The Agent Skills Directory

Command Execution: The skill uses just recipes and shell hooks (build.sh) to automate the VM image construction process. These tools are used to build rootfs and kernel components, with instructions emphasizing the use of stable system paths and avoiding mutable installers.
External Resource Acquisition: The system fetches components from standard package registries (APT, NPM, and PyPI) and official binary archives for tools like Ollama and Claude. It specifically recommends using digest-bound archives and avoiding unverified remote scripts to ensure build integrity.
Indirect Prompt Injection Surface: The build pipeline ingests configuration files and scripts from profile directories. 1. Ingestion points: config/profiles/ files. 2. Boundary markers: capsem-admin validation rules. 3. Capability inventory: Docker builds and shell execution. 4. Sanitization: Pydantic models in src/capsem/builder/models.py. The skill mitigates risks by requiring explicit materialization steps and suggesting black-box verification of the resulting images.

build-images