Skip to content
Back to search

avenoxskills

Production agent skills for Claude Code, Cursor, and any SKILL.md harness — Codex fleets, video pipeline, monorepo review bundles, multi-chain explorer.

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/avenoxai--avenoxskills/badge.svg)](/listings/avenoxai--avenoxskills)

HTML

<a href="/listings/avenoxai--avenoxskills"><img src="/api/listings/avenoxai--avenoxskills/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.

Copy any skill directory into your agent's skills folder:

git clone https://github.com/avenoxai/avenoxskills.git
cp -R avenoxskills/skills/codex-fleet ~/.claude/skills/

Copy the whole skill directory, including any `scripts/` and `references/`. The video trio also shares `avenox-studio/` (see below).

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 · 122 lines

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

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # avenoxskills
  2. Agent skills built and battle-tested in production by [Avenox](https://avenox.lol).
  3. These aren't demos. Each one runs real work — shipping YouTube videos, driving
  4. Codex fleets, packaging codebases for external review — and each carries the
  5. gotchas that only show up after something breaks at 2am. That's the part worth
  6. having.
  7. Compatible with [Claude Code](https://claude.com/claude-code), Cursor, and any
  8. harness that reads `SKILL.md`-style agent skills.
  9. ## Install
  10. Copy any skill directory into your agent's skills folder:
  11. ```bash
  12. git clone https://github.com/avenoxai/avenoxskills.git
  13. cp -R avenoxskills/skills/codex-fleet ~/.claude/skills/
  14. ```
  15. Copy the whole skill directory, including any `scripts/` and `references/`.
  16. The video trio also shares `avenox-studio/` (see below).
  17. ## The skills
  18. ### Agent operations
  19. | Skill | What it does |
  20. |---|---|
  21. | **[codex-fleet](skills/codex-fleet)** | Standalone Codex CLI runner + fleet orchestrator. General `codex exec` tasks, `gpt-image-2` image generation, and parallel multi-lane fleets with worktree isolation. Dependency-free — no control plane required. |
  22. | **[omp-fleet](skills/omp-fleet)** | The same job for [Oh My Pi](https://www.npmjs.com/package/@oh-my-pi/pi-coding-agent) (`omp`) — a second coding-agent CLI onto the same Codex subscription. Provider pinning so a lane can't fall through to a metered aggregator, a 25×-cheaper model tier for recon, and in-process subagent fan-out. Includes the measured RAM comparison that decides which harness you actually want. |
  23. | **[fable-orchestration](skills/fable-orchestration)** | Delegation policy for a multi-model stack: when the main loop runs on a strong but expensive model (Fable 5.2), which work goes to Opus, Sonnet, Haiku, and which to Codex lanes (gpt-6.1-sol, gpt-6-astra, gpt-6-luna), and how tightly to write the brief for each. Routes on *difficulty*, not just task type. |
  24. | **[limit](skills/limit)** | Shows Claude and Codex subscription usage windows on one colored screen — live API/CodexBar cache first, rollout fallback, loud staleness warnings — so an agent knows which pool to delegate to before it burns quota blind. Stdlib Python, no dependencies. |
  25. ### External model review
  26. | Skill | What it does |
  27. |---|---|
  28. | **[gptpro](skills/gptpro)** | Export a monorepo into review-ready zip bundles for a non-agentic frontier model — `node_modules`-free, split by subsystem, with a secret scan that hard-aborts the export rather than shipping a key to a chat UI. |
  29. | **[gptpro-handoff](skills/gptpro-handoff)** | The workflow around it: author the prompt, receive the report, verify every finding against the *live* codebase before implementing. Includes the prompt house style. |
  30. ### Video production
  31. A local-first, agent-operated YouTube pipeline. Nothing uploads to render.
  32. | Skill | What it does |
  33. |---|---|
  34. | **[avenox-video](skills/avenox-video)** | The router — read this first. 7-step pipeline from intake to export. |
  35. | **[avenox-roughcut](skills/avenox-roughcut)** | Transcript-driven rough cut: silence removal *and* flub/retake removal, in the right order. |
  36. | **[avenox-graphics](skills/avenox-graphics)** | Brand-locked motion graphics via HyperFrames, composited onto the cut. |
  37. | **[avenox-thumbnail](skills/avenox-thumbnail)** | High-CTR thumbnail factory for a mascot-driven channel. Parallel `gpt-image-2` jobs, hook-pattern playbook, hard safety rules. |
  38. The three video skills share runtime files in **[`avenox-studio/`](avenox-studio)**
  39. — scripts, the `edit.json` template, and the brand spec:
  40. ```bash
  41. export STUDIO_ROOT="$PWD/avenox-studio"
  42. export STUDIO_JOBS="$HOME/video/projects" # heavy media — keep OUT of cloud sync
  43. cp avenox-studio/brand/frame.template.md avenox-studio/brand/frame.md
  44. cp avenox-studio/brand/caption-corrections.example.json avenox-studio/brand/caption-corrections.json
  45. ```
  46. ### Blockchain
  47. | Skill | What it does |
  48. |---|---|
  49. | **[chainscan](skills/chainscan)** | Multi-chain block explorer via the Etherscan V2 unified API + Foundry `cast` fallbacks. Contract ABI/source, txs, logs, balances, token info across 60+ chains. One key, one endpoint. |
  50. ## Bring your own assets
  51. Two skills expect files this repo deliberately doesn't ship:
  52. - **`avenox-thumbnail`** needs your own mascot reference in `assets/`. The
  53. mascot is channel identity — yours should be yours. It also needs tool logos,
  54. which are third-party trademarks; the fetch recipe is included instead of the
  55. files. See [`skills/avenox-thumbnail/assets/README.md`](skills/avenox-thumbnail/assets/README.md).
  56. - **`avenox-graphics`** reads `avenox-studio/brand/frame.md`, which you create
  57. from the template. Lock it early — visual consistency compounds, and changing
  58. it mid-channel costs more than getting it slightly wrong at the start.
  59. ## Requirements
  60. Varies by skill; each `SKILL.md` states its own.
  61. - `codex-fleet` — Codex CLI 0.128+, authenticated. Uses bash recipes; Windows uses Git Bash. On Windows/Linux omit the
  62. macOS-only `caffeinate` prefix; the two macOS-only helpers, `caffeinate` and `sips`, have bundled
  63. PowerShell replacements in `skills/codex-fleet/scripts/` that need `pwsh` 7+
  64. - `omp-fleet` — `omp` (`@oh-my-pi/pi-coding-agent`), authenticated against a
  65. provider; budget ~0.5GB RAM per concurrent lane bare, ~1.7GB with a typical
  66. MCP set auto-discovered
  67. - `gptpro` — `zip`, `rsync`
  68. - video skills — macOS (hardware encode, `mlx-whisper` on Apple Silicon),
  69. `ffmpeg`, `python3`, MLT/`melt`, Node. Most work on Linux with `libx264` and a
  70. CUDA whisper build substituted in.
  71. - `chainscan` — an Etherscan V2 API key; Foundry for `cast` fallbacks
  72. ## A note on the gotchas
  73. The sections labelled **GOTCHAS** are the highest-value part of this repo. A few
  74. that cost real hours:
  75. - `auto-editor` v29 leaks the last `--cut-out` range as a positional input file.
  76. Use ffmpeg's `select` filter for content cuts.
  77. - Codex's greedy `-i` parse eats your prompt unless you put `--` before it.
  78. - An `omp` lane measures ~1700MB against a Codex lane's ~108MB — but ~75% of
  79. that is MCP servers omp auto-discovers and boots *per lane*, not the harness
  80. (~460MB). Every `npx`-launched MCP server also keeps a resident `npm exec`
  81. parent, so you pay ~50% extra per server for nothing.
  82. - `omp` has no `exec` subcommand — non-interactive is `-p`. And its model ids
  83. fuzzy-match, so an unpinned lane can answer from a metered aggregator
  84. instead of your subscription.
  85. - Parallel `gpt-image-2` jobs share an image cache and can return duplicate
  86. renders — md5 the batch, re-fire dupes solo.
  87. - SVG `feTurbulence` grain must use a fixed seed or rendered frames flicker.
  88. - `auto-editor` and `npx` both need the certifi SSL fix or their downloads fail.
  89. ## License
  90. MIT — see [LICENSE](LICENSE). Use them, fork them, improve them.
  91. Issues and PRs welcome, especially "this broke on my setup" reports.

Ratings and comments

Sign in to rate this skill.

Sign in

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