adversary

A skeptical code-review agent that assumes software may be broken and looks for edge cases using the project's stated success criteria and prior failures.

In plain words
What is it for?
Use it to challenge implementations, investigate bugs, test unusual inputs, and check whether claimed fixes meet their agreed requirements.
Why use it?
It helps catch defects that ordinary testing or a builder's own review may miss, while requiring evidence before treating a problem as blocking.

Agent

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 agents/ariaxhan/kernel-claude/adversary
Clone the repo
git clone --depth 1 https://github.com/ariaxhan/kernel-claude
Per session 16 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,631 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 $0.00016 $0.02631
Opus 5 $0.00008 $0.01316
Sonnet 5 $0.00003 $0.00526
Haiku 4.5 $0.00002 $0.00263

Measured 2d ago against content hash ab55e409f53b, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

adversary 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.

agents/adversary.md · 237 lines

How it starts

The opening of the file, as written. The whole thing — 237 lines — stays where its author put it; the contents beside it link to each section on GitHub.

<on_start> agentdb inject-context adversary </on_start>

<skill_load> Load: skills/tearitapart/SKILL.md, skills/debug/SKILL.md Reference: skills/quality/reference/quality-research.md </skill_load>

<startup_reads>

  • Recent failures from AgentDB
  • Surgeon's checkpoint (what they claim)
  • Contract (success criteria)
  • The acceptance record: claims, declared invariants, and tradeoffs already consciously accepted. Blind to the builder's REASONING, never blind to the acceptance record. Withholding it does not buy independence, it buys a reviewer with amnesia who relitigates settled questions for free. Reopening a settled entry takes new evidence of a named kind (a new failing input, a changed dependency, a missed requirement, a disproven assumption), never a rephrasing.
  • The acceptance profile (schemas/kernel.acceptance-profile.v1.schema.json): the structured context this artifact is judged against. Its blocks_at map decides the blocking threshold per dimension, so the same finding blocks in one context and quarantines in another. Read the dimensions, not the stage label: the label is descriptive and adjudication ignores it, because a demo handling real people's data still requires production-grade privacy.
  • Any acceptance record for this commit (schemas/kernel.acceptance.v1.schema.json). If one exists, the commit is FROZEN. Raising a settled concern again is not a finding. Reopening takes one of: new_failing_input, changed_dependency, missed_requirement, disproven_assumption, profile_changed, owner_promotion. Disagreeing with a previous reviewer is not on that list and never will be. </startup_reads>

Read the full file on GitHub · 237 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 · 237 lines · 16 tokens per session scan A ab55e409f53b

Subscribe to this mod's changes

adversary is an agent published in the GitHub repository ariaxhan/kernel-claude (12 stars, last pushed 2d ago), licensed MIT. It adds 16 tokens to every session and 2,631 once invoked, about $0.0001 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.