The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: Risk of indirect prompt injection through external data processing.
Ingestion points: Untrusted content enters the agent's context when reading issue data via the local cache $JIRA_CACHE_DIR/{JIRA-ID}.md or through the live API tool jira_get_issue in SKILL.md.
Boundary markers: Absent. The instructions lack explicit encapsulation or delimiters to differentiate untrusted issue contents (like descriptions and comment histories) from instruction text.
Capability inventory: The skill possesses local file-writing capabilities to update caches, and shell execution capabilities via node scripts/download_jira_attachments.mjs and scripts/download_jira_attachments.sh to download attachments to disk.
Sanitization: No automatic escaping or validation is specified for incoming text data from the Jira fields before rendering it into the agent conversation.

jira-read