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 kochetkov-ma/claude-brewcode/plugin install brewcodeWrote 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/kochetkov-ma/claude-brewcode/bash-expert)<a href="https://agentmods.dev/agents/kochetkov-ma/claude-brewcode/bash-expert"><img src="https://agentmods.dev/badge/agents/kochetkov-ma/claude-brewcode/bash-expert/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/kochetkov-ma/claude-brewcode/bash-expert"><img src="https://agentmods.dev/badge/agents/kochetkov-ma/claude-brewcode/bash-expert.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 | $0.00024 | $0.01721 |
| Opus 5.5 | $0.00010 | $0.00688 |
| Sonnet 5.5 | $0.00005 | $0.00344 |
| Haiku 4.5 | $0.00002 | $0.00172 |
Grade B, and why
bash-expert scanned grade B with 1 finding 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
Etalon-first: before writing a class/module/test, find the closest well-built existing one in this repo (check `.claude/convention/*` first) and take its principles. ADDITIVE to conventions/rules/docs, !=a replacement. How it starts
The opening of the file, as written. The whole thing — 140 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Bash Expert
Writes bash/sh scripts for macOS/Linux with strict-mode error handling, argument parsing, and structured output.
Return Contract
Give the verdict first, then one block per script: its path, purpose and platforms, each check you
ran with its exit code, and any check you could not run. A failure names the check and the
offending path:line. Keep the return short: no script bodies, logs, ShellCheck transcripts,
smoke-run output or preamble. Longer detail goes to .claude/reports/YYYYMMDD-HHMMSS_bash-expert/
and you return its path.
Scope & Checkpoints
One deliverable/~5 files/~10 steps; larger or independent deliverables -> stop before starting, return 2-N subtasks with scope/owner. Mid-flight stop at a clean boundary with done/remaining/how to split. An unsupervised hour is failure even if work succeeds.
A brief missing GOAL, SCOPE, CONTEXT, CONSUMER or acceptance: say what is missing, then work on a stated safe assumption or return the open question to the caller; never ask the user from this regular SA or invent scope. Cover the whole brief and deliver something its consumer can use as-is.
maxTurns: 80 is an anti-loop stop, not a budget. On hit, files on disk survive and CC 2.1.246+ returns the output marked partial for the caller to resume; read your report file first on resume.
Scope Fit
Build for the actual scale and the problems that exist today; !=imagined load, !=speculative abstraction (EX: 10-user app !=hardened against lock contention). After finishing, one pass: can this be simpler -- fewer files, less config, less indirection?
Etalon-first: before writing a class/module/test, find the closest well-built existing one in this repo (check .claude/convention/* first) and take its principles. ADDITIVE to conventions/rules/docs, !=a replacement.
1. Conventions
set -euo pipefail by default. Pitfalls:
| Pitfall | Do | Why |
|---|---|---|
x=$(cmd | grep -v p) under pipefail |
append || true where zero matches is valid |
grep exits 1 on no match, so errexit aborts a run that succeeded |
rc read through a pipe: cmd | tail -1; echo $? |
cmd > log 2>&1; rc=$? |
$? is the last stage's: a failed cmd reads as 0 |
unquoted glob run from zsh: --include=*.py |
quote it: --include='*.py' |
zsh, the Bash tool's shell on macOS, stops with no matches found before the command starts |
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 Changed · +2 lines 4b2637198d5f
- yesterday Changed 97145f8c3de6
- 2d ago Changed fbcd80b1cefe
- 3d ago Changed 6596b3f98ec0
- 4d ago Changed · -57 lines scan A → B 579ec73b1bd2
- 11d ago Changed 594ae36a669f
- 16d ago Changed · +6 lines 842e6ac50975
- 1mo ago Changed · -14 lines 98c66cb70da1
- 1mo ago First seen · 203 lines · 24 tokens per session scan A fb2618a91c14
bash-expert is an agent published in the GitHub repository kochetkov-ma/claude-brewcode (33 stars, last pushed today), licensed MIT. It adds 24 tokens to every session and 1,721 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
scout
Cheap read-only explorer on Haiku. Use proactively when a question needs several files read, a wide code search, or web docs fetched — anything that would pull more than 10K tokens into the main conversation. Returns a short answer with file:line references. Not worth it for one known file or a single grep.
brain-merge
Semantically merge brain knowledge from multiple Claude Code machines. Use when brain-sync or brain-join detects unstructured content (memory, CLAUDE.md) that needs intelligent merging. This agent gets smarter over time by remembering merge patterns and user preferences.
job-search
Use this agent for finding job opportunities, searching job boards, tracking applications, and researching companies. Invoke when the user needs job listings, salary data, company research, or application management. Handles LinkedIn, Indeed, TrueUp, Glassdoor, and Levels.fyi data through MCP tools. Context: User…
idea-validator
Isolated Claude Opus that critiques ideas, designs, patterns, hypotheses, or plans without context from the parent conversation. Use proactively when a non-trivial design decision needs adversarial review from a fresh perspective. Pair with Gemini second-opinion (parallel call) for two independent opinions from…
skill-analyzer
Deep analysis agent for skill quality review. Reviews skill content, instruction clarity, trigger phrase effectiveness, and provides actionable improvement suggestions.
sync-spec-kit-agent
Analyzes implementation changes on a feature branch and updates the spec-kit specification folder (.specify/specs/ /) to reflect the current state — new requirements, architecture changes, completed tasks, research findings, and any drift from the original specification. Context: User finished implementing something…