The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and summarize content from external sources such as GitHub issues and pull requests.
Ingestion points: Phase 1, Steps 1 and 2 involve reading feature descriptions, issues, or pull request links provided by the user or fetched via the GitHub CLI.
Boundary markers: The instructions do not explicitly specify the use of delimiters (e.g., XML tags or clear separators) to encapsulate untrusted data when the agent processes it for drafting.
Capability inventory: The skill has the capability to execute shell commands via the gh CLI and write new Markdown files to the blogs/ directory.
Sanitization: There are no specific instructions for escaping or sanitizing the content retrieved from external sources.
Context: While this represents a potential surface for indirect instructions, the risk is significantly mitigated by the 'Phase 1' requirement, where the agent must present an outline and wait for explicit human approval before any files are created or modified.
[Command Execution]: The skill utilizes the gh (GitHub) command-line interface to interact with repositories.
Pattern: It uses gh to read source code, issues, and pull requests from microsoft/vscode and microsoft/vscode-copilot-chat.
Context: This is standard functionality for a developer-oriented skill authored by the vendor and is used appropriately for research within the scope of the stated task.

blog-writer