The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/setup.sh script downloads a binary executable (kubeconform) from an external GitHub repository (github.com/yannh/kubeconform) that is not on the trusted vendor list.
[EXTERNAL_DOWNLOADS]: The skill installs the ruamel.yaml Python package from the public PyPI registry using pip3 during its setup phase.
[COMMAND_EXECUTION]: The agent is instructed to use system commands such as grep, ls, and cat to search for and read content from the user's local documentation repository.
[COMMAND_EXECUTION]: The skill executes local shell scripts (scripts/setup.sh) and Python scripts (scripts/yaml_check.py) to manage the environment and validate documentation content.
[REMOTE_CODE_EXECUTION]: The workflow involves downloading an external binary (kubeconform) and subsequently executing it to validate Kubernetes YAML manifests processed or generated by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted documentation files from a local repository and performing actions based on their content.
Ingestion points: Documentation files (.mdx) located in the repository path provided by the user (as seen in SKILL.md Phase 0 and 2).
Boundary markers: The skill uses a structured multi-phase workflow with explicit instructions to load rule files (rules/*.md) and templates, providing organizational structure but no formal isolation.
Capability inventory: The skill possesses the ability to read the filesystem (grep, ls, cat), perform network downloads (curl), and execute arbitrary commands via shell scripts and the kubeconform binary.
Sanitization: There is no evidence of sanitization or validation of the documentation content before it is processed by the agent or passed to validation tools.

doom-doc-assistant