bakeoff

bakeoff is a command for coding agents from MartyBonacci/specswarm. It costs 58 tokens per session (1,375 once invoked), scanned A, original, MIT.

A comparison workflow for choosing between several candidate outputs or implementations. It renders the candidates together, records a human verdict, and saves the chosen result as a regression rule.

In plain words
What is it for?
Use it when a feature has several plausible variants and quality depends on human judgement. It is attached to a SpecSwarm feature and helps select and preserve the preferred version.
Why use it?
It turns subjective choices about design, wording, colours, layouts, or ranking into documented decisions that can be checked later.

Command

Part of the ss plugin — 10 skills, 34 commands, 6 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 commands/martybonacci/specswarm/bakeoff
Clone the repo
git clone --depth 1 https://github.com/MartyBonacci/specswarm

Or install ss, the plugin that ships this one along with the rest of its 10 skills, 34 commands, 6 agents, 4 hooks.

Wrote 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.

agentmods badge for bakeoff

README.md
[![agentmods](https://agentmods.dev/badge/commands/martybonacci/specswarm/bakeoff.svg)](https://agentmods.dev/commands/martybonacci/specswarm/bakeoff)
Your own site
<a href="https://agentmods.dev/commands/martybonacci/specswarm/bakeoff"><img src="https://agentmods.dev/badge/commands/martybonacci/specswarm/bakeoff.svg" alt="Measured on agentmods" height="20"></a>
Per session 58 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,375 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 $0.00058 $0.01375
Opus 5 $0.00029 $0.00687
Sonnet 5 $0.00012 $0.00275
Haiku 4.5 $0.00006 $0.00137

Measured 3d ago against content hash d7f05722013e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

bakeoff 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.

plugins/ss/commands/bakeoff.md · 103 lines

How it starts

The opening of the file, as written. The whole thing — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.

SpecSwarm Bakeoff

Origin lesson: run by hand ~15 times in production, this loop (candidates → contact sheet → human verdict → distilled ruling → pinned winner) was the single most effective pattern for converting fuzzy judgment into deterministic constants. Generic across domains: color pickers, copy tone, layout variants, ranking weights.

Phase 1: Resolve context + workspace

REPO_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
PLUGIN_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
# shellcheck disable=SC1091
source "${PLUGIN_DIR}/lib/features-location.sh"
# shellcheck disable=SC1091
source "${PLUGIN_DIR}/lib/bakeoff.sh"

# Resolve feature dir (branch NNN-slug → --feature override → latest feature)
BRANCH=$(git rev-parse --abbrev-ref HEAD 2>/dev/null)
FEATURE_NUM="${FEATURE_OVERRIDE:-$(echo "$BRANCH" | grep -oE '^[0-9]{3}' || true)}"
[ -z "$FEATURE_NUM" ] && FEATURE_NUM=$(list_features "$REPO_ROOT" | grep -oE '^[0-9]{3}' | sort -nr | head -1)
find_feature_dir "$FEATURE_NUM" "$REPO_ROOT" || { echo "❌ no feature dir; run /ss:specify first (bakeoffs attach to a feature)"; exit 1; }

SLUG=$(ss_bakeoff_slug "<the user's description>")
BAKEOFF_DIR=$(ss_bakeoff_dir "$FEATURE_DIR" "$SLUG")
echo "🍞 Bakeoff workspace: ${BAKEOFF_DIR}"

N = --candidates (default 3, clamp to 2–6).

Phase 2: Generate N candidates (parallel, deliberately diverse)

Claude — dispatch N parallel Task subagents (subagent_type: general-purpose) in ONE message. Each gets the same brief plus a DISTINCT angle so the candidates genuinely differ (not one idea ×N):

  • Angle rotation (pick N): conservative/on-convention · bold/high-contrast · minimal · dense/information-rich · playful · borrowed-from-best-in-class.
  • Each agent's brief:
    You are candidate <i> of <N> in a SpecSwarm bakeoff: <description>.
    Your angle: <angle>. Read the relevant project context (spec.md, existing code)
    but DO NOT modify project files. Write your candidate into
    <BAKEOFF_DIR>/candidate-<i>/ :
      - NOTES.md — 3-6 lines: your angle, the key choices, the constants you chose
      - the candidate artifact(s): code file(s), copy text, config/constants file —
        whatever form the choice takes. Self-contained; runnable/viewable in isolation
        where possible.
    Return one line: CANDIDATE <i> READY <one-line self-description>.
    
  • Visual choices + browser MCP available: after the agents finish, render each candidate (dev server or static open) and save a screenshot as candidate-<i>/preview.png (read the browser console while you're there — console errors disqualify a candidate).

Read the full file on GitHub · 103 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. 3d ago First seen · 103 lines · 58 tokens per session scan A d7f05722013e

Subscribe to this mod's changes

bakeoff is a command published in the GitHub repository MartyBonacci/specswarm (65 stars, last pushed 1mo ago), licensed MIT. It adds 58 tokens to every session and 1,375 once invoked, about $0.0003 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.