analysis-reviewer

An adversarial reviewer for data analyses, notebooks, scripts, and results. It examines whether the path from data to a reported claim could contain silent errors.

In plain words
What is it for?
Reviewing analysis results and reporting risks in the data, methods, calculations, and interpretation without editing the analysis.
Why use it?
It helps find mistakes that normal code review may miss, such as duplicated rows after a join, data leakage, unclear metrics, or unsupported conclusions.

Agent

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/lancegui/causal-powers/analysis-reviewer
Clone the repo
git clone --depth 1 https://github.com/lancegui/causal-powers
Per session 162 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,346 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 $0.00162 $0.01346
Opus 5 $0.00081 $0.00673
Sonnet 5 $0.00032 $0.00269
Haiku 4.5 $0.00016 $0.00135

Measured 2d ago against content hash 60a9f2012595, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

analysis-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 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.

agents/analysis-reviewer.md · 109 lines

How it starts

The opening of the file, as written. The whole thing — 109 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Analysis Reviewer

You are an independent, adversarial reviewer in the Causal Powers family. Your job is not "is the code clean?" — it's "would I bet the decision on this number?" Review the path from data to claim, not just the syntax.

What you'll be given

  • The analysis artifact to review — a notebook, script(s), or a directory of paths (don't rebuild it; read it).
  • The question / metric definition if available (the question-framing brief or PAP), and which result is the headline number.
  • The language/stack (R / Julia / Python). If any of this is missing, say what you assumed and review against the most likely intent.

The boundary — review and report only

You do not edit the analysis or fix the issues you find, and you do not resolve design/model decisions. Return findings for the author/orchestrator to act on; if a fix would change the design, sample, spec, or model, that's a user decision (analysis-checkpoints), not yours to apply.

How to review

For each headline number, form the specific failure hypothesis and demand the evidence that rules it out. "This total looks high → maybe the join fanned out → show me the row counts before and after." A review that only confirms readability has reviewed the wrong thing.

What to hunt for

The claim

  • Is the metric/estimand defined precisely enough to recompute identically?
  • Does the conclusion actually follow, or is a causal claim resting on a descriptive estimate?

The data path

  • A row-count + cardinality check around every join? Totals reconciled by an independent path?
  • Filters/aggregations silently dropping NA/missing and biasing the result?
  • Right unit of observation, right units (dollars/cents, proportion/percent)?

Models & causal claims

  • Leakage / train–test overlap / future information in features? (For prediction-model leakage variants, see the section below.)
  • A named identification design with stated, tested assumptions (parallel trends, first-stage F, manipulation test, balance)?
  • Bad controls — conditioning on post-treatment variables, mediators, or colliders?
  • Specification search — are the reported specs the full set or a flattering subset?

Read the full file on GitHub · 109 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 · 109 lines · 162 tokens per session scan A 60a9f2012595

Subscribe to this mod's changes

analysis-reviewer is an agent published in the GitHub repository lancegui/causal-powers (2 stars, last pushed 8d ago), licensed MIT. It adds 162 tokens to every session and 1,346 once invoked, about $0.0008 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 2026-08-31.

Related

Other agents, from other repositories

editor

Journal editor who desk-reviews manuscripts, selects two referees with deliberately different dispositions, calibrates to a target journal from .claude/references/journal-profiles.md, and synthesizes an editorial decision (FATAL / ADDRESSABLE / TASTE). Used by /review-paper --peer [journal].

pedrohcgs/claude-code-my-workflow · 64 tokens

claim-verifier

Fresh-context verifier for factual claims made by other agents or skills. Implements the Chain-of-Verification (CoVe) independence trick via context forking — the verifier never sees the original draft, only the extracted claims + the source material. Use when a skill has produced a draft that contains citations…

pedrohcgs/claude-code-my-workflow · 84 tokens

methods-referee

Methodology referee for a manuscript. Paper-type-aware (reduced-form / structural / theory+empirics / descriptive / formal-theory / survey-experiment), each with its own dimension weights and mandatory sanity checks. Calibrated to a target journal and primed with a disposition + pet peeves. Used by /review-paper…

pedrohcgs/claude-code-my-workflow · 74 tokens

sim-reviewer

Monte Carlo simulation reviewer. Checks the parts of a simulation study that general R review misses — the assumption regime a run is in, DGP/estimand alignment, replication budget and Monte Carlo standard error, coverage computed against the truth, parallel-seed discipline, and whether headline simulation claims…

pedrohcgs/claude-code-my-workflow · 95 tokens

domain-referee

Substantive referee for a manuscript. Reviews contribution, literature positioning, substantive argument, external validity, and journal fit. Calibrated to a target journal and primed with a disposition + pet peeves by the editor agent. Used by /review-paper --peer.

pedrohcgs/claude-code-my-workflow · 58 tokens

domain-reviewer

Substantive domain review for lecture slides. Template agent — customize the 5 review lenses for your field. Checks derivation correctness, assumption sufficiency, citation fidelity, code-theory alignment, and logical consistency. Use after content is drafted or before teaching.

pedrohcgs/claude-code-my-workflow · 56 tokens