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.
npx skills add CardSorting/LUMI-JOY --skill adversarial-auditorgit clone --depth 1 https://github.com/CardSorting/LUMI-JOYWrote 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.
[](https://agentmods.dev/skills/cardsorting/lumi-joy/adversarial-auditor)<a href="https://agentmods.dev/skills/cardsorting/lumi-joy/adversarial-auditor"><img src="https://agentmods.dev/badge/skills/cardsorting/lumi-joy/adversarial-auditor.svg" alt="Measured on agentmods" height="20"></a>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 5.1 | $0.00036 | $0.00349 |
| Opus 5 | $0.00018 | $0.00175 |
| Sonnet 5 | $0.00007 | $0.00070 |
| Haiku 4.5 | $0.00004 | $0.00035 |
Grade A, and why
adversarial-auditor 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 8d 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.
What it actually says
Adversarial Auditor Skill
Deep forensic and adversarial scrutiny skill distilled from the BroccoliDB compaction learnings and Osmosis methodology.
📋 Operational Protocol
-
Plan Red-Teaming (
adversarial_scrutinize_plan):- Inspect proposed architecture against 5 vulnerability vectors (verification omission, rollback gap, ungrounded metrics, context amnesia, edge case absence).
- Reject fail-closed any plan lacking concrete test commands or safety boundaries.
-
Provenance Grounding Verification (
adversarial_audit_provenance):- Verify that all numbers, claims, and conclusions have character-level or semantic grounding in source evidence.
- Reject synthetic placeholders (
TBD,TODO_LATER, mocked numbers).
-
Cognitive Spend Decomposition (
adversarial_decompose_spend):- Separate compressible fluff (conversational filler, nested quotes, redundant whitespace) from irreducible task invariants.
- Target >35% token reduction on bloated prompts.
-
Completion Receipt Validation (
adversarial_verify_completion):- Inspect stdout receipts from compiler and test suites.
- Fail closed if receipts are missing, simulated-only, or contain error indicators.
🛡️ Invariants
- Zero subshell execution overhead (<0.1 ms in-memory assertions).
- Fail-closed gate enforcement.
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.
- 8d ago First seen · 34 lines · 36 tokens per session scan A 6d56cde9ebde
adversarial-auditor is a skill published in the GitHub repository CardSorting/LUMI-JOY (2 stars, last pushed 10d ago), licensed Apache-2.0. It adds 36 tokens to every session and 349 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.
Other skills, from other repositories
sast-businesslogic
Detect business logic vulnerabilities in a codebase using a three-phase approach: threat modeling (domain analysis and attack scenarios), batched verify (check exploitable gaps in parallel subagents, 3 scenarios each), and merge (consolidate batch results). Covers price manipulation, workflow bypass, limit violations…
sast-idor
Detect Insecure Direct Object Reference (IDOR) vulnerabilities in a codebase using a three-phase approach: recon (find candidates), batched verify (check authorization in parallel subagents, 3 candidates each), and merge (consolidate batch results). Checks endpoints for missing ownership or authorization checks on…
sast-sqli
Detect SQL injection vulnerabilities in a codebase using a three-phase approach: recon (find unsafe SQL construction sites), batched verify (trace user input to those sites in parallel subagents, 3 sites each), and merge (consolidate batch results). Covers string concat, f-strings, unsafe ORM methods, and dynamic…
sast-graphql
Detect GraphQL injection vulnerabilities in a codebase using a three-phase approach: recon (confirm GraphQL usage and find unsafe operation document assembly sites), batched verify (trace user input to those sites in parallel subagents, up to 3 candidate sites each), and merge (consolidate batch results). Requires…
sast-ssrf
Detect Server-Side Request Forgery (SSRF) vulnerabilities in a codebase using a three-phase approach: recon (find outbound call sites), batched verify (trace user input to destinations in parallel subagents, 3 sites each), and merge (consolidate batch results). Requires sast/architecture.md (run sast-analysis first).…
github
GitHub operations via gh CLI: issues, PRs, CI runs, code review, API queries. Use when: (1) checking PR status or CI, (2) creating/commenting on issues, (3) listing/filtering PRs or issues, (4) viewing run logs. NOT for: complex web UI interactions requiring manual browser flows (use browser tooling when available)…