The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes Jira issue summaries, descriptions, and comments which are external, untrusted data sources. An attacker could embed malicious instructions in a Jira ticket that the agent then processes. \n
Ingestion points: Issue details retrieved via jira issue view (CLI) or mcp__atlassian__getJiraIssue (MCP) in SKILL.md and references/mcp.md. \n
Boundary markers: The skill lacks explicit instructions to ignore embedded commands within ticket data, although it does mandate manual approval for most actions. \n
Capability inventory: The skill possesses extensive capabilities including shell command execution (jira CLI), file reading/writing, and network operations via MCP tools. \n
Sanitization: No explicit sanitization or filtering of fetched ticket content is described. \n- [DATA_EXFILTRATION]: The skill provides instructions for reading local files and writing to temporary directories. \n
The jira issue create and jira issue comment add commands in references/commands.md support a --template flag for reading files from the local filesystem. This capability could be abused to read sensitive files if a malicious path is provided. \n
The CLI workflow in references/commands.md includes writing multi-line content to /tmp/jira_body.md, which is a shared temporary directory. \n- [COMMAND_EXECUTION]: The skill relies on executing the jira CLI and a bundled script located at ~/.claude/skills/jira/jira-link-issues as described in references/mcp.md. \n- [EXTERNAL_DOWNLOADS]: The skill recommends installing the jira-cli tool from the well-known GitHub repository ankitpokhrel/jira-cli using the Homebrew package manager.

jira