The Agent Skills Directory

[OBFUSCATION]: The scripts/clay-to-deepline.mjs script contains a hideSpecifiers function that uses Unicode escapes to replace JavaScript keywords like import, from, and require with encoded versions (e.g., \u0069mport). This is used when generating migration modules to hide these keywords from static analysis scans.
[DYNAMIC_EXECUTION]: The skill generates JavaScript modules (.clay.js files) based on formulas extracted from the Clay platform and dynamically loads them using import() during its parity verification phase. It also uses the new Function() constructor to evaluate code strings at runtime during self-tests. The core workflow involves generating TypeScript 'play' files which are subsequently executed in the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes large volumes of untrusted external data, including social media profiles, company website content, and complex platform exports. This creates a significant surface for indirect prompt injection, particularly as the skill has capabilities for network operations, database writes, and CRM integration.
[COMMAND_EXECUTION]: Multiple scripts in the skill, such as scripts/clay-to-deepline.mjs and scripts/claude-deepline-statusline.mjs, use child_process (execSync, spawn) to execute external binaries. scripts/claude-deepline-statusline.mjs specifically executes commands defined in a user-controlled local file (~/.claude/statusline-user-command.txt).
[PERSISTENCE]: The scripts/claude-deepline-statusline.mjs script maintains state across agent sessions by writing data to ~/.claude/deepline-statusline-state.json.

deepline-gtm