The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted LLM trace and span data which could contain malicious instructions designed to influence the agent's behavior.\n
Ingestion points: Commands such as px trace get and px span list retrieve input.value and output.value attributes from external traces (identified in SKILL.md and references/open-coding.md).\n
Boundary markers: The instructions lack specific delimiters or guardrail prompts to distinguish untrusted trace content from system instructions.\n
Capability inventory: The agent has the ability to execute shell commands, perform network operations via the Phoenix CLI, and write to local sidecar files in .px/coding/.\n
Sanitization: No explicit sanitization or filtering of the trace data is described before it is used for note-taking and taxonomy creation.\n- [COMMAND_EXECUTION]: The skill relies on shell command execution to interface with the Phoenix CLI and process data with tools like jq.\n
Evidence: Multiple command examples are provided throughout the skill files, including complex pipelines involving px and jq.\n- [REMOTE_CODE_EXECUTION]: The skill utilizes npx to fetch and execute external code from the NPM registry.\n
Evidence: SKILL.md specifies the use of npx @arizeai/phoenix-cli for running the debugger.\n- [EXTERNAL_DOWNLOADS]: The skill fetches documentation and packages from remote sources.\n
Evidence: The use of npx to download the CLI and the px docs fetch command to download documentation markdown files.

phoenix-cli