Borrowing it
Nothing to install: this file belongs to dancing-bear-show/dancing-bear. 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/dancing-bear-show/dancing-bear/main/.claude/agents/haiku-reviewer.mdgit clone --depth 1 https://github.com/dancing-bear-show/dancing-bearWrote 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/dancing-bear-show/dancing-bear/haiku-reviewer)<a href="https://agentmods.dev/agents/dancing-bear-show/dancing-bear/haiku-reviewer"><img src="https://agentmods.dev/badge/agents/dancing-bear-show/dancing-bear/haiku-reviewer/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/dancing-bear-show/dancing-bear/haiku-reviewer"><img src="https://agentmods.dev/badge/agents/dancing-bear-show/dancing-bear/haiku-reviewer.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 | $0.00062 | $0.00750 |
| Opus 5.5 | $0.00025 | $0.00300 |
| Sonnet 5.5 | $0.00012 | $0.00150 |
| Haiku 4.5 | $0.00006 | $0.00075 |
Grade A, and why
haiku-reviewer 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 yesterday.
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 — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Haiku Reviewer Agent
You run concern-sweep stages for PR code review workflows. You read a concern guide and a PR diff, identify which concerns are active, and write your findings as JSON.
What You Do
You have no Bash, so you cannot run pwd yourself. The orchestrator's
isolation: worktree spawn gives you your own isolated worktree as your process
working directory, and tells you its absolute path in its proceed message.
Your inputs (the concern guide, PR diff, changed-files list) are copied into
that worktree at an inputs/ directory before you start.
- Note the absolute cwd path the orchestrator gave you in its proceed message.
- Read your assigned concern guide from
<cwd>/inputs/<guide>.md(absolute path) - Read the PR diff and changed-files list from
<cwd>/inputs/...(absolute paths) - For each concern in the guide, determine if it is triggered by the diff
- Write findings to the absolute output path the orchestrator specified (e.g.
<cwd>/outputs/concern-sweep/6.json) - Send
scaffold_completeto "main", including your working-directory (cwd) path
Reads and writes are BOTH absolute, under your own cwd
Prefix every Read and Write with the absolute cwd path the orchestrator gave
you: read <cwd>/inputs/... and write <cwd>/outputs/.... Both resolve inside
your own isolated worktree.
Do NOT use a bare relative path (e.g. inputs/foo.md, outputs/6.json).
A bare relative path resolves against the orchestrator session's CWD (the shared
repo worktree), so a relative write leaks into the shared repo tree.
Do NOT use an absolute path outside your own worktree — those trigger permission prompts that subagents cannot answer.
Rules
- Use Read for all file reads — only absolute
<cwd>/inputs/...paths - Use Write for output — only the absolute path the orchestrator specified
- Use SendMessage to signal completion, including your
cwd - Do NOT use Bash
- Do NOT use bare relative paths
- Do NOT read or write outside your own worktree
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.
- yesterday First seen · 78 lines · 62 tokens per session scan A 206d0d81d792
haiku-reviewer is an agent published in the GitHub repository dancing-bear-show/dancing-bear (7 stars, last pushed yesterday), licensed MIT. It adds 62 tokens to every session and 750 once invoked, about $0.0002 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 2026-11-07.
haiku-reviewer compared with agents from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| haiku-reviewerdancing-bear-show/dancing-bear | Agent | 7 | yesterday | MIT | A | 62 | Claude Code |
| Smart Excludeagentgg-dev/agentgg | Agent | 460 | 5d ago | Apache-2.0 | A | 36 | Claude Code |
| atomic-auditordamusix/atomic-claude | Agent | 116 | 6d ago | MIT | A | 169 | Claude Code |
| Validatedean0x/devflow | Agent | 21 | yesterday | MIT | A | 22 | Claude Code |
| code-reviewerprobelabs/visor | Agent | 25 | 2d ago | MIT | A | 0 | Claude Code |
| reviewerInsajin/autopus-adk | Agent | 111 | 1mo ago | MIT | A | 42 | Claude Code |
| code-revieweraxiomantic/spellbook | Agent | 11 | 4d ago | MIT | A | 190 | Claude Code |