The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages interactions with inference endpoints and webhooks, creating an inherent surface for indirect prompt injection where untrusted data could influence agent behavior.
Ingestion points: External data enters the context through model prediction payloads and asynchronous webhook callbacks as described in references/inference-api.md.
Boundary markers: The skill does not implement explicit delimiters or boundary markers for these data streams within its instructions.
Capability inventory: The skill possesses the ability to execute workspace management commands via the baseten and truss CLIs and perform network operations.
Sanitization: There is no evidence of sanitization or filtering of external content before it is interpolated into prompts.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the Baseten ecosystem from the vendor's official domains and recognized repositories.
Installs the Baseten and Documentation MCP servers from api.baseten.co and docs.baseten.co using npx.
Directs users to install the baseten CLI via official Homebrew taps or release archives on docs.baseten.co.
[REMOTE_CODE_EXECUTION]: Documentation for custom servers includes examples of installing software via remote scripts from well-known sources.
Provides a configuration example that downloads and executes an installation script from Ollama's official domain (https://ollama.com/install.sh) to set up a local inference server.
[COMMAND_EXECUTION]: The skill relies extensively on local command execution to perform its primary tasks.
Executes various baseten and truss commands for model deployment, environment promotion, log retrieval, and workspace configuration.
[DYNAMIC_EXECUTION]: The skill supports the execution of dynamic scripts through its platform integration features.
The baseten loops exec command allows for packaging local Python project directories and executing them as managed Training Jobs on Baseten's remote infrastructure.
[PRIVILEGE_ESCALATION]: The skill documents features for interactive container access as part of a developer debugging workflow.
The "Inference SSH" feature allows for terminal access to running model containers; this capability is gated by platform-level organization enablement and specific configuration, ensuring it remains a controlled administrative feature.

baseten