The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process data from various external sources, which creates an inherent surface for indirect prompt injection where malicious instructions could be embedded in the processed content.
Ingestion points: The documentation in SKILL.md and references/workflows.md describes adding sources via nlm source add using flags for web URLs (--url), text content (--text), and Google Drive documents (--drive).
Capability inventory: The CLI tool enables capabilities such as deleting notebooks and sources (nlm notebook delete), extracting data into tables (nlm data-table create), and querying source content (nlm notebook query).
Boundary markers: The skill includes explicit instructions for the agent to ask for user confirmation before executing destructive commands and notes the requirement for the --confirm flag.
Sanitization: No specific content sanitization or escaping procedures are described in the documentation for handling the imported source text.
[COMMAND_EXECUTION]: The skill documents the execution of the nlm CLI and related shell commands for administration and troubleshooting.
Authentication: The nlm login command is described as launching a Chrome session to extract cookies, which is the primary mechanism for the tool to authenticate with the target service.
Process Management: In references/troubleshooting.md, standard commands like pkill -f "Chrome" and kill -9 <PID> are suggested to resolve port conflicts that may occur during browser automation tasks.
Environment-Specific Flags: The troubleshooting guide references the use of the --sandbox danger-full-access flag for OpenAI Codex environments to enable network connectivity required for the tool's operation.

nlm-cli-skill