The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external Jira issues and Confluence pages and has the capability to create or modify content on those platforms. This creates a surface for indirect prompt injection where malicious instructions embedded in a ticket or page could influence the agent's behavior during workflows like status report generation or task extraction.
Ingestion points: confluence.mjs get-page, jira.mjs get, jira.mjs search (specified in SKILL.md and various workflow files).
Boundary markers: Absent. The instructions do not explicitly prompt the agent to ignore instructions embedded within the retrieved Atlassian content.
Capability inventory: jira.mjs create, jira.mjs edit, confluence.mjs create-page, confluence.mjs update-page across all script files.
Sanitization: Uses the marked and marklassian libraries to convert markdown to platform-specific formats (ADF/XHTML), providing structural validation but not semantic instruction filtering.
[COMMAND_EXECUTION]: The skill uses execFileSync in scripts/sync.mjs to execute sibling Node.js scripts (jira.mjs, confluence.mjs). While these are internal scripts, it constitutes a local command execution pattern.
[DYNAMIC_EXECUTION]: The skill dynamically resolves the paths to its internal scripts using the current file URL to facilitate execution between modules.

atlassian-rest