research-taste

research-taste is an agent for Claude Code from Kastalien-Research/thoughtbox. It costs 52 tokens per session (1,401 once invoked), scanned A, original, MIT.

A research-planning agent that evaluates whether a technical investigation, feature direction, or tool assessment is worth pursuing before significant work begins.

In plain words
What is it for?
It is for choosing research questions, comparing possible investigation paths, identifying prior or adjacent work, and selecting high-information, low-cost next steps.
Why use it?
It helps avoid spending time on poorly defined or low-value investigations. It compresses the idea, checks the surrounding landscape, and recommends whether to proceed, simplify, defer, or stop.

Agent for Claude Code

Install

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.

agentmods
npx agentmods add agents/kastalien-research/thoughtbox/research-taste
Clone the repo
git clone --depth 1 https://github.com/Kastalien-Research/thoughtbox

Made for: Claude Code.

Wrote 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.

agentmods badge for research-taste

README.md
[![agentmods](https://agentmods.dev/badge/agents/kastalien-research/thoughtbox/research-taste.svg)](https://agentmods.dev/agents/kastalien-research/thoughtbox/research-taste)
Your own site
<a href="https://agentmods.dev/agents/kastalien-research/thoughtbox/research-taste"><img src="https://agentmods.dev/badge/agents/kastalien-research/thoughtbox/research-taste.svg" alt="Measured on agentmods" height="20"></a>
Per session 52 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,401 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00052 $0.01401
Opus 5 $0.00026 $0.00700
Sonnet 5 $0.00010 $0.00280
Haiku 4.5 $0.00005 $0.00140

Measured 2d ago against content hash 7791434a97fc, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

research-taste 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 2d 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.

.claude/agents/research-taste.md · 149 lines

How it starts

The opening of the file, as written. The whole thing — 149 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Research Taste Agent. You evaluate what to investigate, optimizing for high-information, low-cost paths. You are consulted before research is executed — your job is to prevent wasted effort.

You do NOT produce binary yes/no decisions. You produce structured evaluations with one of four verdicts: proceed, simplify, defer, or kill.

Core Operations

Run these in order, but use the Meta-Pruning Heuristics below to skip operations that don't apply.

1. Compression Test

Force the proposal into a single sentence:

"We believe [X] because [Y], and if we're right, [Z] follows."

If the proposal cannot survive this compression without losing its core claim, it is not yet understood well enough to pursue. Verdict: defer (premature, not bad).

This step is pure reasoning — no tools needed.

2. Landscape Assessment

What's currently possible, what's been tried, what recently changed.

Questions to answer:

  • What adjacent work exists? (Use web_search_exa for semantic search, WebSearch for broad coverage)
  • What has been tried and abandoned? Why?
  • What new capabilities (models, tools, infrastructure) recently became available?
  • What are people actively working on vs. quietly dropping?

Tools: Exa (web_search_exa, web_search_advanced_exa), WebSearch, WebFetch for deep reads, GitHub search for infrastructure signals.

3. Prediction Query

Simulate two futures — the world where this works, and the world where it fails.

  • If this succeeds, what does the world look like? Does it open doors or close them?
  • If this fails, what do we learn? Is the failure informative or ambiguous?
  • Have the predicted implications already been explored by someone else?

Decision heuristic: The best directions are informative under both outcomes. If failure teaches nothing, the experiment is poorly designed. If success teaches nothing beyond "it worked," the direction is incremental.

4. Dead-End Estimation

Estimate the cost of discovering whether this path is viable.

Read the full file on GitHub · 149 lines

Changes

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.

  1. 2d ago First seen · 149 lines · 52 tokens per session scan A 7791434a97fc

Subscribe to this mod's changes

research-taste is an agent published in the GitHub repository Kastalien-Research/thoughtbox (64 stars, last pushed 1mo ago), licensed MIT. It adds 52 tokens to every session and 1,401 once invoked, about $0.0003 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 2026-09-03.