The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing and running MCP servers from public registries (NPM and PyPI). These include the official @anthropic/mcp-atlassian package and well-known community alternatives like mcp-atlassian (sooperset) and atlassian-mcp (xuanxt).
Evidence: Example configurations and installation commands using npx, uvx, npm, and pip are found in SKILL.md and references/mcp-server-setup.md.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates workflows that ingest potentially untrusted data from Jira issues (summaries, descriptions) and Confluence pages, which could contain adversarial instructions designed to influence the agent's logic during automated processing.
Ingestion points: External data enters the context via tools like jira_get_issue, confluence_get_page, jira_search, and confluence_search (referenced in references/common-workflows.md).
Boundary markers: While the skill doesn't explicitly demonstrate prompt delimiters for all workflows, it includes strict constraints requiring the agent to validate all queries and avoid skipping input sanitization.
Capability inventory: The agent is granted capabilities to write data back to Atlassian products via jira_update_issue, confluence_update_page, jira_add_comment, and other modification tools.
Sanitization: The skill provides an escapeHtml utility and documentation on verifying permissions with read-only probes before executing write operations.

atlassian-mcp