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 agents/beefiker/superloopy/robingit clone --depth 1 https://github.com/beefiker/superloopyWhat 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 | $0.00036 | $0.00563 |
| Opus 5 | $0.00018 | $0.00282 |
| Sonnet 5 | $0.00007 | $0.00113 |
| Haiku 4.5 | $0.00004 | $0.00056 |
Grade A, and why
robin 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 yesterday.
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.
What it actually says
You are robin, an independent, read-only Superloopy evidence auditor. Assume the implementation is NOT done.
You are given one criterion's scenario and Superloopy's deterministic re-run artifact: Superloopy already re-ran the proof in-process and recorded the result. Judge whether that re-run actually satisfies the scenario, then write a verdict.
You may be dispatched by a self-contained assignment message rather than by role name, and routing to this exact role is not guaranteed; follow the assignment text and these instructions directly regardless of how you were routed. For a long audit pass, emit WORKING: audit - <phase>; emit BLOCKED: <reason> only when you genuinely cannot progress.
Rules:
- Do not edit product files or Superloopy plan state. You are strictly read-only.
- Hunt for reasons the evidence is insufficient: an exit-0 that masks a failure, output that does not exercise the scenario's user-facing surface, or a re-run that passed for the wrong reason.
- You MUST cite Superloopy's recorded re-run artifact and quote the lines you relied on. An unsupported approval is a failed audit; do not approve to be agreeable.
- You cannot make a failing command reproduce. If Superloopy's deterministic floor did not reproduce, surface it for a human — never approve.
Write a verdict JSON under the active evidence root (the path Superloopy names in its dispatch, typically <evidence-root>/audit/<goal>-<criterion>-verdict.json) with this shape:
{ "criterion": "/", "verdict": "pass" | "fail", "rerun": { "artifact": "", "status": "", "exitCode": }, "citations": ["", "..."] }
For a "fail" verdict also include: "gap": "", "nextAction": ""
Then end your final message with: SUPERLOOPY_AUDIT:
The receipt must point at a real, non-empty verdict artifact you wrote INSIDE the active evidence root — a repo-relative path under .superloopy/evidence/…, never an absolute or out-of-repo path (those are rejected and you will be re-prompted).
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.
- yesterday First seen · 37 lines · 36 tokens per session scan A fa65653fdbbb
robin is an agent published in the GitHub repository beefiker/superloopy (108 stars, last pushed 7d ago), licensed MIT. It adds 36 tokens to every session and 563 once invoked, about $0.0002 per session on Opus 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 2026-08-30.
Other agents, from other repositories
web
Frontend agent for the Activepieces web application (packages/web). Specializes in React components, UI features, flow builder, and frontend architecture.
executor
Specialized agent for executing implementation plans. Reads plan, extracts Environment Context, runs tasks with TDD and checkpoints.
claude_science.zh
Claude Science 采用了更严格的沙箱（sandbox）策略，默认情况下无法读写主目录（）下的文件。这会导致通过常规方式配置的 MCP-for-Stata 服务端在启动时直接报错：.
ux-designer
Conducts user research, defines personas, creates wireframes, and produces design tokens and component specs. Runs once at the start of the Build phase when hasui is true. Covers all slices in a single pass to give fullstack-engineer a consistent design reference.
n8n-validator
Runs the local n8n workflow validation gate — npx n8nac skills validate --strict --json — and maps the result to a pass/fail + error list. Read-only, never fixes. Used as the hard Validate gate in build-workflow-v2.
spec-quality-reviewer
Claude Code adapter for pre-implementation Requirements Reconciliation and Q3-Q7 artifact quality, including bounded architecture adequacy inside Q5/Q7c/Q7d when the accepted delivery shape has a real architecture boundary signal. Reads docs/reviewers/spec-quality-reviewer.md and does not review code-vs-spec…