scout

A command that sends a software review to several AI models from different providers, then has a second review challenge the first findings before combining the results.

In plain words
What is it for?
Use it to divide a review into areas such as security, correctness, architecture, product, or documentation, gather findings, and produce a verified summary for human decision-making.
Why use it?
It provides independent opinions and an adversarial check, helping expose defects or missed risks in a review.

Command

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/askwigconsulting/cohort/scout
Clone the repo
git clone --depth 1 https://github.com/askwigconsulting/cohort
Per session 36 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,005 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 $0.00036 $0.01005
Opus 5 $0.00018 $0.00502
Sonnet 5 $0.00007 $0.00201
Haiku 4.5 $0.00004 $0.00101

Measured 2d ago against content hash 30543619ac13, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

scout 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.

canonical/commands/scout.md · 79 lines

How it starts

The opening of the file, as written. The whole thing — 79 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Send a panel after it. /scout runs a multi-vendor adversarial review: a coordinator decomposes the target into disjoint areas and routes each to the cheapest capable model from any vendor, then has a second round try to break the first round's findings. Reviewers are advisory and read-only — they find; the coordinator verifies; the human decides. This is the review sibling of /crew.

Run /scout only from a coordinator tier (Fable, preferred, or Opus) — decomposition, routing, and adversarial synthesis are the judgments a lower tier gets wrong.

1. Scope — carve disjoint areas

Define the target and split it into areas that don't overlap, so reviewers don't collide and coverage is legible: e.g. security/trust, correctness, architecture, product, docs. Assign the hardest, most ambiguous areas to the strongest models and the mechanical ones to the cheapest — that is the token-optimization the panel exists for.

2. Round 1 — fan out across vendors (≤20 in flight)

Each reviewer gets its area, all three lenses (defects weighted highest), and the operational gates (scope, evidence, adversarial self-check, verify, calibrate). Route by fit and cost:

  • Claude subagents — Fable for architecture-critical/ambiguous, Opus for complex, Sonnet for well-scoped, Haiku for mechanical/doc areas. They read the repo directly.

  • ChatGPT/consult-gpt (Codex CLI, read-only). Explores the repo itself.

  • Grokcohort engine review grok --tier flagship for areas needing repo exploration, or cohort engine consult grok --tier flagship with a packaged bundle for a bounded question. When grok-cli + bwrap are installed these run through the local bubblewrap-sandboxed grok CLI (real, worktree-scoped repo access); if it is missing — or present but broken — they fall back to the API-direct agentic loop (gated tools, transcript recorded). Read-only, advisory; every path is egress-gated.

    The fallback is always announced on stderr, never silent. Read the note before the answer: the two channels differ in what the model could see (worktree-scoped file access versus the gated read-only toolbox), so an area reviewed over the fallback was reviewed with less context than one reviewed over the CLI. A gate refusal is different again — it is a refusal, never retried on the other channel.

Read the full file on GitHub · 79 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 · 79 lines · 36 tokens per session scan A 30543619ac13

Subscribe to this mod's changes

scout is a command published in the GitHub repository askwigconsulting/cohort (2 stars, last pushed 26d ago), licensed MIT. It adds 36 tokens to every session and 1,005 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.