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 dryvist/claude-code-plugins --skill local-subagentsgit clone --depth 1 https://github.com/dryvist/claude-code-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/dryvist/claude-code-plugins/local-subagents)<a href="https://agentmods.dev/skills/dryvist/claude-code-plugins/local-subagents"><img src="https://agentmods.dev/badge/skills/dryvist/claude-code-plugins/local-subagents/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/dryvist/claude-code-plugins/local-subagents"><img src="https://agentmods.dev/badge/skills/dryvist/claude-code-plugins/local-subagents.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 | $0.00058 | $0.02761 |
| Opus 5.5 | $0.00023 | $0.01104 |
| Sonnet 5.5 | $0.00012 | $0.00552 |
| Haiku 4.5 | $0.00006 | $0.00276 |
Grade A, and why
local-subagents 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 10d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -fsS --max-time 10 -H "Authorization: Bearer $ROUTER_KEY" \ How it starts
The opening of the file, as written. The whole thing — 238 lines — stays where its author put it; the contents beside it link to each section on GitHub.
local-subagents
Doctrine: the project's delegation guidance. This skill is how you act on it — when to hand work out, how to find out what is serving right now, and how to place the call.
Your own inference is the scarcest and most expensive thing in the session. Local models may cost nothing per token and sit idle; use them when they can do the work.
1. When to delegate — the default is yes
Delegate a step when its result can be checked from concrete evidence:
| Delegate | Keep |
|---|---|
| Summarizing a file, log, or diff | What the user actually wants |
| Classifying or triaging a batch | Architecture and design choices |
| Extracting structured fields into a schema | Tradeoffs, risk, security judgment |
| A first pass over unfamiliar code ("where is X handled") | Resolving contradictory evidence |
| Drafting boilerplate text from a stated pattern | Reviewing anything risky |
| Reading and reducing command or test output | The final answer to the user |
| Code edits — the router never writes code; implementation chunks go to Codex or a scoped worker subagent |
Two rules that keep this honest:
- Split the job; never escalate the whole job. One hard step inside a task does not make the task premium work. Delegate the other eight steps.
- Capable judges the subtask, not the parent task. A 9B model summarizing a log is the right tool even inside a hard architectural task.
- Router tiers earn re-checkable lookups and bulk reads, not code edits. Delegate to a router tier only when the result can be verified against concrete evidence (a grep hit, a test pass, a schema match). Route code edits to Codex or a scoped worker subagent. A misread search or a silently wrong edit from a cheap tier costs more than doing it yourself.
1b. Cheap first, escalate on checked failure
When a delegated batch has a verification command (a test, a lint, a diff against expected output), run the cheapest capable tier first and re-run only the failures at the next tier up:
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 10d ago Changed · +1 lines 8de9fbe5c6e2
- 11d ago Changed d9107aee6913
- 13d ago Changed 161f344eb035
- 15d ago Changed 4b00bb20a43a
- 1mo ago Changed · +22 lines · -62 tokens per session c665bd5150cf
- 1mo ago Changed · +30 lines 41968ecc9c05
- 1mo ago First seen · 185 lines · 120 tokens per session scan A 827ffc8d897e
local-subagents is a skill published in the GitHub repository dryvist/claude-code-plugins (3 stars, last pushed today), licensed Apache-2.0. It adds 58 tokens to every session and 2,761 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-06.
Other skills, from other repositories
detect-github-url
Detects when the user pastes or mentions a GitHub URL that may contain Claude skills or plugins. Triggers on messages containing github.com URLs, especially when user says things like "here's a repo", "check out this repo", "load this", "use these skills", "I found these skills", or simply pastes a github.com link.…
example-feature-generator
Generate boilerplate for new features following project conventions. Use when user says "create feature", "add feature", or "new feature scaffolding".
media-gemini-3-image-prompt
A prompt builder for Gemini 3 Pro Image, also known as Nano Banana Pro, Google's image-generation model. It expands a short request into five parts covering the subject, action, setting, composition, lighting, style, and text constraints.
commerce-jtbd-persona
An e-commerce customer-analysis skill that studies reviews and questions to identify why people buy and to create concrete customer profiles. JTBD means “jobs to be done”: the practical goal a customer is trying to accomplish.
data-realestate
A lookup tool for reported South Korean property transactions using Ministry of Land data. It covers sales and rental deals for apartments, officetels, multi-family homes, houses, and commercial or office properties.
design-sync-upload
An uploader for design-system files such as DESIGN.md, tokens, logos, fonts, and images into Claude Design. It can either use an authenticated connection or prepare a folder and guide for manual upload.