cas-supervisor

A supervisor guide for planning and coordinating multi-agent software projects, where separate workers handle assigned tasks.

In plain words
What is it for?
Use it to plan project tasks, assign workers, coordinate their branches, and verify the finished work.
Why use it?
It helps divide large work into coordinated pieces while reducing conflicts and keeping track of completion.

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/codingagentsystem/cas/cas-supervisor
Any agent
npx skills add codingagentsystem/cas --skill cas-supervisor
Clone the repo
git clone --depth 1 https://github.com/codingagentsystem/cas

Made for: Claude Code, Codex.

Per session 60 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,712 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. 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 $0.00060 $0.01712
Opus 5 $0.00030 $0.00856
Sonnet 5 $0.00012 $0.00342
Haiku 4.5 $0.00006 $0.00171

Measured yesterday against content hash adeebccb0c22, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

cas-supervisor scanned grade B with 1 finding 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 yesterday.

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.

Unrestricted tool accessmediumExcessive agency

A wildcard tool grant or "run any command" leaves no least-privilege boundary at all.

5. **End your turn immediately.** Stop here. Do not monitor, poll, or run any commands. Workers will push a message to you when done or blocked. Your next action is triggered by their message, not by checking.
.claude/skills/cas-supervisor/SKILL.md · 152 lines

How it starts

The opening of the file, as written. The whole thing — 152 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Factory Supervisor

You coordinate workers to complete EPICs. You are a planner, not an implementer.

Hard Rules

  • Never use SendMessage. Use mcp__cas__coordination action=message target=<name> message="..." summary="<brief summary>" for all communication. SendMessage is blocked in factory mode.
  • Never implement tasks yourself. Delegate ALL coding to workers.
  • Never close tasks for workers. Workers own their closes via mcp__cas__task action=close. When a worker reports completion, tell them to close it themselves. If they hit "verification required", the task-verifier runs in the worker's session — the worker must follow the verification flow, not you.
  • Never monitor, poll, or sleep. The system is push-based. After assigning tasks, you MUST stop responding and wait for an incoming message. Workers will message you when they complete tasks, hit blockers, or have questions. You do NOT need to check on them.
  • Epics are yours to verify and close. Only the supervisor verifies and closes the epic task itself (after all subtasks are done and merged).

What "end your turn" means

After you assign tasks and send context to workers, produce no more output. Do not:

  • Run git log, git diff, or any git command to check for worker commits
  • Run mcp__cas__task action=list to see if task statuses changed
  • Run mcp__cas__coordination action=worker_status to check worker activity
  • Use any tool "just to see" what's happening

Your next action should ONLY happen in response to a worker message or a user prompt. Between those events, you are idle. This is correct behavior — you are not "waiting", you are done until someone contacts you.

Worker Modes

Workers can run in two modes:

  • Isolated (isolate=true): Each worker gets its own git worktree and branch. Use when workers will modify overlapping files or when you need clean branch-based merging.
  • Shared (isolate=false or omitted): Workers share the main working directory. Simpler setup, but workers must coordinate to avoid editing the same files simultaneously.

Read the full file on GitHub · 152 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. yesterday First seen · 152 lines · 60 tokens per session scan B adeebccb0c22

Subscribe to this mod's changes

cas-supervisor is a skill published in the GitHub repository codingagentsystem/cas (151 stars, last pushed 5mo ago), licensed MIT. It adds 60 tokens to every session and 1,712 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (unrestricted tool access). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.