The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to interact with Jira by executing the twg CLI tool. It provides specific instructions for executing the binary from standard local paths such as $HOME/.local/bin/twg on macOS/Linux and $env:LOCALAPPDATA\Programs\twg\bin\twg.exe on Windows.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Jira (issue summaries and descriptions) which could contain adversarial instructions aimed at manipulating the agent's behavior.
Ingestion points: Untrusted content is ingested via twg jira workitem get in references/workitems.md and semantic search results from jira workitem similar in references/duplicates.md.
Boundary markers: The skill recommends using field projection (e.g., --agent-fields) in references/duplicates.md to restrict the metadata visible to the model, which provides a limited logical boundary.
Capability inventory: The skill possesses significant capabilities including the ability to create, update, transition, and delete workitems across Jira projects.
Sanitization: There are no explicit instructions to sanitize or filter potential prompt injection patterns within the retrieved Jira content.

twg-jira