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 softspark/ai-toolkit --skill swarmgit clone --depth 1 https://github.com/softspark/ai-toolkitWrote 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/softspark/ai-toolkit/swarm)<a href="https://agentmods.dev/skills/softspark/ai-toolkit/swarm"><img src="https://agentmods.dev/badge/skills/softspark/ai-toolkit/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/softspark/ai-toolkit/swarm"><img src="https://agentmods.dev/badge/skills/softspark/ai-toolkit/swarm.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 9 Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.Fix: Remove the model/provider override or disclose it prominently and require explicit operator approval before invoking an external coding CLI or billed model.
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 | $0.00035 | $0.01774 |
| Opus 5.5 | $0.00014 | $0.00710 |
| Sonnet 5.5 | $0.00007 | $0.00355 |
| Haiku 4.5 | $0.00003 | $0.00177 |
Grade A, and why
swarm 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 13d 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 — 184 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/swarm - Parallel Agent Swarm
$ARGUMENTS
Only in Claude Code, apply the model-routing-patterns skill when choosing
executors or creating agent definitions. Delegate to codex:codex-rescue only
when its plugin is installed, enabled and callable in this session. Otherwise
use the installed native agents and their configured models. A context without
the Agent tool returns the dispatch decision to its supervisor; it does not
invent a tool or bypass the client. Preserve explicit user choices and verify
actual completion before accepting a delegated result.
MANDATORY: You MUST use the Agent tool
When delegation is available, decompose the task and invoke agents via parallel
Agent tool calls for independent work. If the runtime has no delegation support,
use the current session and explicitly report single-agent execution. Never
present simulated agents as a real swarm.
Modes
Map-Reduce (default)
Split task into N independent sub-tasks. Launch ALL agents in a single response (parallel execution).
# Single response with N Agent tool calls:
Agent(subagent_type="...", prompt="sub-task 1 — own files: path/a/")
Agent(subagent_type="...", prompt="sub-task 2 — own files: path/b/")
Agent(subagent_type="...", prompt="sub-task N — own files: path/n/")
After all complete: aggregate results (see Aggregation section below), produce synthesis report.
Consensus
Same problem, 3 independent agents from different angles. Launch all 3 in a single response.
Agent(subagent_type="backend-specialist", prompt="[problem] — approach from data layer angle. Output: solution + confidence 0.0–1.0")
Agent(subagent_type="tech-lead", prompt="[problem] — approach from architecture angle. Output: solution + confidence 0.0–1.0")
Agent(subagent_type="performance-optimizer", prompt="[problem] — approach from performance angle. Output: solution + confidence 0.0–1.0")
After all complete: compare evidence and acceptance criteria, record dissent, and validate the proposed result. Self-reported confidence is advisory; it is not a calibrated probability and must not decide the winner on its own.
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.
- 13d ago Changed · +13 lines 6d0ab73daa3a
- 22d ago Changed · +2 lines 215977c5762f
- 1mo ago First seen · 169 lines · 35 tokens per session scan A 0767d98c9c4f
swarm is a skill published in the GitHub repository softspark/ai-toolkit (180 stars, last pushed yesterday), licensed Apache-2.0. It adds 35 tokens to every session and 1,774 once invoked, about $0.0001 per session on Opus 5.5. 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 2026-09-03.
Other skills, from other repositories
cline-fix-volatile-msg
Ladder-aware Cline Anthropic caching — verify the rolling read/write ladder on the wire, then add the tools breakpoint and tune TTL. Updated for the 2026-08 AI-SDK monorepo.
continue-gemini-explicit
Continue's Gemini provider doesn't use the cachedContents API at all. Add explicit caching for sessions over the minimum token threshold.
roo-fix-volatile-msg
Ladder-aware Roo Code Anthropic caching — verify the rolling read/write ladder on the wire, then close the real gaps (Vertex 4-block budget, MiniMax path).
aider-1h-ttl
Aider uses 5min TTL by default and works around long pauses with keepalive pings. Wire up the 1h TTL beta instead.
cline-openai-cache-key
Cline OpenAI native provider sends no promptcachekey. Add a stable per-task key so cachedtokens stops being zero.
continue-enable-defaults
Continue's prompt caching is opt-in via config and off by default. Flip the default to systemAndTools.