The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references Chart.js and an annotation plugin from the JSDelivr CDN. These external resources are loaded using pinned versions and Subresource Integrity (SRI) hashes, which is a standard security best practice for managing external dependencies.
[COMMAND_EXECUTION]: The skill uses standard shell utilities including mkdir and cat to maintain a local data cache in /tmp, which helps prevent data loss during long context window operations. It also uses the open command to present the final generated report to the user.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the following mechanisms:
Ingestion points: External data is ingested from earnings transcripts and Kensho search results, which are stored in /tmp/earnings-preview/transcript-extracts.txt and /tmp/earnings-preview/kensho-findings.txt.
Boundary markers: While the skill has strong instructions regarding data grounding and verbatim quotes, it lacks explicit instructions to ignore or strip executable code or instructions embedded within the ingested data.
Capability inventory: The agent can write to the local filesystem and invoke the open command on the resulting file, which serves as the execution vector for any potential payload.
Sanitization: There is no evidence of sanitization or escaping of the ingested text before it is inserted into the HTML report template, creating a surface for potential Cross-Site Scripting (XSS) if the source data is malicious.

earnings-preview-single