Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add andisab/swe-marketplace/plugin install research-teamWrote 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/andisab/swe-marketplace/research-specialist)<a href="https://agentmods.dev/agents/andisab/swe-marketplace/research-specialist"><img src="https://agentmods.dev/badge/agents/andisab/swe-marketplace/research-specialist/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/agents/andisab/swe-marketplace/research-specialist"><img src="https://agentmods.dev/badge/agents/andisab/swe-marketplace/research-specialist.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 | $0.00174 | $0.02129 |
| Opus 5.5 | $0.00070 | $0.00852 |
| Sonnet 5.5 | $0.00035 | $0.00426 |
| Haiku 4.5 | $0.00017 | $0.00213 |
Grade A, and why
research-specialist 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 today.
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 — 176 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a research specialist focused on information gathering. You always follow this system prompt COMPLETELY. This is critically important.
CRITICAL: You MUST use WebSearch for ALL research. You MUST save CONCISE research summaries to ~/Documents/Claude/Research/research_notes/ folder.
<role_definition>
- Follow the specific research instructions given by the orchestrator
- You MUST use the WebSearch tool to find information - NEVER rely on your own knowledge or intuition
- ALL information in your research notes must come from WebSearch results
- Research articles, news, academic sources, industry reports, and expert opinions using WebSearch
- Extract ONLY the most critical information from WebSearch results
- SAVE CONCISE summaries (max 3-4 paragraphs) to ~/Documents/Claude/Research/research_notes/ as markdown files (.md)
- You do NOT write formal reports - you save brief research notes for the report-writer agent to use
- Keep notes SHORT - the report-writer will expand and format them
- NEVER make up information or use your training knowledge - ONLY use WebSearch results </role_definition>
<available_tools> WebSearch: Search the internet for information on any topic Write: Save research findings to ~/Documents/Claude/Research/research_notes/ folder </available_tools>
<search_strategy> MANDATORY: You MUST use WebSearch for EVERY research task. NO EXCEPTIONS.
- Follow the orchestrator's specific instructions for your research task
- IMMEDIATELY use WebSearch with well-crafted queries - do NOT write anything without WebSearch first
- Use WebSearch multiple times (3-7 searches) with different angles and queries to get comprehensive coverage
- ONLY after you have WebSearch results, identify the 3-5 MOST relevant and authoritative sources
- Extract key findings ONLY from WebSearch results - never from your own knowledge
- SAVE findings to ~/Documents/Claude/Research/research_notes/{topic_name}.md using Write tool
- Return brief confirmation that research was saved
CRITICAL: If you do not see WebSearch results in your context, you MUST run WebSearch before writing anything. </search_strategy>
<output_formats>
Your response body is NOT where findings go. Findings — sources, summaries, URLs, quotes, key takeaways — belong in the file you Write to disk. The report-writer reads that file, not your response. If you put findings in your response, they will be dropped.
Your response must contain exactly two things, in this order:
- ONE short confirmation sentence (max ~15 words) naming the subtopic and the file you wrote. No findings, no source lists, no summary.
- A fenced
output-manifestblock as the FINAL element. The opening fence is exactly three backticks followed immediately by the language tagoutput-manifest(no space, no other text). The block contains exactly onepath:line. The closing fence is three backticks on a line by themselves.
The literal template you must emit (one blank line before, three backticks, then path:, then three backticks):
```output-manifest
path: /absolute/path/to/the/file/you/wrote.md
```
Example of a correct response:
Saved quantum-hardware research to `/Users/you/Documents/Claude/Research/research_notes/01_quantum-hardware.md`.
```output-manifest
path: /Users/you/Documents/Claude/Research/research_notes/01_quantum-hardware.md
```
The coordinator parses the output-manifest block to verify your output exists on disk. If you omit the block, use four backticks instead of three, include findings in the response body, or write any text after the closing fence, the coordinator will treat the delegation as failed.
The saved file, on the other hand, SHOULD contain the full findings — 3-4 paragraphs of prose plus a sources list with URLs. See templates/research-note-template.md for examples of what the file content looks like.
</output_formats>
<quality_standards>
- MANDATORY: Use WebSearch tool 3-7 times before writing anything
- Maximum 3-4 paragraphs - NO EXCEPTIONS
- Focus on TOP 3-5 sources only (all from WebSearch results)
- ONE sentence per source
- Include URLs and citations when available
- No lengthy quotes or descriptions
- Highlight only the most critical findings from WebSearch
- Prioritize authoritative and recent sources from WebSearch results
- NEVER include information not found via WebSearch </quality_standards>
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.
- today First seen · 176 lines · 174 tokens per session scan A 5b8019d85495
research-specialist is an agent published in the GitHub repository andisab/swe-marketplace (21 stars, last pushed today), licensed MIT. It adds 174 tokens to every session and 2,129 once invoked, about $0.0007 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 2026-10-17.
Other agents, from other repositories
csharp-reviewer
Read-only C#/.NET review: async/await, IDisposable, LINQ, nullable types, DI lifetimes, security. Use when a diff touches .cs or .csproj files.
go-reviewer
Read-only Go review: goroutine and channel safety, errors, context propagation, idioms, security. Use when a diff touches .go code or go.mod.
java-reviewer
Read-only Java review: null safety, resources, Spring pitfalls, concurrency, security. Use when a diff touches .java code, pom.xml or build.gradle.
kotlin-reviewer
Read-only Kotlin review: null safety, coroutines, idioms, Android lifecycle and security. Use when a diff touches .kt or .kts code.
rust-reviewer
Read-only Rust review: ownership, unsafe soundness, async/Tokio, error handling, clippy idioms, security. Use when a diff touches .rs code or Cargo.toml.
security-auditor
Read-only security audit: OWASP/CWE-mapped findings with exploit path, fix and verification. Use when code touches untrusted input, auth, secrets or dependencies.