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.
npx agentmods add skills/jmosier69/refute/deep-reviewnpx skills add Jmosier69/refute --skill deep-reviewgit clone --depth 1 https://github.com/Jmosier69/refuteWrote 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/skills/jmosier69/refute/deep-review)<a href="https://agentmods.dev/skills/jmosier69/refute/deep-review"><img src="https://agentmods.dev/badge/skills/jmosier69/refute/deep-review.svg" alt="Measured on agentmods" 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 | $0.00097 | $0.06143 |
| Opus 5 | $0.00048 | $0.03071 |
| Sonnet 5 | $0.00019 | $0.01229 |
| Haiku 4.5 | $0.00010 | $0.00614 |
Grade A, and why
deep-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 5d 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 — 461 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Deep Review
One review command, four modes. Absorbs the strongest parts of six prior review skills; see
DESIGN.md for what came from where and why.
- Never spawn Fable. Not at any tier, not under any flag. When a finding warrants a stronger independent read, STOP and print the escalation block (§10). The operator invokes Fable themselves, in a fresh context.
- Never block a merge. This command is advisory. It may emit a verdict; it may not act on one.
- Never auto-fix without re-reviewing the fix. A fix for a review finding is not pre-vetted.
- Agents are read-only. Parallel agents share one worktree — a writing agent corrupts the others' view.
1. Resolve mode and base ref
| Invocation | Mode | Base |
|---|---|---|
/deep-review |
A — increment | git merge-base origin/main HEAD, includes uncommitted |
/deep-review --pr [N] |
B — pre-prod gate | origin/main (three-dot) |
/deep-review --sweep [--since X] |
C — integration sweep | union diff of merges since marker |
/deep-review --security |
D — security-only | any of the above scopes |
Always resolve origin/main, never local main — a stale local main silently corrupts the
diff base. Run git fetch origin first. For Mode B, cross-check the computed base against
gh pr view <N> --json baseRefName,baseRefOid; if they disagree, trust gh and say so.
For Mode C, read the marker from <REVIEWS>/state.json (see §11). If absent, default --since
to the last merge commit on origin/main and say which ref you chose.
Amplification (§8) runs by default in Mode B only. --amplify enables it in Modes A, C, and D.
--no-amplify disables it in every mode, including the Mode A --fix path. Absent
--no-amplify, --fix in Mode A amplifies the findings it fixed (§12) without needing
--amplify.
Fail fast, before spawning anything: confirm the ref resolves (git rev-parse), confirm the
diff is non-empty, confirm the working tree state matches the mode. A bad ref must fail here, not
inside fourteen sub-agents.
What ships with it
5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 5d ago First seen · 461 lines · 97 tokens per session scan A 52e52a64c68e
deep-review is a skill published in the GitHub repository Jmosier69/refute (9 stars, last pushed 26d ago), licensed MIT. It adds 97 tokens to every session and 6,143 once invoked, about $0.0005 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 skills, from other repositories
river-review-code
コードの可読性、保守性、型安全性を検証する。他の専門エージェントに該当しない場合のデフォルトフォールバック先。.
Standard Review Policy for Midstream
Applies standard AI review policy guidelines for midstream (implementation) phase reviews.
codex-adversarial-review-lite
Codex Adversarial Review - Lite: user-invoked audit workflow for Claude Code users who want Codex CLI to independently review AI-generated code, plans, test expectations, and scope before fixes are applied. Cross-platform (Windows, macOS, Linux, WSL). Use only when the user explicitly invokes audit or selftest.
workflow-audit
Systematic UI workflow auditing for SwiftUI applications. Discovers entry points, traces user flows, detects dead ends and broken promises, audits data wiring, evaluates from user perspective. Triggers: "workflow audit", "audit flows", "find dead ends", "check navigation".
plan
Epic decomposition into trackable, right-sized tasks. Three modes — audit-aware (codebase-audit reports), workflow-audit-aware (handoff.yaml with pre-rated findings), standalone (from scratch). Light convention scanning for projects without CLAUDE.md.
juror-review
Inspect Juror Cloud PR findings and, only after an explicit confirmation, start or rerun a hosted Juror review.