The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external GitHub issues (titles, bodies, and acceptance criteria), which creates a surface for instructions embedded in that data to manipulate the task planning logic.
Ingestion points: SKILL.md Step 2 specifies gathering data from GitHub issue fields including title, body, acceptance criteria, and scenarios.
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the ingested issue data.
Capability inventory: The skill has the ability to write draft files to the local .tmp/ directory and delegate repository mutations (issue creation/linkage) to the fusion-issue-authoring sub-agent.
Sanitization: There are no explicit content filtering or sanitization rules defined for the processed issue data.
[COMMAND_EXECUTION]: The skill requires and interacts with the GitHub Model Context Protocol (MCP) server to perform repository-related tasks.
Evidence: SKILL.md frontmatter lists github as a required MCP server and references mcp_github tools for workflow execution.

fusion-issue-task-planning