The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests data originating from Jira issues. The skill explicitly addresses this risk by mandating manual sanitization by a human operator.
Ingestion points: SKILL.md (Workflow steps 3 and 4) and references/044-planning-jira.md (Steps 2 and 3) where the agent receives a "sanitized issue inventory" and "sanitized issue context" provided by the user.
Boundary markers: Present. Constraints in SKILL.md and references/044-planning-jira.md explicitly warn the agent to treat raw Jira summaries, descriptions, and comments as untrusted data and not to ingest them directly.
Capability inventory: The skill can execute various jira CLI commands, including create, assign, and transition, as detailed in references/044-planning-jira.md (Step 4).
Sanitization: Required. The instructions mandate that the Jira project maintainer/operator must author or approve sanitized summaries outside of the agent context before the data is analyzed or handed off to other workflows.
[COMMAND_EXECUTION]: The skill facilitates the execution of the Jira CLI (jira) to perform issue management tasks.
[CREDENTIALS_UNSAFE]: The skill references the use of Jira API tokens. It follows security best practices by directing users to use the jira configure flow and OS-level credential stores rather than providing or pasting tokens into the agent context.
[EXTERNAL_DOWNLOADS]: The skill provides guidance for installing the Jira CLI through well-known package managers such as Homebrew (brew install) and Chocolatey (choco install), or via official release binaries.
[PRIVILEGE_ESCALATION]: The documentation includes standard manual installation steps for users, such as using chmod +x on downloaded Linux binaries to make them executable.

044-planning-jira