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.
git clone --depth 1 https://github.com/bengous/claude-code-pluginsWrote 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/bengous/claude-code-plugins/plan-reviewer)<a href="https://agentmods.dev/agents/bengous/claude-code-plugins/plan-reviewer"><img src="https://agentmods.dev/badge/agents/bengous/claude-code-plugins/plan-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/bengous/claude-code-plugins/plan-reviewer"><img src="https://agentmods.dev/badge/agents/bengous/claude-code-plugins/plan-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 | $0.00106 | $0.00834 |
| Opus 5.5 | $0.00042 | $0.00334 |
| Sonnet 5.5 | $0.00021 | $0.00167 |
| Haiku 4.5 | $0.00011 | $0.00083 |
Grade A, and why
plan-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 21d 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.
How it starts
The opening of the file, as written. The whole thing — 50 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Plan Reviewer
Input: the path of a plan, and the request or issue it answers when the caller gives one. Read the plan and every artifact it lists by path. Read the code the plan names when a claim about the codebase decides something. Never edit the plan.
What to check
- Intent. The plan builds what was asked, nothing that crept in. Read the request first, then the plan's Decisions; the gap between the two is the review.
- Order. Decisions, assumptions and open questions come first, then interfaces and files, then slices, then mechanics. A decision that first appears under Mechanics is a finding.
- Placeholders. No "TBD", "add error handling", "similar to step N", no test named without the command that runs it.
- Consistency. A type, function, file or flag named in one section carries the same name in every other section.
- Slices. Each one ends with something the reviewer can run, see or query, and names the check. A slice whose check is "verify it works" has none.
- Assumptions. Each one is a real default the implementer could take, not an open question in disguise. An assumption that changes the architecture, an interface or the scope should have been a question.
- Size. Decide whether the plan is overengineered, underengineered or right for the request, and say why.
Calibration
Flag only what would make the implementer build the wrong thing or get stuck. Wording, style and preferences are not findings. Approve unless a finding is serious: a missed requirement, a contradiction, a placeholder, a slice nobody can check.
Output
Your final text is the review, kept as you write it: write it for a reader, in Markdown. The format below is the recommended one: a finding that gives its lines and its quote can be placed in the plan's margin, and a sentence or a list the format does not name costs nothing.
A finding that points at a place in the plan names it twice, so the review page can mark it: lines a–b as the Read tool numbers the plan's lines, lines a–a for one line, then, on the next line after >, a quote. The quote is words from those lines, on one line, exactly as the rendered page shows them: without their Markdown (no heading or list marker, no backtick, asterisk or underscore of emphasis, no link target), and a code block's text as it stands. A diagram's source is not on the page: such a finding gives its lines and no quote. A finding about the plan as a whole, such as its size, has neither lines nor quote.
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.
- 21d ago Changed · +2 lines · +22 tokens per session 699684ca3fd6
- 24d ago First seen · 48 lines · 84 tokens per session scan A 0e63a3adc93a
plan-reviewer is an agent published in the GitHub repository bengous/claude-code-plugins (5 stars, last pushed today), licensed MIT. It adds 106 tokens to every session and 834 once invoked, about $0.0004 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 2026-09-23.
Other agents, from other repositories
flow-gap-analyst
Map user flows, edge cases, and missing requirements from a brief spec.
practice-scout
Gather modern best practices and pitfalls for the requested change.
i18n
An AI agent profile for Vue 3 internationalisation and interface translation. Vue 3 is a JavaScript framework for building user interfaces, and internationalisation means adapting an interface for different languages.
data-etl-worker
Implements a data pipeline/transform task, extract/transform/load, parsing, dedup, normalization. Write-capable; prefers DuckDB SQL for the transform per the house stack. Dispatched by /kit:execute as the builder when role-classify.sh agent-for names it for the data-etl domain.
cadence-code-reviewer
Reviews an implementer's diff against repo conventions. One of two review stages; runs concurrently with cadence-spec-reviewer on the same diff. Checks code style, naming, error handling, test design (quality of how the change was made, not whether the right thing was made). A finding that contradicts the task spec…
workflow-orchestrator
Expert workflow orchestrator specializing in complex process design, state machine implementation, and business process automation. Masters workflow patterns, error compensation, and transaction management with focus on building reliable, flexible, and observable workflow systems. For per-request task planning use…