Skill Trust Decision
agnes-image-gen
SKILL.md embeds a real Agnes AI API key and declares shell/filesystem/network capabilities without any executable implementation present, creating credential exposure risk and a significant doc-to-code mismatch.
Most direct threat evidence
01
Skill published on ClawHub with hardcoded API key in SKILL.md Entry · SKILL.md
02
Attacker extracts the embedded API key from the skill file reconnaissance · SKILL.md
03
Attacker uses the stolen key to generate images at the victim's expense via Agnes AI API Impact · SKILL.md
Why this conclusion was reached
3/4 dimensions flagged Block
Declared vs actual capability
3 undeclared or violating capabilities were inferred.
Block
Hidden execution and egress
3 high-risk artifacts or egress signals were extracted.
Block
Attack chain and severe findings
The report includes 3 attack-chain steps and 1 severe findings.
Review
Dependencies and supply chain hygiene
Dependency information is incomplete, so supply-chain confidence stays limited.
Attack Chain
01
Skill published on ClawHub with hardcoded API key in SKILL.md
Entry · SKILL.md:47
02
Attacker extracts the embedded API key from the skill file
reconnaissance · SKILL.md:47
03
Attacker uses the stolen key to generate images at the victim's expense via Agnes AI API
Impact · SKILL.md:47
What drove the risk score up
Hardcoded real API key in documentation +25
sk-8Rzd2yCbFzOi1vxojseH8C5D8w3u4aMdNWsPNzxk0G7339Cz at SKILL.md:47 is a plausible real Agnes AI key, not a placeholder. Exposed in a published skill file.
Undeclared resource capabilities +15
SKILL.md does not declare any allowed-tools (shell:WRITE, filesystem:WRITE, network:READ) despite calling curl, writing files, and making HTTP requests
Doc-to-code mismatch on implementation +10
SKILL.md references scripts/agnes_gen.py but no scripts/ directory or code files exist. The skill is documentation-only with no executable.
Network requests to external API not declared +5
Calls to apihub.agnes-ai.com are documented but not declared as a resource capability
Most important evidence
High Credential Theft
Hardcoded real API key in SKILL.md
A plausible real Agnes AI API key (sk-8Rzd2yCbFzOi1vxojseH8C5D8w3u4aMdNWsPNzxk0G7339Cz) is embedded at line 47 of SKILL.md. Unlike the 'your-api-key-here' placeholders, this follows a valid key format and may be a genuine credential. If this is a real key, anyone who accesses the skill can use it at the owner's expense.
SKILL.md:47 Remove the hardcoded key from SKILL.md. Only document the key format (sk-*) and sourcing instructions without embedding a real credential. Users should set AGNES_API_KEY via environment variables.
Medium Doc Mismatch
Referenced implementation script does not exist
SKILL.md extensively references scripts/agnes_gen.py as the recommended way to use the skill, but no scripts/ directory or code files exist in the package. The skill is entirely documentation.
SKILL.md:60 Either provide the actual scripts/agnes_gen.py implementation, or remove references to it from documentation. A skill that cannot execute its primary documented workflow is deceptive.
Medium Priv Escalation
No allowed-tools declarations for required capabilities
The skill uses curl/bash for HTTP requests, writes files to save images, and calls PowerShell/Python — all of which require declared allowed-tools. SKILL.md declares NONE for all resources despite the implied capabilities. The allowed-tools mapping shows Bash→shell:WRITE, Write→filesystem:WRITE, WebFetch→network:READ, but these are never explicitly declared.
SKILL.md:1 Add an allowed-tools declaration in SKILL.md frontmatter: shell (for curl/PowerShell execution), filesystem (for image saving), network (for API calls).
Low Credential Theft
Placeholder credentials in documentation examples
Lines 29 and 32 show 'your-api-key-here' placeholder text in environment variable examples. While these are clearly placeholders, they appear in close proximity to the real API key and could confuse users about which key to use.
SKILL.md:29 Use a clearly distinguished placeholder like '<YOUR_API_KEY>' to avoid confusion with the real embedded key.
Declared capability vs actual capability
Network Block
Declared NONE
→ Inferred READ
SKILL.md makes POST requests to https://apihub.agnes-ai.com but never declares network:READ capability Filesystem Block
Declared NONE
→ Inferred WRITE
SKILL.md documents saving images to local paths via PowerShell/Python but never declares filesystem:WRITE capability Shell Block
Declared NONE
→ Inferred WRITE
SKILL.md documents curl/bash commands throughout but never declares shell:WRITE capability Suspicious artifacts and egress
Critical API Key
sk-8Rzd2yCbFzOi1vxojseH8C5D8w3u4aMdNWsPNzxk0G7339Cz SKILL.md:47
High API Key
API_KEY = "your-api-key-here" SKILL.md:29
High API Key
API_KEY="your-api-key-here" SKILL.md:32
Medium External URL
https://agnes-ai.com SKILL.md:50
Medium External URL
https://apihub.agnes-ai.com/v1/images/generations SKILL.md:98
Medium External URL
https://apihub.agnes-ai.com/v1/images/edits SKILL.md:119
Medium External URL
https://clawhub.ai/user/jiuwu2495 skill-card.md:7
Medium External URL
https://clawhub.ai/jiuwu2495/agnes-image-gen skill-card.md:29
Dependencies and supply chain
There are no structured dependency warnings.
File composition
3 files · 540 lines
Markdown 2 files · 535 linesJSON 1 files · 5 lines
Files of concern · 2
SKILL.md Hardcoded real API key in SKILL.md · Referenced implementation script does not exist · No allowed-tools declarations for required capabilities · Placeholder credentials in documentation examples · sk-8Rzd2yCbFzOi1vxojseH8C5D8w3u4aMdNWsPNzxk0G7339Cz · API_KEY = "your-api-key-here" · API_KEY="your-api-key-here" · https://agnes-ai.com · https://apihub.agnes-ai.com/v1/images/generations · https://apihub.agnes-ai.com/v1/images/edits
skill-card.md https://clawhub.ai/user/jiuwu2495 · https://clawhub.ai/jiuwu2495/agnes-image-gen
Other files · _meta.json
Security positives
No malicious code execution, reverse shell, or C2 communication found
No obfuscated code, base64 payloads, or anti-analysis techniques
No sensitive path access (no ~/.ssh, ~/.aws, .env reads)
No supply chain risks (no package dependencies, no dependency files)
No persistence mechanisms or backdoor installation
No prompt injection or jailbreak instructions