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 skills add kelp/agent-plugins --skill fleet-efficiencygit clone --depth 1 https://github.com/kelp/agent-pluginsWrote 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/kelp/agent-plugins/fleet-efficiency)<a href="https://agentmods.dev/skills/kelp/agent-plugins/fleet-efficiency"><img src="https://agentmods.dev/badge/skills/kelp/agent-plugins/fleet-efficiency/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/kelp/agent-plugins/fleet-efficiency"><img src="https://agentmods.dev/badge/skills/kelp/agent-plugins/fleet-efficiency.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00000 | $0.01117 |
| Opus 5.5 | $0.00000 | $0.00447 |
| Sonnet 5.5 | $0.00000 | $0.00223 |
| Haiku 4.5 | $0.00000 | $0.00112 |
Grade A, and why
fleet-efficiency 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 1mo 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.
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.
How it starts
The opening of the file, as written. The whole thing — 93 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fleet Efficiency (delegation rules, context handoff)
Any dispatch (one agent or many)
- Delegate work that is large and independent. Anything you can finish in a handful of tool calls, do inline. Never spawn an agent to double-check your own work.
- Pick the most specific agent type before falling back to
general-purpose. - Run independent subagents in parallel: one message, multiple Agent calls.
- Run subagents in the background and keep working. For a follow-up, continue a long-lived agent via SendMessage instead of spawning a fresh one.
- Brief the agent like a colleague who just walked in: goal, context, what has been ruled out, expected output shape, and a length cap if you want one.
- Verify what the agent did; never trust the summary blind.
- Forks (
subagent_type: "fork") always run on the parent model;modeloverrides are ignored.
Fleets (3+ agents, Workflows, audits, migrations)
When fanning out many agents (audits, migrations, multi-stage pipelines), the token bill lives in the subagents, not the main thread. Rules:
- Scout once, brief many. The orchestrator (or one scout agent) builds the repo map / file partition ONCE; every worker gets an explicit file list scoped to its task. Never let N agents each rediscover the repo.
- Shared preamble, identical bytes. Put the shared brief (architecture nutshell, conventions, rubric) as a byte-identical block at the TOP of every fleet prompt, per-agent task at the BOTTOM. Identical prefixes hit the prompt cache across the whole fleet; a reordered or reworded brief pays full price N times.
- Excerpt, don't point. Paste the ten relevant lines into the
prompt instead of "read CLAUDE.md first". A 9k-token doc read
by 150 agents is >1M tokens. If a project has a condensed
agent brief (e.g.
docs/agent-brief.md), paste that. - Hand artifacts forward. Every pipeline stage returns structured output (schema) carrying exactly what the next stage needs: code excerpts, file:line, diffs, commands run. A downstream agent re-reads source only when its JOB is to distrust the upstream one (adversarial verification); formatters, dedupers, and drafters should need zero file reads.
- Continue, don't respawn. For fix loops on the same artifact, SendMessage the original agent; its context is intact. Inside Workflow scripts (no continuation), include the prior diff and the reviewer's issue list in the fresh agent's prompt so it doesn't re-derive them.
- Keep fleet prompts byte-stable. No timestamps, run ids, or
other volatile values inside prompts; they bust Workflow
resume caching and cross-agent prompt caching. Pass volatile
values via
argsand reference them once. - Don't optimize away independence. Verifiers re-reading the code their finder cited is intentional redundancy; cut the briefing waste, not the adversarial checks.
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.
- 1mo ago Changed · +26 lines · -59 tokens per session 550d63b11961
- 1mo ago First seen · 67 lines · 59 tokens per session scan A c4e955e512d6
fleet-efficiency is a skill published in the GitHub repository kelp/agent-plugins (2 stars, last pushed 1mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,117 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.
Other skills, from other repositories
import-audio
Moves audio files to the correct album location with proper path structure. Use when the user has downloaded WAV files from Suno or other sources that need to be organized.
claude-code-bash-patterns
Claude Code Bash tool patterns with hooks, automation, git workflows. Use for PreToolUse hooks, command chaining, CLI orchestration, custom commands, or encountering bash permissions, command failures, security guards, hook configurations.
slicing-code-context
Use when an exact symbol, path, entrypoint, or line range can bound a focused code question or patch proposal under a fixed source budget. Not for source changes or broad repository exploration.
handoff
Use when work reaches a session boundary or the user asks to hand off, delegate, or get a clipboard-ready prompt. Mode brief packages continuity; mode prompt builds the delegation prompt.
spring-boot-scanner
Smart code scanner that detects Spring Boot patterns and routes to appropriate skills. Use when editing Java or Kotlin files in Spring Boot projects, working with pom.xml/build.gradle containing spring-boot-starter, or when context suggests Spring Boot development. Detects annotations (@RestController, @Entity…
askme
Explore intent with Verbalized Sampling before planning. Use when the task is ambiguous, you need maximum clarifying questions, or you want collaborative or adversarial mode via /askme.