the agents the loop spawns. Don't blur it in either direction.

1 — Define roles as profiles

A role is a profile: which skills, which servers, and (by convention) which model. In .agentstack/agentstack.toml:

[profiles.implementer]
skills  = ["porting-guide"]        # the task's context artifacts
servers = ["github"]

[profiles.reviewer]
skills  = ["adversarial-review"]   # ships in this catalog
servers = []                       # reviewers judge the diff; no tools needed

Keep reviewer profiles minimal on purpose — a reviewer with no servers can't be tool-poisoned, and the diff is all it should trust anyway.

2 — Bind models to roles

Pick per role, not per run: bulk/mechanical implementation → a cheap strong coder; review → a different model family than the implementer when possible (diverse failure modes). Record the binding wherever the executor configures each agent (sandcastle's agent: option, a Workflow model: param, a --model flag). If the route-by-cost skill is loaded, apply its ladder.

3 — Provision the sandbox (works today, no extra tooling)

The box needs three things: the harness CLIs, the rendered capabilities, and secrets that never touch disk.

# on the host — render the role's capabilities into the worktree the
# sandbox will mount (repeat per worktree):
cd <worktree> && agentstack use implementer --write

# start the container with secrets injected from the keychain at run time —
# no .env file, nothing baked into the image or committed:
docker run \
  -e GH_PAT="$(agentstack secret get GH_PAT)" \
  -e ANTHROPIC_API_KEY="$(agentstack secret get ANTHROPIC_API_KEY)" \
  -v <worktree>:/work ...

Read the full file on GitHub · 108 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 108 lines · 80 tokens per session scan A 4dc405376b95

Subscribe to this mod's changes

orchestrate-workflow is a skill published in the GitHub repository Tarekkharsa/agentstack (3 stars, last pushed 19d ago), licensed Apache-2.0. It adds 80 tokens to every session and 1,118 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.