The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): A hardcoded Google API key (AIzaSyDvvGGRbH4Os3Er0dYi0kE_AzE3_2b_Az8) was found in the references/skill-template.md file. This credential will be embedded into the configuration of any new skill generated by this assistant.
[COMMAND_EXECUTION] (HIGH): The references/skill-template.md file instructs the agent to execute scripts/generate_image.py using shell commands with interpolated user-controlled variables (e.g., --prompt "{{prompt}}"). If the agent does not properly sanitize these inputs, an attacker could achieve arbitrary command execution via shell metacharacters.
[PROMPT_INJECTION] (HIGH): As per Category 8 (Indirect Prompt Injection), the workflows defined in references/skill-template.md (Template A, C, and E) involve ingesting untrusted external data such as articles and stories to generate prompts. This untrusted data directly influences the execution of a script that performs filesystem operations.
Ingestion points: references/skill-template.md (Phase 1 of Templates A, C, E) specifies reading user-provided articles, documents, and story descriptions.
Boundary markers: None present in the templates to delimit external content from instructions.
Capability inventory: scripts/generate_image.py uses os.makedirs and open(..., "wb").write() to write data to the local filesystem.
Sanitization: No sanitization or validation logic is defined for the external content before it is processed or passed to the generation script.

image-skill-builder