The Agent Skills Directory

[DYNAMIC_EXECUTION]: The library active_lib.py and test script test_offline.py utilize exec() and compile() to load and execute core logic. This implementation ensures that the identical business logic used for local previews and smoke tests is successfully deployed to the remote Clay workflow steps.
[CREDENTIALS_UNSAFE]: The skill manages API keys for Jev (TypeSafe/OpenRouter). It follows security best practices by storing them locally in .env files with 0600 permissions. However, it explicitly writes these keys into the first step of the generated Clay workflow. The skill provides clear documentation that this behavior exposes the key to any user with access to the Clay workspace, noting it as a temporary workaround for limitations in the Clay CLI.
[COMMAND_EXECUTION]: The skill uses subprocess.run to interact with the clay CLI for workspace management and to trigger system-level UI dialogs (such as osascript, zenity, or kdialog) for secure key entry. These operations are scoped to the skill's primary installation and management tasks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests LinkedIn profile data, which is considered untrusted external input. This data is processed and interpolated into questions for the Jev decision model. The risk of prompt injection is mitigated by the use of structured Jev question types (Noul and Choice), which restrict the model's output to specific probabilities or categories rather than free-form text.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to api.typesafe.ai and openrouter.ai to access the Jev model. These are legitimate endpoints for well-known services required for the skill's stated functionality.

check-employment-with-jev