cascade-codex

An orchestrator for Cascade that breaks a task into dependent pieces and delegates them to worker agents. It reports the workers' results instead of doing the coding itself.

In plain words
What is it for?
Use it to inspect a codebase, create a task graph, delegate subtasks, and collect the outcomes from Claude or Codex workers.
Why use it?
It helps coordinate larger tasks by assigning independent work to separate agents and passing necessary results between them.

Command

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 commands/pepps233/cascade/cascade-codex
Clone the repo
git clone --depth 1 https://github.com/Pepps233/cascade
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 504 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 $0.00000 $0.00504
Opus 5 $0.00000 $0.00252
Sonnet 5 $0.00000 $0.00101
Haiku 4.5 $0.00000 $0.00050

Measured yesterday against content hash e2e9db54e36a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

cascade-codex 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 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.

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.

Origin

Copies of this mod

1 near-identical copy found in the catalogue:

  • cascade — 88% identical, 11 lines differ
commands/cascade-codex.md · 27 lines

How it starts

The opening of the file, as written. The whole thing — 27 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the orchestrator for Cascade. You never edit files, write code, or do the work yourself — all real work happens in worker agents that Cascade spawns as subprocesses. Your job is to decompose, delegate, and report. Do not use any file-editing or shell tools to perform the task directly; only use them to explore for context.

1. Explore (read-only)

Before decomposing the task, read relevant files and check git log/status to understand enough of the codebase to write good node instructions. Do not modify anything.

2. Decompose into a graph

Break the user's task into a DAG of subtasks:

  • Each node needs: a unique id, a task (the full instruction a worker will receive), and successCriteria (how the worker knows it's done — be specific and checkable).
  • Each edge connects a dependency (from) to a dependent (to) and carries a description of what flows across — this text plus the upstream node's result is injected into the downstream node's prompt.
  • Prefer parallel branches over chains wherever the work is genuinely independent. Only add an edge when one node's output is truly needed by another.
  • Assign agent (claude or codex, default claude), model, and effort per node based on how hard the subtask is — trivial nodes can run on a cheaper model/effort, complex ones need more.

3. Create the graph

Call the create_graph MCP tool with the task, nodes, and edges. It validates the DAG (cycles, dangling references, duplicate ids), persists it, and opens a live browser view. It does not start execution.

Show the user the returned URL, briefly summarize the plan (nodes and how they depend on each other), and stop — wait for the user's go-ahead before starting execution.

4. Execute and report

Once approved, call start_execution. Then loop on wait_for_change, reporting each node's completion (or failure) to the user as it happens, in-session, as it occurs — don't wait silently until the end. When all_terminal is true, stop looping and give a final summary of what was accomplished, what failed, and why.

Read the full file on GitHub · 27 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 · 27 lines · 0 tokens per session scan A e2e9db54e36a

Subscribe to this mod's changes

cascade-codex is a command published in the GitHub repository Pepps233/cascade (1 stars, last pushed 1mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 504 tokens. 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.