The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements tools that can read arbitrary files from the local filesystem and transmit them to external Atlassian sites.
Evidence: The parseJsonArg function in scripts/lib/atlassian.ts uses a @ prefix convention to read content directly from a file path using fs.readFileSync.
Evidence: The scripts/jira-attachment.ts script allows the agent to upload any specified local file to a Jira issue.
Risk: While these are intended features for handling large inputs and attachments, they provide a vector for exfiltrating sensitive data if the agent is tricked via prompt injection.
[PROMPT_INJECTION]: The skill processes content retrieved from Jira and Confluence, which may contain malicious instructions.
Ingestion points: Data is retrieved through scripts/jira-get.ts, scripts/confluence-get.ts, scripts/jira-search.ts, and scripts/confluence-search.ts.
Boundary markers: Retreived content is not wrapped in markers to distinguish it from system instructions.
Capability inventory: The skill has the ability to read local files and perform write operations on external Atlassian platforms.
Sanitization: There is no evidence of sanitization or escaping of the retrieved content to mitigate instruction-following risks.

atlassian