The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read product marketing context from external files, which creates a surface for indirect prompt injection if those files contain malicious instructions.
Ingestion points: Read operations are performed on .agents/product-marketing.md, .claude/product-marketing.md, and product-marketing-context.md (found in SKILL.md).
Boundary markers: No boundary markers or "ignore instructions" delimiters are specified for the ingested content.
Capability inventory: The skill possesses the capability to execute shell commands and perform network requests (via curl), which could be abused if malicious instructions are successfully injected via the context files.
Sanitization: There is no evidence of sanitization or validation of the data read from the external files.
[COMMAND_EXECUTION]: The skill provides several shell commands for image optimization and web inspection, which may be executed by the agent or user.
Evidence: Commands include cwebp, mogrify, and jpegoptim (found in SKILL.md).
[DATA_EXFILTRATION]: Includes network operations that target non-whitelisted domains.
Evidence: A curl command targeting https://yoursite.com is provided in SKILL.md for the purpose of inspecting website images.

image