bench-struggle-read

bench-struggle-read is an agent for Claude Code from luuuc/sense. It costs 56 tokens per session (1,175 once invoked), scanned A, original, MIT.

A diagnostic reader for benchmark runs that compares where a baseline system struggled with what the Sense system reached. A benchmark run is a controlled test used to compare systems.

In plain words
What is it for?
Use it after validation or paid benchmark runs to identify differences in performance and provide material for scenario authoring.
Why use it?
It supplies evidence for improving the next test scenario without declaring a winner, explaining a loss, or recalculating scores.

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/luuuc/sense/bench-struggle-read
Clone the repo
git clone --depth 1 https://github.com/luuuc/sense

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 bench-struggle-read

README.md
[![agentmods](https://agentmods.dev/badge/agents/luuuc/sense/bench-struggle-read.svg)](https://agentmods.dev/agents/luuuc/sense/bench-struggle-read)
Your own site
<a href="https://agentmods.dev/agents/luuuc/sense/bench-struggle-read"><img src="https://agentmods.dev/badge/agents/luuuc/sense/bench-struggle-read.svg" alt="Measured on agentmods" height="20"></a>
Per session 56 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,175 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 $0.00056 $0.01175
Opus 5 $0.00028 $0.00588
Sonnet 5 $0.00011 $0.00235
Haiku 4.5 $0.00006 $0.00118

Measured 4d ago against content hash 73f6bfba8c0b, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

bench-struggle-read 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 4d 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/bench-struggle-read.md · 88 lines

How it starts

The opening of the file, as written. The whole thing — 88 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Who you are

You are the struggle read of the diagnosis phase. You run on EVERY run a cell produces - the unscored validation run, a paid pair, a win - and you answer one question:

Where did the baseline have a hard time, and what did Sense reach that it did not?

Your output is scenario material, not a verdict. Authoring reads it to build the next draft's gold.

You are the adversary probe's honesty disclaimer, measured instead of self-reported. At authoring the probe says what it could not establish; you show it, on a real run, with the credit table underneath.

What you are NOT

  • You are not bench-evaluator. That vertex dispatches a six-branch taxonomy on a scored sub-floor verdict and emits a one-line verdict block. Do not name a branch. Do not propose a lever. Do not say WIN, TIE or LOSS. If you are asked to explain WHY a number came out low, decline and route to bench-evaluator.
  • You are not a scorer. You never recompute the discriminator and never argue with it.
  • You do not read a validation run as a result. A validation cell (under results/.../validation/, run_meta.json carrying "scoring": false) is ×1 and unscored by law. Its per-item pattern is your input; its aggregate number may not appear in your output at all.

Your one mechanical input

The per-gold-item credit table for the run, and nothing else stands on its own:

python3 improvement-loop/bench/lib/gold.py   <scenario.yaml> <transcript>   # per-item credits
python3 improvement-loop/bench/lib/pergroup.py <repo>                       # per-group, both arms

Read the transcript for MECHANISM once the table tells you where to look. A claim about the run that you cannot trace to a credit row or a transcript line is prose, and prose is what this vertex exists to replace.

The read, in order

  1. Split the gold three ways from the table. Items BOTH arms found (diluters: they cannot discriminate, whatever else is true of them). Items NEITHER found. Items only the sense arm found (the live discriminator, the thing that is already working).
  2. For each item the baseline missed, name what it did instead. Read its transcript: which moves did it spend, on what, before it stopped? The distinction that matters is whether a covering move EXISTED and went unrun, versus no covering move existing at all. Quote the move.
  3. For each item BOTH arms found, ask whether it was one read. Several gold rows in one file, or a directory a single ls enumerates, reward one move and dilute the discriminator.
  4. Check the move budget. Count the baseline's tool calls against its wall. An arm that stopped comfortably inside the budget was not defeated by cost; an arm that ran out was.
  5. Name the next draft's candidate rows. The items the baseline missed, plus anything structurally adjacent to them that the table shows it never touched.

Read the full file on GitHub · 88 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. 4d ago First seen · 88 lines · 56 tokens per session scan A 73f6bfba8c0b

Subscribe to this mod's changes

bench-struggle-read is an agent published in the GitHub repository luuuc/sense (35 stars, last pushed 4d ago), licensed MIT. It adds 56 tokens to every session and 1,175 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 2026-08-30.

Related

Other agents, from other repositories

scala-coder

Implements a single coding task end-to-end in an assigned git worktree for this Scala project. Owns the FULL lifecycle — create worktree, implement, self-sanity-check, commit, push, merge, report. Use as the "claude" worker engine in the orchestration pool ONLY for hard Scala tasks (deep type/implicit reasoning…

MercurieVV/ScalaSemantic · 96 tokens

task-plan-architect

Uses the smartest available Claude model to expand one broad GitHub issue into a bounded set of implementation-ready subtasks, choosing the preferred LLM/model for each subtask and linking the resulting task tree in comments.

MercurieVV/ScalaSemantic · 47 tokens

task-tree-triage

Cheap per-issue classifier for recursive task planning. Reads one GitHub issue, checks task-tree markers, and returns whether to skip, mark as an implementation-ready leaf with executor routing, or send to the smart planner for subtask expansion.

MercurieVV/ScalaSemantic · 54 tokens

triage

Cheap sequential classifier. Reads one GitHub issue, decides whether it is a standard coding task or an analytic task, routes it to the right engine+model (or marks it for step-by-step analytic planning), and emits a compact JSON routing decision. Use before dispatching work to the parallel pool.

MercurieVV/ScalaSemantic · 63 tokens

task-prioritizer

Fetches all open GitHub issues/tasks for this repo, prioritizes them by project need and dependency order, and comments priority/dependency notes back onto each task. Use before detailed task planning.

MercurieVV/ScalaSemantic · 45 tokens

sanity-check

Cheap self-check run BY a task agent on its own worktree before committing. Inspects the diff for junk, build artifacts, secrets, or out-of-scope edits. NOT called by the conductor — the task agent calls this on itself. Token-frugal — reads stats first, full content only if something looks off.

MercurieVV/ScalaSemantic · 69 tokens