The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data to automate issue creation and organization. It reads local contributor guides (CONTRIBUTING.md) to resolve routing and queries existing GitHub issues to perform duplicate checks. This creates a surface where maliciously crafted external content could attempt to influence agent behavior. However, this risk is effectively mitigated by mandatory human-in-the-loop checkpoints.
Ingestion points: Workspace files (CONTRIBUTING.md, contribute/ directory), existing GitHub issue titles and bodies retrieved during duplicate searches, and user-provided context for drafting.
Boundary markers: The skill explicitly instructs the agent to ask for content edits and requires a separate, explicit publish confirmation step (SKILL.md Step 5) before any GitHub mutation commands are executed.
Capability inventory: Access to mcp_github::issue_write, mcp_github::sub_issue_write, mcp_github::add_issue_comment, and gh api for GraphQL fallback operations.
Sanitization: Relies on the user's mandatory review of the generated draft and commands before execution.
[SAFE]: The skill utilizes official GitHub MCP tools and the gh CLI. Caching mechanisms for labels and assignees follow best practices by using session memory or local .tmp/ files (noted as never committed) to minimize token consumption and avoid hitting rate limits. All referenced URLs target official GitHub or Copilot services.

fusion-issue-authoring