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.
npx agentmods add skills/himkt/cafleet/skill-authornpx skills add himkt/cafleet --skill skill-authorgit clone --depth 1 https://github.com/himkt/cafleetWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/himkt/cafleet/skill-author)<a href="https://agentmods.dev/skills/himkt/cafleet/skill-author"><img src="https://agentmods.dev/badge/skills/himkt/cafleet/skill-author.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00126 | $0.10327 |
| Opus 5 | $0.00063 | $0.05164 |
| Sonnet 5 | $0.00025 | $0.02065 |
| Haiku 4.5 | $0.00013 | $0.01033 |
Grade A, and why
skill-author scanned grade A 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 6d ago.
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.
Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
2. **Mkdir on first write**: `(Path(BASE) / "prompts").mkdir(parents=True, exist_ok=True)`. Use `pathlib`, not `subprocess.run(["mkdir", ...])`. This is enforced by the no-bypass write protocol. How it starts
The opening of the file, as written. The whole thing — 539 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill Author — Integrating the CAFleet-Orchestrated Pattern
You are about to write a new skill that drives a Director and one or more spawned members through the CAFleet message broker. This guide walks you through every sub-system you need to wire up, explains why each step exists, and finishes with a worked example you can read end-to-end. It is a teaching document — not a paste-this template — because the canonical rules drift quickly and skills that copy a prefab spawn prompt rot the fastest.
Read this whole document before you start writing your SKILL.md. The rules are short individually but the failure modes when any of them is missed are loud, and most of them have already bitten earlier authors.
1. What "CAFleet-orchestrated" means
A CAFleet-orchestrated skill is one where the Director (the main Claude session running your slash command) bootstraps a fresh CAFleet fleet, spawns one or more members as separate claude (or codex) processes inside dedicated tmux panes, and coordinates the work through the CAFleet message broker (cafleet message send / cafleet message poll / cafleet message ack). Members are real, isolated coding-agent processes — not in-process subagents — and the broker delivers each message as a 2-line keystroke preview into the recipient's tmux pane.
The shape always looks like this:
User
+-- Director (main Claude — runs cafleet fleet create / member create; coordinates via the broker)
+-- member-1 (claude pane)
+-- member-2 (claude pane)
+-- ...
Use this pattern when:
- The work needs parallelism that the harness cannot provide on its own — multiple members drafting in parallel, multiple researchers investigating sub-topics, multiple verifiers running on different slide ranges.
- The work needs role specialization — Director / Drafter / Reviewer / Programmer / Tester are roles that justify separate processes with separate role files.
- The work needs persistent inter-member memory — the broker stores every message in SQLite and the audit-file path under
${BASE}/prompts/is a permanent record.
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.
- 6d ago First seen · 539 lines · 126 tokens per session scan A 71f88a395615
skill-author is a skill published in the GitHub repository himkt/cafleet (49 stars, last pushed today), licensed MIT. It adds 126 tokens to every session and 10,327 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
anycap-human-interaction
Collect structured visual feedback from humans using AnyCap's annotation tool, or create and iterate on diagrams using the interactive whiteboard (Excalidraw). Covers image annotation, URL/web page review with screen recording, video review, audio feedback, and collaborative diagramming with Mermaid input. Use when…
anycap-media-production
Produce media assets using AnyCap: generate images, videos, music, speech, dialogue, and complete audio scenes from text or reference inputs, refine images through interactive visual annotation, and deliver finished assets. Covers the full production workflow from concept to delivery across all media types (image…
anycap-blog-production
Turn user-provided data, notes, research, product facts, or structured inputs into AnyCap-style blog posts and then enrich those posts with first-party evidence blocks when needed. Use when a user wants a blog, article, tutorial, learn page, or SEO content draft written in the tone of the AnyCap website. Best fit when…
anycap-worldcup-predict
World Cup match prediction and pre-match intel for football (soccer) fans: predict who will win and a likely scoreline, backed by structured, verifiable evidence -- player profiles, squad dossiers, current form, head-to-head records, and past-tournament history -- using AnyCap web search and crawl. Weighs FIFA…
anycap-ai-tool-seo
Guide for planning and auditing SEO for AI tool, SaaS, and product-led websites. Powered by AnyCap -- the capability runtime that equips AI agents with web search and web crawl through a single CLI. Use when Codex needs to define SEO ICPs, map search intent to page types, inspect live SERPs, write page briefs for…
anycap-deepresearch
Guide for conducting thorough, multi-source research and producing comprehensive, well-sourced reports. Powered by AnyCap -- the capability runtime that equips AI agents with web search (including AI Grounded citations), web crawl, image generation, cloud storage, and one-click web publishing through a single CLI. Use…