coder-eval-review

A command that creates a JSON review for each failed task in a completed Coder Eval run, including a summary and issue labels.

In plain words
What is it for?
Use it to review a chosen or latest evaluation run, classify failures with tags, and create review index files.
Why use it?
It turns failed evaluation results into structured records that can be compared, searched, and aggregated.

Command 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 commands/uipath/coder_eval/coder-eval-review
Clone the repo
git clone --depth 1 https://github.com/UiPath/coder_eval

Made for: Claude Code.

Per session 14 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,595 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 $0.00014 $0.01595
Opus 5 $0.00007 $0.00797
Sonnet 5 $0.00003 $0.00319
Haiku 4.5 $0.00001 $0.00160

Measured yesterday against content hash 1ff86c493206, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

coder-eval-review 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.

.claude/commands/coder-eval-review.md · 104 lines

How it starts

The opening of the file, as written. The whole thing — 104 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Context

If $ARGUMENTS is empty or blank, resolve the run to review per the "No target given" rule in .claude/shared/run-layout.md — discover the run root, then take latest if that symlink resolves and the newest run directory by name if it does not. Tell the user which run you picked and how you got there before reading anything.

You are producing post-run reviews for a coder_eval evaluation run. The target path is: $ARGUMENTS

For each failed task in the run you write a structured review.json next to that task's task.json, plus a small review_index.json digest at the run root for fast aggregation. Output is JSON only — no narrative.

Step 1: Load the suggested vocabulary

Read src/coder_eval/resources/tags.yaml (relative to the repo root). It has a top-level tags: list of {name, definition, examples} entries. Prefer these names when classifying — but the vocabulary is a suggestion, not a strict allowlist. If nothing fits, emit a kebab-case slug that does. Drift is caught by a lint, not blocked at write time.

If the vocabulary file is missing or unparseable, continue with an empty suggestion list (don't abort) — drift lint will surface every tag you emit.

Step 2: Discover failed tasks

The run layout (runs/<run_id>/<variant_id>/<task_id>/<NN>/…, <NN> a zero-padded replicate index) is defined in .claude/shared/run-layout.md.

  1. Read <run_path>/run.json if present (for context — run_id, start_time).
  2. Glob <run_path>/*/*/*/task.json and read each one.
  3. Read <run_path>/analysis.md if present — it already diagnoses many failures; lean on its findings rather than re-deriving them.
  4. A task counts as failed if final_status != "SUCCESS" or weighted_score < 0.9. Skip passing tasks for now (we may extend to passing tasks later — the schema supports it).

If no task.json files exist, write an empty review_index.json ({"reviews": []}) and exit.

Step 3: Write per-task review.json

For each failed task, write <run_path>/<variant_id>/<task_id>/<NN>/review.json with this exact shape:

Read the full file on GitHub · 104 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. yesterday First seen · 104 lines · 14 tokens per session scan A 1ff86c493206

Subscribe to this mod's changes

coder-eval-review is a command published in the GitHub repository UiPath/coder_eval (119 stars, last pushed 3d ago), licensed Apache-2.0. It adds 14 tokens to every session and 1,595 once invoked, about $0.0001 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.