Skip to content
Back to search

astra-flash-orchestrator

Coordinate your models from Codex. Plan, delegate, use host tools, and review work across workspaces. Formerly Astra Flash Orchestrator.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/ethanplusai--astra-flash-orchestrator/badge.svg)](/listings/ethanplusai--astra-flash-orchestrator)

HTML

<a href="/listings/ethanplusai--astra-flash-orchestrator"><img src="/api/listings/ethanplusai--astra-flash-orchestrator/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

You need **Node.js 24.15+**, a Codex client with plugin support, and your own model access. Native workers need a compatible configured route. Claude mode needs the official Claude Code CLI, authenticated through your subscription.

From this checkout in a stable folder:

codex plugin marketplace add "$PWD"
codex plugin add codex-on-crack@codex-on-crack

Add the panel if you want it:

codex plugin add codex-on-crack-panel@codex-on-crack

Start a fresh conversation:

> Help me plan this project. Check which models I have available, recommend > who should do what, and agree on the plan with me before building. > Connect this workspace and its runs to the Codex on Crack panel.

The three core skills are **crack-plan** (plan and choose), **crack-setup** (configure workers with preview and undo), and **crack** (execute and review).

For existing Astra Flash Orchestrator installations, read the [upgrade guide](docs/UPGRADING.md) before switching policies. Setup does not run paid model requests. Builds use the model access you authorize.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 141 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Codex on Crack
  2. **Run your models together. Keep the work in Codex.**
  3. Previously **Astra Flash Orchestrator**. The original Astra + Flash workflow grew
  4. into a package for planning, delegation, tool handoffs, and reviewing work across
  5. your configured models. [Upgrading from the original? Start here.](docs/UPGRADING.md)
  6. > Run subagents through Opus 5.5.
  7. > Have Opus plan this and Flash implement it. Bring the result back for review.
  8. > Help me choose which of my models should work on this project.
  9. You describe the work. The skills find the supported route in your setup, agree
  10. on the plan with you, and carry the assignment through implementation and review.
  11. You don't have to explain the CLI or your subscription in every prompt.
  12. ![15-second demo: type a prompt, open the panel, follow agents, and send design feedback](docs/media/panel-demo.gif)
  13. *15-second illustrated walkthrough with a recreated Codex interface and synthetic
  14. project data. The workflow is condensed; this is not a live build recording.*
  15. [Install](INSTALL-IN-CODEX.md) · [Upgrade](docs/UPGRADING.md) ·
  16. [Workflow](docs/WORKFLOW.md) · [Panel](plugins/codex-on-crack-panel/README.md) ·
  17. [OpenCode / original installer](INSTALL-IN-OPENCODE.md)
  18. ## One conversation, different models
  19. Your selected **Codex model** keeps the conversation, coordinates the work, and
  20. uses the host's available tools. It can delegate to a compatible native worker,
  21. including a model configured through Codex Router.
  22. **Opus 5.5** can implement a whole assignment through your authenticated Claude
  23. Code client. It can also plan and review the build while Codex dispatches a native
  24. worker, such as **DeepSeek V4.1 Flash**, for implementation. Codex returns worker
  25. results and host-tool evidence to the lead.
  26. That means you can use Claude for a build and still ask the Codex host to check
  27. it in the browser, use available computer tools, or generate an image. External
  28. Claude sessions do not directly inherit Codex tools or replace the model selected
  29. in your composer. The host performs those steps.
  30. | Arrangement | Who does what |
  31. | --- | --- |
  32. | Codex host + native worker | The host scopes and reviews; a compatible configured worker implements. |
  33. | Codex host + Opus | Opus implements through Claude Code; Codex coordinates and validates. |
  34. | Codex host + Opus + native worker | Opus plans and reviews; Codex handles dispatch and tool requests; the worker implements. |
  35. | Just the current model | Small tasks and explicit solo work stay in the current conversation. |
  36. Model choice is yours. Recommendations use your configured roles, preferences,
  37. and declared capabilities. The package doesn't benchmark every model or silently
  38. substitute another model or billing route. The current Claude adapter targets
  39. `claude-opus-5-5`; other Claude models are not automatically supported by it.
  40. ## See what is happening
  41. The optional panel lives alongside your conversation as an embedded MCP App in
  42. supported Codex hosts. A localhost view is available as a fallback.
  43. Choose a workspace first. Its **Overview**, **Review**, and **Usage & routes**
  44. then show only that project's registered work:
  45. - Follow phases, agents, model handoffs, and recorded tool activity.
  46. - Inspect per-run timing and available input, output, and cache counters.
  47. - Answer visual questions and leave feedback while the conversation continues.
  48. - Keep decisions tied to the screenshot revision you actually reviewed.
  49. - Record registered sources and export an offline replay for later inspection.
  50. Feedback is saved for the agent to read at a natural checkpoint. Sending it does
  51. not interrupt the conversation or count as approval. Explicit approval is a
  52. separate action. Register the sessions and runs you want to observe; the panel
  53. doesn't automatically read every conversation on your computer.
  54. ## Install
  55. You need **Node.js 24.15+**, a Codex client with plugin support, and your own model
  56. access. Native workers need a compatible configured route. Claude mode needs
  57. the official Claude Code CLI, authenticated through your subscription.
  58. From this checkout in a stable folder:
  59. ```sh
  60. codex plugin marketplace add "$PWD"
  61. codex plugin add codex-on-crack@codex-on-crack
  62. ```
  63. Add the panel if you want it:
  64. ```sh
  65. codex plugin add codex-on-crack-panel@codex-on-crack
  66. ```
  67. Start a fresh conversation:
  68. > Help me plan this project. Check which models I have available, recommend
  69. > who should do what, and agree on the plan with me before building.
  70. > Connect this workspace and its runs to the Codex on Crack panel.
  71. The three core skills are **crack-plan** (plan and choose), **crack-setup**
  72. (configure workers with preview and undo), and **crack** (execute and review).
  73. For existing Astra Flash Orchestrator installations, read the
  74. [upgrade guide](docs/UPGRADING.md) before switching policies.
  75. Setup does not run paid model requests. Builds use the model access you authorize.
  76. ## What changed from Astra Flash Orchestrator?
  77. The original package prescribed one arrangement: Astra plans and reviews, Flash
  78. builds. That is still a useful arrangement. Codex on Crack extends the idea to
  79. other configured workers, a Claude Code execution route, project-lead handoffs,
  80. and a panel for following and reviewing the work.
  81. The name changed because the package is no longer tied to two models.
  82. It is still built around a scoped assignment and a lead checking the result.
  83. ## Judge the result, then the usage
  84. Does the finished product work? Does it look right? How much repair did it need?
  85. Those questions come before token totals.
  86. Moving work to another provider can preserve Codex allowance while adding costs
  87. elsewhere. Tokens are not weekly allowance, and subscription dollar estimates
  88. are not API bills. Resumed sessions may contain cumulative counters. The panel
  89. keeps missing information unknown rather than estimating it.
  90. The original usage result described a particular test. It is not a universal
  91. savings promise for this package or every model combination.
  92. [Record your own build](docs/RECORDING.md) to see what helps your work.
  93. ## Before you use it
  94. This is a preview release. See [troubleshooting](docs/TROUBLESHOOTING.md) for
  95. setup and compatibility issues. Instructions and file scopes are not an OS sandbox.
  96. Your selected model, credentials, and permissions stay under your control.
  97. Use one active orchestration policy per task.
  98. [Documentation](docs/README.md) · [Troubleshooting](docs/TROUBLESHOOTING.md) ·
  99. [Contributing](CONTRIBUTING.md) · [Security](SECURITY.md) · [Sources](SOURCES.md)
  100. MIT licensed. Not affiliated with OpenAI, Anthropic, or DeepSeek.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.