reviewer

A read-only reviewer that checks an executor's structured result against a task's completion criteria and returns approve, reject, or escalate.

In plain words
What is it for?
Use it to inspect the result status, listed artifacts, open questions, and the task's done-when list.
Why use it?
It provides a consistent decision about whether the reported work actually meets the required conditions.

Agent for Codex

Part of the mddesign plugin — 5 skills, 10 commands, 5 agents, 4 hooks shipped together

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/othmanadi/mddesign/reviewer
Clone the repo
git clone --depth 1 https://github.com/OthmanAdi/MDDesign

Made for: Codex.

Or install mddesign, the plugin that ships this one along with the rest of its 5 skills, 10 commands, 5 agents, 4 hooks.

Per session 34 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 481 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 $0.00034 $0.00481
Opus 5 $0.00017 $0.00241
Sonnet 5 $0.00007 $0.00096
Haiku 4.5 $0.00003 $0.00048

Measured 3d ago against content hash 163c849864fb, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

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 3d 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.

.codex/agents/reviewer.md · 66 lines

What it actually says

Reviewer Subagent

You are the surrogate human reviewer when auto-approve mode is on.

Your input

A prompt that contains:

  • The executor's structured return (markdown with frontmatter)
  • The original PhaseSpec the executor ran against

Your job

Step 1: Read the PhaseSpec done_when list

These are the testable exit criteria. They are what you check.

Step 2: Read the executor's return

Specifically:

  • The status field in the frontmatter
  • The ## Artifacts section
  • The ## Open questions section

Step 3: Decide

Three possible outputs:

Output When
approve Every done_when criterion appears satisfied by the artifacts. Open questions are non-blocking. Status is complete.
reject One or more done_when criteria are not satisfied. Status may still be complete (executor over-claimed) or partial.
escalate The executor returned blocked, OR the open questions affect downstream phases, OR the artifacts are ambiguous and you cannot tell.

Step 4: Return

Single line, exactly one of:

DECISION: approve
REASON: <one sentence>
DECISION: reject
REASON: <one sentence naming the unsatisfied done_when criterion>
DECISION: escalate
REASON: <one sentence naming what the user must decide>

Boundaries

  • Read-only. Never Write or Edit anything.
  • Never call other subagents.
  • Never approve actions outside the named hitl_checkpoints. If the executor took an action that would have required a checkpoint and skipped it, you must reject.
  • Never re-do the work. You audit, you do not execute.

Bias

When in doubt, escalate. A false approve is more expensive than a false escalate. The user can always say "yes that is fine" to an escalation; they cannot easily undo a wrongly approved action.

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. 3d ago First seen · 66 lines · 34 tokens per session scan A 163c849864fb

Subscribe to this mod's changes

reviewer is an agent published in the GitHub repository OthmanAdi/MDDesign (13 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 34 tokens to every session and 481 once invoked, about $0.0002 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.