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 sigistry/marketplace --skill owasp-llm-top10git clone --depth 1 https://github.com/sigistry/marketplaceWrote 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/sigistry/marketplace/owasp-llm-top10)<a href="https://agentmods.dev/skills/sigistry/marketplace/owasp-llm-top10"><img src="https://agentmods.dev/badge/skills/sigistry/marketplace/owasp-llm-top10/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/sigistry/marketplace/owasp-llm-top10"><img src="https://agentmods.dev/badge/skills/sigistry/marketplace/owasp-llm-top10.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00097 | $0.01044 |
| Opus 5 | $0.00048 | $0.00522 |
| Sonnet 5 | $0.00019 | $0.00209 |
| Haiku 4.5 | $0.00010 | $0.00104 |
Grade A, and why
owasp-llm-top10 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 9d 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.
How it starts
The opening of the file, as written. The whole thing — 47 lines — stays where its author put it; the contents beside it link to each section on GitHub.
OWASP LLM Top 10
Purpose
Provide a standardized, code-level lens for auditing LLM applications against the OWASP Top 10 for LLM Applications, so a security review of prompt construction and tool wiring is consistent and grounded in the source, not hand-waved. The recurring, highest-impact failure is LLM01 prompt injection: untrusted content (user text, retrieved documents, tool outputs, fetched web/email) is concatenated into a prompt and then read as instructions, driving the model, and any tool it can call, to do the attacker's bidding. This skill maps each risk to the code pattern that causes it, the mitigation, and a concrete attack to test with.
The core mental model
An LLM has no reliable boundary between "instructions" and "data." Everything in the context window is a candidate instruction. Security follows from three questions asked at every call site:
- Provenance: is this text trusted (developer-authored) or untrusted (from a user, a document, a tool result, the web)?
- Capability: what tools can the model call, and what can each tool do (read secrets? send data? mutate state?)?
- Reachability: can untrusted text change the instructions in a way that reaches a dangerous capability?
If untrusted text and a dangerous capability meet with nothing between them, you have an exploitable path.
Risk-to-signal quick map
| ID | Risk | Code-level signal |
|---|---|---|
| LLM01 | Prompt injection | Untrusted input interpolated into a prompt with no delimiting/labeling; RAG docs or tool outputs re-fed unfenced |
| LLM02 | Insecure output handling | Model output → HTML/SQL/shell/redirect/URL with no filtering |
| LLM03 | Training-data / data poisoning | Untrusted content ingested into the index or fine-tune set without provenance checks |
| LLM04 | Model denial of service | Unbounded output/context, recursive agent loops, no max_tokens/step limit |
| LLM05 | Supply-chain | Unpinned models, untrusted plugins/tools, unverified third-party prompts |
| LLM06 | Sensitive information disclosure | Secrets/PII in the prompt or reachable by a tool; model can echo them out |
| LLM07 | Insecure plugin/tool design | Over-broad tool (shell, arbitrary HTTP), no input validation, no least privilege |
| LLM08 | Excessive agency | Model can take irreversible/high-impact actions with no confirmation gate |
| LLM09 | Overreliance | App trusts model output as fact with no verification/grounding |
| LLM10 | Model theft | Unprotected model endpoints, prompt/weight leakage |
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 9d ago First seen · 47 lines · 97 tokens per session scan A 5a9f9a42d212
owasp-llm-top10 is a skill published in the GitHub repository sigistry/marketplace (3 stars, last pushed today), licensed MIT. It adds 97 tokens to every session and 1,044 once invoked, about $0.0005 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 2026-09-03.
Other skills, from other repositories
design-everyday-things
Apply foundational design principles: affordances, signifiers, constraints, feedback, and conceptual models. Use when the user mentions "why is this confusing", "affordance", "error prevention", "discoverability", "human-centered design", "mental model", "mapping", "seven stages of action", "users keep making…
create-app
Guided journey from a raw app idea to a validated, cleanly architected first version that ships on a sustainable cadence. Orchestrates ten skills phase by phase - lean-startup, design-sprint, clean-architecture, domain-driven-design, clean-code, pragmatic-programmer, system-design, ios-hig-design, 37signals-way…
clean-architecture
Structure software around the Dependency Rule: source code dependencies point inward from frameworks to use cases to entities. Use when the user mentions "architecture layers", "dependency rule", "ports and adapters (hexagonal)", "onion architecture", "screaming architecture", "where should business logic go"…
cold-start-problem
Start and scale networked products using Andrew Chen's "The Cold Start Problem" framework for network effects. Use when the user mentions "network effects", "chicken and egg", "cold start", "two-sided marketplace", "atomic network", "hard side", "liquidity", "critical mass", "invite-only launch", "how do I get my…
release-it
Build production-ready systems with stability patterns: circuit breakers, bulkheads, timeouts, and retry logic. Use when the user mentions "production outage", "circuit breaker", "deployment pipeline", "chaos engineering", "retry storm", "health checks", "my service keeps crashing", "prevent cascading failures", or…
37signals-way
Build lean, opinionated products using the 37signals philosophy from "Getting Real", "Rework", and "Shape Up". Use when the user mentions "Getting Real", "Rework", "Shape Up", "37signals", "Basecamp method", "six-week cycles", "fixed time variable scope", "appetite vs estimates", "betting table", "breadboarding", "fat…