The Agent Skills Directory

[SAFE]: The skill implements a robust design-approval gate which prohibits the AI from performing filesystem mutations, Git operations, or creating development artifacts until the user explicitly confirms the proposed architecture. Additionally, the skill performs mandatory read-only collision checks to ensure that the target directory is empty or absent before any write operations occur, protecting existing data.
[EXTERNAL_DOWNLOADS]: The skill resolves framework and dependency versions live by querying official registries such as npm, PyPI, and crates.io using standard commands like 'npm view'. These network operations are limited to metadata retrieval for version resolution.
[COMMAND_EXECUTION]: Following design approval, the skill executes standard development tasks including initializing Git repositories and running scaffolding CLI tools. It is also designed to execute verification commands like linting and testing to ensure the integrity of the generated codebase.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied goals to recommend stacks. The risk of malicious instructions within these project goals is mitigated by several factors: 1. Ingestion points: User-provided goal input. 2. Boundary markers: The 'design-approval' gate ensures a human reviews the interpreted requirements and proposed architecture before implementation. 3. Capability inventory: Filesystem writes, command execution for scaffolding, and network lookups for package versions. 4. Sanitization: Instructions mandate the validation of configuration at startup and the redaction of sensitive telemetry.

woostack-bootstrap