The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the user to execute sudo mv ./kind /usr/local/bin/kind. This requires elevated privileges to move a downloaded binary into a system-protected directory.
[REMOTE_CODE_EXECUTION]: The installation guide includes a pattern of downloading a binary from an external URL (kind.sigs.k8s.io), granting it execution permissions via chmod +x, and moving it to the system execution path.
[EXTERNAL_DOWNLOADS]: Downloads the kind binary from kind.sigs.k8s.io, which is a well-known service managed by the Kubernetes project.
[EXTERNAL_DOWNLOADS]: Clones the project source code from the vendor's repository at github.com/WeaveMindAI/weft.git.
[DYNAMIC_EXECUTION]: The skill documentation describes a Code node that allows for the execution of Python code within the defined AI pipelines.
[DYNAMIC_EXECUTION]: The provided dev.sh script is intended to automate the installation of dependencies such as Rust and Restate.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by processing external data through LLM nodes.
Ingestion points: Untrusted data enters the agent context through nodes such as WebSearch, HumanQuery, Discord, and Slack (SKILL.md).
Boundary markers: The skill utilizes LlmConfig nodes to provide systemPrompt instructions to the models.
Capability inventory: The skill has access to network operations (API nodes), database persistence (Postgres), and arbitrary code execution (Python node).
Sanitization: No explicit sanitization or input validation for external data is documented before the data is interpolated into prompts.
[COMMAND_EXECUTION]: The skill documentation includes various shell commands and scripts (./dev.sh, ./cleanup.sh, cargo, docker) for managing the lifecycle of the services and infrastructure.

weft-ai-language