The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands and scripts to perform its primary functions.
It executes linkedin-cli (the linkedin binary) for all LinkedIn interactions (searches, profile access, connection requests).
It utilizes system commands for environment setup and management, including node, npm, which, and platform-specific commands for background task management (launchctl on macOS, systemctl or crontab on Linux, and schtasks on Windows).
The scripts/tick.mjs script manages a dispatcher-worker model by spawning detached Node.js processes for each registered LinkedIn account to ensure parallel processing.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external dependency to interact with LinkedIn.
The documentation and scripts/doctor.mjs instruct the user to install @linkedapi/linkedin-cli globally via npm.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through external data ingested during the lead qualification phase (Category 8).
Ingestion points: Candidate data (full name, position, location) is fetched from LinkedIn searches via linkedin-cli and stored in temporary JSON files (qualify-<batch-id>.candidates.json) by scripts/import.mjs.
Boundary markers: The config/qualification-prompt.md file defines a structured JSON output format and provides instructions for the agent to follow, though it lacks explicit delimiters to ignore instructions embedded within the lead data.
Capability inventory: The agent uses sub-agents to classify leads. The results are committed to a local SQLite database, and the qualified leads are then processed by the background scheduler to send automated connection invites.
Sanitization: There is no evidence of sanitization or escaping of the LinkedIn profile data (e.g., job titles or names) before it is interpolated into the qualification prompt for the AI to judge.
[SAFE]: The persistence mechanisms and automated execution are core, transparently documented features of the lead management pipeline.
Persistence is achieved via standard operating system tools to run the scripts/tick.mjs task periodically (defaults to every 5 minutes).
The scripts/query.mjs tool includes a protection layer (containsWrite) that prevents the execution of data modification or DDL statements, enforcing read-only access to the local database.

linkedin-growth