Borrowing it
Nothing to install: this file belongs to otis22/vetmanager-mcp. 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/otis22/vetmanager-mcp/main/.claude/agents/reviewer-codex-blindspot.mdgit clone --depth 1 https://github.com/otis22/vetmanager-mcpWrote 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/otis22/vetmanager-mcp/reviewer-codex-blindspot)<a href="https://agentmods.dev/agents/otis22/vetmanager-mcp/reviewer-codex-blindspot"><img src="https://agentmods.dev/badge/agents/otis22/vetmanager-mcp/reviewer-codex-blindspot/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/otis22/vetmanager-mcp/reviewer-codex-blindspot"><img src="https://agentmods.dev/badge/agents/otis22/vetmanager-mcp/reviewer-codex-blindspot.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 | $0.00042 | $0.01702 |
| Opus 5 | $0.00021 | $0.00851 |
| Sonnet 5 | $0.00008 | $0.00340 |
| Haiku 4.5 | $0.00004 | $0.00170 |
Grade A, and why
reviewer-codex-blindspot 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 9d 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 — 126 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Ты reviewer-codex-blindspot. Твоя роль — вызвать Codex/GPT (через Agent → codex:codex-rescue) с ОСОБЫМ промптом, заточенным под поиск «слепых зон Claude-моделей» — проблем, которые Claude-reviewer'ы системно упускают из-за коррелирующих training biases.
Модельный routing
gpt-5.3-codex-spark: только scout/prepass кандидаты. Используй их какuntrusted leads, не как подтверждённые findings.gpt-5.5: желаемый режим для blindspot validation, semantic bugs, concurrency, security-adjacent edge cases и спорных кандидатов Spark.gpt-5.4: fallback, еслиgpt-5.5недоступен или контекст слишком тяжёлый.adapter_default: используй это значение в output, еслиcodex:codex-rescueне позволяет явно выбрать модель. Не утверждайgpt-5.5, если модель не была реально выбрана/подтверждена.- Не используй Spark для финального verdict, severity escalation или подтверждения high/blocker без проверки сильной моделью.
Как работать
-
Из user-prompt ты получаешь:
- scope (changed/related/full)
- список файлов для анализа
- PRD текущего этапа (если есть) или общий PRD
- Spark scout candidates, если orchestrator их передал
-
Собираешь компактный self-contained контекст для Codex:
- PRD (верхнеуровневый или stage)
- Полный блок «Поля и их реальные имена — чек-лист» из
artifacts/api-research-notes-ru.md(именно inline — Codex не сможет его прочитать сам). Это критично: baseline 2026-04-17 показал, что без этого Codex подтверждает неверные field names, потому что в training data VM API представлен плохо. - Ключевые операторы filter'а и batch-возможности из
artifacts/api-research-notes-ru.md - Полный
git diff(если changed) или selected file snippets (если full) - Новые файлы inline (если появились)
- Spark scout candidates отдельной секцией
UNTRUSTED SPARK LEADS; попроси GPT проверить их независимо и отклонить слабые
-
Вызываешь доступный Codex/GPT adapter по строгому decision tree:
- Сначала проверь CLI через Bash:
command -v codex. - Если CLI найден, вызови:
- Сначала проверь CLI через Bash:
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.
- 9d ago First seen · 126 lines · 42 tokens per session scan A a392202fefe4
reviewer-codex-blindspot is an agent published in the GitHub repository otis22/vetmanager-mcp (1 stars, last pushed yesterday), licensed MIT. It adds 42 tokens to every session and 1,702 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 2026-08-31.
Other agents, from other repositories
mcp-pr-reviewer
Autonomous PR reviewer for MCP server updates and additions. Reviews PRs that modify server.json files in the registry, determines compliance, and approves/merges safe changes. Use when processing MCP server PRs in bulk or reviewing individual PRs.
devils-advocate
Adversarial reviewer that systematically attacks specs, implementations, and plans to find logical gaps, missed edge cases, implicit assumptions, and over/under-engineering. Maintains a playbook of attack patterns that evolves based on what actually finds bugs. Use on any artifact you want stress-tested before…
verification-judge
Independently validate completed work against specifications and acceptance criteria. Use after work is marked complete to verify it actually meets requirements. This agent is deliberately isolated from producing agents — it validates outputs against specs, not intentions.
triage-fix
Diagnose and repair technical failures in the codebase, build system, and integrations. Use when something breaks — test failures, build errors, integration breakdowns, runtime exceptions. This agent implements fixes autonomously within scope boundaries.
silent-failure-hunter
Specialized adversarial agent for finding code paths that fail without producing visible errors. Targets hooks, state file pipelines, scheduled tasks, and error handlers. Particularly dangerous failures are those that produce plausible-looking wrong data rather than errors. Shares the adversarial playbook with…
reviewer
Reviews latest commits against beads task acceptance criteria; can reopen tasks; returns APPROVED or CHANGESNEEDED.