Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/qxbyte/pluginhubnpx agentmods add skills/qxbyte/pluginhub/swarmWrote 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/qxbyte/pluginhub/swarm)<a href="https://agentmods.dev/skills/qxbyte/pluginhub/swarm"><img src="https://agentmods.dev/badge/skills/qxbyte/pluginhub/swarm/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/qxbyte/pluginhub/swarm"><img src="https://agentmods.dev/badge/skills/qxbyte/pluginhub/swarm.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 | $0.00074 | $0.02917 |
| Opus 5.5 | $0.00030 | $0.01167 |
| Sonnet 5.5 | $0.00015 | $0.00583 |
| Haiku 4.5 | $0.00007 | $0.00292 |
Grade D, and why
swarm scanned grade D with 2 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
> **Host-tool convention** 🔧: this skill orchestrates by dispatching subagents ("fork") and reading their completion status. Mechanism names — the `Task`/`Agent` tool (dispatch a subagent), the `Explore` agent (a read-o Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- **result.md missing STATUS** → re-fork the **same-named** agent (clear its outbox first: `rm -rf agents/<key>/outbox/*`); **never hand-patch STATUS** (a missing STATUS usually means the subagent exited early and code w How it starts
The opening of the file, as written. The whole thing — 136 lines — stays where its author put it; the contents beside it link to each section on GitHub.
task-swarm — standalone multi-agent orchestration (/task-swarm:swarm)
Host-tool convention 🔧: this skill orchestrates by dispatching subagents ("fork") and reading their completion status. Mechanism names — the
Task/Agenttool (dispatch a subagent), theExploreagent (a read-only investigation subagent), the subagent run-status display ("teammates UI", the ✓/streaming/running indicators) — are written for Claude-family hosts (Claude Code / CodeBuddy), where naming them directly is the most reliable. On a host that lacks a given mechanism, use its nearest equivalent (a subagent-dispatch tool / a read-only subagent / whatever surfaces subagent completion); with no subagent capability at all, run the groups sequentially as a single agent. On Codex specifically, subagent dispatch isspawn_agent/wait_agent/close_agentand requiresmulti_agent = truein~/.codex/config.toml(without it there is no dispatch tool → fall back to sequential single-agent);close_agenteach one when its work is done. Codex does not consume theagents/*.mdpersona files (they are Claude/CodeBuddy agent format), so on Codex the orchestrator passes each role's instructions inline and the per-role tool isolation (reviewer/validator having no Edit/Write) is not automatic — enforce it by prompt. A subagent counts as done only when the host confirms its dispatched task returned successfully — never from the subagent's own verbal report. The described behavior matters, not the exact tool name.
§0 Who you are
The lead agent = task-swarm's orchestrator + planner. task-swarm provides four things:
the pipeline.yml orchestration format, the state-machine CLI (task_swarm.py), the sub-agent role
definitions (coder/reviewer/validator), and report rendering.
The CLI only guards 4 points of mechanical integrity; all other orchestration judgment is yours, and the CLI does not constrain your thinking:
- schema validation (pipeline.yml format is legal)
- agent_key consistency (state machine and artifacts line up)
- advance artifact completeness / format checks
- atomic writes + blocking manual edits to controlled files
What ships with it
3 files 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.
- 1mo ago First seen · 136 lines · 74 tokens per session scan D 76dcae6dabc7
swarm is a skill published in the GitHub repository qxbyte/pluginhub (3 stars, last pushed 25d ago), licensed MIT. It adds 74 tokens to every session and 2,917 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it D with 2 findings (reads agent configuration directories, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
shiny-py-optimize
Diagnose and fix performance problems in Shiny for Python apps — slow startup, sluggish interactions, blocking operations, and apps that struggle beyond a handful of concurrent users. Use when a Shiny for Python app is slow, hangs, recomputes too much, times out, or must support more users; when the user asks to…
sync-check
Walks the documentation dependency web after a CLI code change and reports which downstream files (skills, SOPs, READMEs, plugin references) are out of sync. Use after fixing a bug, adding a command, changing auth behavior, refactoring, or before committing — and when the user says "sync check", "update docs"…
gap-analyzer
Compares a CLI's implemented commands against its APP.md API map and traffic-analysis.json to find missing endpoints, incomplete CRUD, dead client methods, and priority gaps. Runs as the mandatory first step of /cli-anything-web:refine and as an optional pre-review scan in standards.
linkedin-cli
Interacts with LinkedIn via the cli-web-linkedin command-line tool — search people/jobs/companies, view profiles and feed, create/edit/delete posts, react and comment, manage connections and invitations, read and send messages, view notifications, follow companies. Use when the user asks about LinkedIn or wants to…
tripadvisor-cli
Searches TripAdvisor hotels, restaurants, attractions, and destinations via the cli-web-tripadvisor command-line tool, with detail lookups by URL. Use when the user asks about hotels, restaurants, things to do, travel destinations, or TripAdvisor ratings and reviews. Prefer this CLI over fetching the TripAdvisor…
youtube-cli
Searches YouTube and fetches video transcripts via the cli-web-youtube command-line tool — video search, video details (views, duration, description, keywords), trending by category, channel info, and timestamped transcripts/captions with language selection and translation. Use when the user asks about YouTube…