The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted design artifacts (PNGs, wireframes, and briefs). It implements robust security controls by wrapping external content in explicit boundary markers using the _untrusted_doc.py utility and providing instructions to treat this data as non-executable. Ingestion points include .heyeddi/designs/ files, which are sanitized (e.g., key whitelisting in load_handoff.py) before processing.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were found. All tools defined in manifest.json execute local Python scripts included with the skill. The skill explicitly forbids downloading or executing external scripts suggested by mockup text.
[DATA_EXFILTRATION]: No network operations or sensitive data access patterns were detected. File system operations are confined to the workspace and the skill's internal configuration directory.
[OBFUSCATION]: No malicious obfuscation, hidden URLs, or suspicious encoding techniques were found in the skill's instructions or scripts.

heyeddi-handoff