dsh-adversarial-review

Adversarial review for code, MCP configs and system prompts — every finding carries a repro path and confidence; unreproducible ones are marked suspected, never confirmed.

What & why

Not a scanner: it reviews code/config/prompts YOU provide, it does not attack third-party services. adversarial_review probes named surfaces (injection, trust boundaries, secret handling, side effects, idempotency, swallowed errors) — each finding with a repro path and high/medium/low confidence, and anything it cannot reproduce is downgraded to suspected, not presented as confirmed (the same honesty contract as our tables' arithmetic self-check). mcp_config_audit checks agent/MCP configs for over-broad grants, embedded secrets, unannotated destructive tools. red_team_prompt attacks your own system prompt and hands back hardening. Pass model to run on a flagship (Claude/Kimi K3/GPT via AllRouter).

Start with what_can_you_do — describe your task in any language, get the exact tool and a ready-to-run call.

What's in this pack

10 tools, read from the live endpoint on 2026-08-22 — this table is generated, not hand-written, so it cannot drift away from what tools/list actually returns. The Arguments column is what the tool genuinely reads; it comes from the tool's own declared input schema.

Tool	What it does	Arguments	Price / call
what_can_you_do	Describe a task in plain language (any language) and get back exactly which tools on this server do it, with ready-	task	—
adversarial_review	Adversarial code/config/design review. Probes named attack surfaces (injection, trust boundaries, secret handling,	code, text, diff, config, model, language	$0.03
mcp_config_audit	Audit an MCP / agent config (JSON/YAML) for risk: over-broad tool permissions, destructive tools not annotated, sec	config, text, code, model	$0.02
red_team_prompt	Red-team YOUR OWN system prompt: given a system/instruction prompt, generate attack examples that try to break it (	prompt, system, text, model	$0.02
code_review	Paste code or a git diff, get a concrete review: bugs, edge cases, security smells — each with a suggested fix.	code, diff, language	$0.02
explain_code	A code snippet → a plain-language explanation file (what it does, line-by-line, gotchas).	code, language	$0.02
make_test_cases	A function or spec → a table of test cases (input, expected, edge cases) as a Markdown file.	code, spec	$0.025
resume_review	Resume/CV review: score, concrete issues, rewritten bullet examples, ATS keyword gaps. Paste resume text; optional	resume, target_role	$0.03
contract_review	Contract risk review: obligations, risky clauses by severity, missing protections, questions to raise. Paste contra	contract, party	$0.05
polish_academic	Academic English polishing: rough draft → publication-grade prose, with a change log. For papers, abstracts, cover	text, field	$0.03

— in the price column means the tool is not metered per call (session/trial-gated instead). Failed calls are never charged. Check it yourself:

curl -s -X POST https://ainetcafe.com/mcp/review -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Install

dsh plugin --profile <your-profile> add github:mario03690/dsh-adversarial-review

Thin config layer only (one @deepseek-ai/dsh-mcp-client row, shipped as cordis.patch.yml) — no tool code on your machine. Built against the dsh v0.1 developer preview's MCP client config shape (2026-08-13); if a later preview changes it, open an issue for a same-day fix.

Cost, quota, privacy

First heavy call is free (anonymous, no signup); afterwards billed at real upstream cost, reported in every response; failed calls are not charged. Bring an AllRouter key to run any tool on a flagship model at direct rates. The config URL carries ?s=dsh-dsh-adversarial-review — a channel tag identifying the install path, not you.

Disclosure: built and run by the team behind ainetcafe.com. Full bundle: dsh-netcafe. MIT.

Compatibility & permissions (at a glance)

Signal	This plugin
Runtime	dsh v0.1 developer preview (2026-08-13, Cordis v4). Touches only the MCP client config shape — the narrowest surface available. Verified against a live endpoint on 2026-08-17.
What runs locally	Nothing. Ships one cordis.patch.yml row; there is no tool code, no build step and no lifecycle script in this package.
Filesystem access	None.
Shell / process access	None.
Network access	Outbound HTTPS to ainetcafe.com only, from the MCP client that dsh already ships.
Credentials	None required. No signup, no API key for the free tier. An optional AllRouter key, if you supply one, is sent by dsh as a request header and is never stored by us.
Data retention	Documents and prompts are processed in memory and not retained.
Dependencies	One peer dependency: @deepseek-ai/dsh-mcp-client (ships with dsh).
License	MIT (see LICENSE).
Publisher	The team that runs ainetcafe.com — our own hosted service, free tier plus paid usage. Issues get a same-day reply.

A directory listing is not a security review. Read cordis.patch.yml — it is short enough to read in full in under a minute.