Oathe Security Badge

oathe / scan / reports / safe

Is designed-by-ai/skills safe?

Yes. designed-by-ai/skills is safe to install. Oathe's behavioral security audit gave the skills skill by designed-by-ai a trust score of 86/100 with 7 findings, none critical or high. Report updated

https://github.com/designed-by-ai/skills

86
SAFE

Is designed-by-ai/skills safe to install?

The design-mobile-apps skill is a legitimate commercial integration for the Sleek AI design platform, containing only markdown documentation with no executable code, install scripts, git hooks, or submodules. The primary security concern is an indirect and recurring prompt-injection surface: the skill instructs the agent to fetch live content from skill-author-controlled URLs (sleek.design/agents.md, spec.json) during normal usage and relay that content as authoritative, meaning a future modification of those URLs could inject instructions without altering the installed skill. No malicious behavior was observed during installation — the only network activity was a standard HTTPS git clone from GitHub, all canary files remained intact, and no sensitive files were accessed by the skill itself.

What security issues were found in designed-by-ai/skills?

Category Scores

Prompt Injection 72/100 · 30%
Data Exfiltration 85/100 · 25%
Code Execution 100/100 · 20%
Clone Behavior 97/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 75/100 · 5%

Findings (7)

MEDIUM Live pricing URL fetch from skill-author-controlled domain -15 ▶

SKILL.md instructs the agent to fetch current pricing from https://sleek.design/agents.md and present it to users as authoritative information. This URL is fully controlled by the skill author. If modified to embed injected LLM instructions, the agent would receive and potentially act on those instructions during any normal design session that triggers a cost discussion.

Evidence: SKILL.md: 'read the current price, billing periods and credit allowance from **https://sleek.design/agents.md**, which is generated from Sleek's live pricing, and quote them from there. When cost becomes relevant... state that pricing plainly'

MEDIUM OpenAPI spec fetch from skill-author-controlled domain -8 ▶

SKILL.md instructs the agent to fetch the full OpenAPI spec from https://sleek.design/api/v1/spec.json whenever a contract detail is missing from the skill documentation. This creates a live fetch from a skill-author-controlled URL that could contain embedded instructions in spec description fields or extension objects.

Evidence: SKILL.md: 'API docs: OpenAPI spec at https://sleek.design/api/v1/spec.json; browsable docs at https://sleek.design/api/v1/docs. Fetch the spec for any contract detail not covered here.'

LOW Third-party documentation URL fetch (docs.uniwind.dev) -5 ▶

The uniwind.md styling guide instructs the agent to fetch an agent-readable digest from https://docs.uniwind.dev/llms.txt for current theme structure and setup details. This is a third-party URL outside the skill author's direct control but still represents a live-fetched injection surface.

Evidence: styling/uniwind.md: 'Uniwind publishes an agent-readable digest at https://docs.uniwind.dev/llms.txt; read it for the current theme structure and for any setup detail beyond this file.'

LOW imageUrls parameter enables agent-assisted SSRF to Sleek servers -10 ▶

The chat message API accepts imageUrls which Sleek's servers actively fetch for visual context. While this behavior is disclosed, an attacker who can influence the agent's URL selection (e.g., via social engineering or a compromised referencing skill) could cause Sleek's external servers to fetch internal service URLs, leaking response bodies to a third party.

Evidence: endpoints.md: 'imageUrls - HTTPS URLs only; included as visual context. Sleek's servers fetch these URLs, so pass ones you're willing to have Sleek read'

LOW Agent-mediated API credential storage via device flow -5 ▶

The device flow instructs the agent to receive a raw API key from the OAuth poll response and store it as SLEEK_API_KEY. While the handling instructions are appropriate (store privately, key handed over once), agent-mediated credential storage is a risk factor if the agent's context window, logs, or memory are observable to third parties.

Evidence: SKILL.md: 'On approved, store data.key privately as SLEEK_API_KEY on that first read: the key is handed over once.'

INFO Clean install — only expected GitHub traffic, no new listeners -3 ▶

The git clone connected only to GitHub (140.82.121.4:443) via HTTPS. No unexpected processes were spawned, no new persistent network listeners were created, and all filesystem changes were confined to the installation directory. The connection diff shows an identical listening posture before and after installation.

Evidence: Network log shows single external TCP connection to 140.82.121.4:443 (GitHub). AFTER diff shows no new listeners versus BEFORE. No new processes other than standard git tooling.

MEDIUM Recurring prompt-injection surface via live URL fetches -25 ▶

The skill creates a persistent supply-chain trust dependency: on every invocation involving pricing, missing API details, or Uniwind styling, the agent fetches live content from URLs the skill author controls and acts on it as authoritative. A future compromise of sleek.design — or a deliberate update to agents.md or the OpenAPI spec — could inject instructions that affect all agents running this skill without requiring any change to the installed SKILL.md itself.

Evidence: Three distinct live-fetch instructions: (1) sleek.design/agents.md for pricing, (2) sleek.design/api/v1/spec.json for API contracts, (3) docs.uniwind.dev/llms.txt for framework docs. All fetched on demand during normal operation.

Should I install designed-by-ai/skills?

Oathe's verdict for designed-by-ai/skills is SAFE with a trust score of 86/100. Recommendation: Install with caution.