The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions provide a resolution path that executes a binary located within the local workspace directory. Executing binaries from a workspace path is a security risk as the workspace contents could be manipulated by external contributors or malicious actors.
Evidence: SKILL.md instructs to use <workspace-root>/.deepline/runtime/bin/deepline if the global command is missing.
[EXTERNAL_DOWNLOADS]: The skill prompts for the installation of a package from a non-standard third-party NPM registry rather than the public registry.
Evidence: SKILL.md contains the installation command npm install -g deepline --registry https://code.deepline.com/api/v2/npm/.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from external analytics platforms, including Snowflake warehouse tables and Axiom production logs, which are potential vectors for indirect prompt injection.
Ingestion points: Snowflake tables (e.g., AERO_DB.CONVEX_RAW.PLAY_RUNS, AERO_DB.CONVEX.USAGE_EVENTS) and Axiom logs (vercel dataset) described in SKILL.md.
Boundary markers: None identified in the processing instructions for external data.
Capability inventory: The skill has the capability to execute commands via the deepline CLI and perform network operations to retrieve analytics data.
Sanitization: No sanitization or validation steps are described for data retrieved from external sources before it is processed by the agent.

deepline-analytics