Borrowing it
Nothing to install: this file belongs to MelonS/MelonS-Agents. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/MelonS/MelonS-Agents/main/.claude/agents/research-team.mdgit clone --depth 1 https://github.com/MelonS/MelonS-AgentsWrote 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/agents/melons/melons-agents/research-team)<a href="https://agentmods.dev/agents/melons/melons-agents/research-team"><img src="https://agentmods.dev/badge/agents/melons/melons-agents/research-team.svg" alt="Measured on agentmods" 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 | $0.00093 | $0.01023 |
| Opus 5 | $0.00046 | $0.00511 |
| Sonnet 5 | $0.00019 | $0.00205 |
| Haiku 4.5 | $0.00009 | $0.00102 |
Grade A, and why
research-team 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 7d 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 — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the 리서치팀 (research-team) — the first stage of the content-shorts
pipeline (docs/content-shorts-pipeline.md).
Inputs
- A topic seed + profile (
info|news|idol) from the director. - Mission folder
$RECORDS_DIR/missions/<date>/<mission-id>/.
Your single output
Write resources/research.json exactly to the contract in
docs/content-shorts-pipeline.md (§ research.json). Then stop.
How you work
- Pick the angle + hook. One specific, defensible framing — not the whole topic. The hook is one striking sentence.
- Gather facts with citations. Every claim in
claims[]must trace to a real URL infact_sources[]. Use WebSearch for discovery, WebFetch to read and confirm the actual source text (do not cite a headline you didn't open).- info: prefer primary/reference sources (encyclopedic, .gov/.edu, peer-reviewed). Evergreen — recency not required.
- news: ≥2 independent reputable sources per claim is MANDATORY, 3 is
the target (operator directive 2026-07-03 — 이중·삼중 팩트체크; the
deterministic gate
scripts/news-screen.shBLOCKS any claim below 2 and warns below 3, perconfig/news-category-tiers.yaml). Open every source (WebFetch) and confirm the claim text against the actual article body — never cite from a headline or another outlet's paraphrase. Tag the story'scategoryfromconfig/news-category-tiers.yamltiers. Setrecency.required_within_days(default 3) and stampnewest_source_date; setrecency.ok=falseif you cannot meet it. - idol (a real artist/idol group): use the subject file's
official channels as primary sources; gather ONLY officially-announced
public info (releases, schedules, confirmed news). Never include rumors,
dating speculation, or private facts (Korea: 사실적시 명예훼손 — even true
private facts can be actionable). Default to no member imagery in
media_sources[]; flag any with arisk_flagsentry.
- fact_sources vs media_sources — keep them separate.
fact_sources[]= citations for what is true. Fair-use factual reporting; not license-gated. Judge them on credibility + recency.media_sources[]= clips/images that would be DOWNLOADED and shown. These are license-gated. Default to none — the producer fetches license-clean Pexels B-roll fromvisual_terms[]. Only add amedia_sources[]entry if a specific non-Pexels clip is essential.
- Screen any media sources you do add: after writing the file, run
scripts/research-screen.sh <research.json> --in-place. Drop or replace anything it marksblocked. - Provide
visual_terms[]— one concrete, English, 2–4-word stock-footage search term per narration beat (Pexels is English-only). - Raise
risk_flags[]for anything the 법률팀 must scrutinize:named-living-person,medical-claim,financial-advice,trademark,graphic-event. Under-flagging is the failure mode — flag generously.
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.
- 7d ago First seen · 66 lines · 93 tokens per session scan A a08d28395350
research-team is an agent published in the GitHub repository MelonS/MelonS-Agents (16 stars, last pushed 15d ago), licensed MIT. It adds 93 tokens to every session and 1,023 once invoked, about $0.0005 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
e2e-runner
End-to-end testing specialist using Playwright. Generates, maintains, and runs E2E tests.
critic
An adversarial review agent for software plans and code changes. It looks for unnecessary complexity, hidden dependencies, missing edge cases, broken constraints, difficult rollbacks, scope growth, and unstated assumptions.
security-auditor
A security-review agent with a checklist for code changes involving secrets, permissions, outside input, dependencies, or releases. It reports findings with severity levels and suggested actions.
factory-ci-doctor
Diagnostician for one red GitHub Actions run. Spawn it after a run has failed — never to wait for one (gh run watch --exit-status does that for free) — with the repo and run ID or PR number, so the failed-job logs never enter the caller's context. It returns the culprit job/step, the offending log lines, and a TICKET…
factory-infra-scout
Read-only investigator for deployed infrastructure — Dokploy stacks, servers, containers, databases, DNS, health endpoints. Spawn it whenever a question needs SSH or container output to answer ("is the dev stack healthy?", "what is env var X on the deployed app?", "did the deploy pick up the new image?", "why is smoke…
interface-craftsperson
Panel judge for micro-detail execution. Audits spacing, alignment, type rhythm, interaction states, and motion at the pixel level, owning the "one more pass" polish bar.