The Agent Skills Directory

[PROMPT_INJECTION]: No patterns of behavior override, safety filter bypass, or role-play injection were detected. The skill uses instructional language consistent with its technical documentation purpose.
[DATA_EXFILTRATION]: No network operations or commands to access sensitive local files were found. All email header examples (Message-ID, etc.) use placeholder data and standard RFC formats.
[CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or secrets are present. Placeholders like '{uuid}' or '{timestamp}' are used for demonstration purposes only.
[EXTERNAL_DOWNLOADS]: The skill references established open-source libraries from well-known technology organizations (GitHub, Lever, Mailgun) for email parsing. These are provided as educational resources rather than automated installation commands.
[REMOTE_CODE_EXECUTION]: There are no patterns suggesting the download or execution of remote scripts or arbitrary commands.
[OBFUSCATION]: No Base64, hex encoding, zero-width characters, or homoglyph substitutions were detected in the instructions or code examples.
[COMMAND_EXECUTION]: The skill contains illustrative TypeScript and SQL snippets, but does not invoke subprocesses, shells, or administrative commands (sudo).
[INDIRECT_PROMPT_INJECTION]: While the skill describes processing untrusted external email data, it specifically teaches security mitigations such as content stripping, anomaly detection for thread hijacking, and human-in-the-loop review for suspicious inputs.
[DYNAMIC_EXECUTION]: No usage of eval, exec, or runtime code generation was observed.
[METADATA_POISONING]: The skill metadata (name, description, author) accurately reflects the technical content of the file.

thread-management