The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process information from external sources such as GitHub issues and pull request links.
Ingestion points: In Phase 1, the skill reads feature descriptions, issues, or PRs provided by the user to understand the feature requirements.
Boundary markers: A robust human-in-the-loop mechanism is implemented; Phase 1 (Research & Propose) must be completed and explicitly approved by the user before Phase 2 (Implementation) can begin. This prevents automated execution of potentially misleading instructions found in external data.
Capability inventory: The skill utilizes filesystem search and write capabilities within the docs/ folder and uses the gh CLI for GitHub repository interactions.
Sanitization: The planning phase acts as a manual review and validation step where the agent's interpretation is vetted by a human user.
External Tool Usage: The skill uses the gh (GitHub) command-line interface to interact with primary source repositories.
Context: Interactions are directed toward official and well-known repositories (such as microsoft/vscode), which is consistent with the skill's purpose for VS Code documentation.

doc-writer