agentsclimarketplace

Mission control cuda tile refactor

Skill MN755/Codex-Mission_Control/plugins/mission-control/skills/mission-control-cuda-tile-refactor

An Open Source Platform For Building Projects With OpenAI's Codex, Claude CLI, And Other Coding LLMs Using Multi-Agent Swarms And Customizable Workflows. Runs On Windows, MacOS, And Linux.

Install
npx -y skills add MN755/Codex-Mission_Control --skill mission-control-cuda-tile-refactor

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when CUDA Tile or tile-shaped GPU refactors need Mission Control planning, validation, and benchmark discipline.

SKILL.md

2.2 KB, as published. Nobody here has run it

Mission Control CUDA Tile Refactor

Purpose

Route CUDA Tile refactors through Mission Control so tile shape, launch configuration, and benchmark deltas stay explicit.

The Codex chat agent is not the Mission Control Manager. It is the bridge between the user and the Mission Control Manager.

Use when

  • The repo uses CUDA Tile or similar tile-programming patterns.
  • The user wants a kernel reshaped for throughput or memory behavior.
  • A refactor needs performance proof, not just syntactic churn.

Workflow

  1. Read the current codebase and validation context for the kernel path.
  2. Ask Mission Control for a bounded tile-refactor plan with correctness checks first.
  3. Require benchmark and profiling loops after the refactor.
  4. Surface any GPU infrastructure blockers before blaming the tile change.
  5. Review the final evidence and known tradeoffs.

Mission Control calls

Tools:

  • mission_control_start_task
  • mission_control_get_status
  • mission_control_request_snapshot

Resources:

  • mission-control://projects/{project_id}/codebase-map
  • mission-control://projects/{project_id}/validation-summary
  • mission-control://projects/{project_id}/decision-ledger

User-facing output

  • Show the tile-refactor goal and affected kernel area.
  • Show the benchmark and profile loop Mission Control expects.
  • Show whether the result is blocked by code, infrastructure, or mixed evidence.

Approval behavior

Risky refactors, benchmark commands, profilers, and rollback points still require whatever approval Mission Control policy demands.

Never do

  • Do not refactor tile code without a validation loop.
  • Do not confuse a style cleanup with a performance refactor.
  • Do not hide benchmark regressions behind pretty code.

Failure and fallback

If Mission Control cannot confirm a tile-oriented code path, fall back to a narrower CUDA-kernel review before proposing a tile refactor.

Example invocation

Use Mission Control to refactor this CUDA Tile kernel and prove the result.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.