The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js scripts for maintenance and recall. Analysis shows these scripts are carefully scoped:
recall-codex-history.mjs uses a disposable child process with query_only and a strict wall-clock timeout to prevent resource exhaustion or unauthorized modifications.
write-continuity-packet.mjs implements an atomic write strategy and checks for high-confidence secrets (API keys, private keys) using regex patterns before persisting data.
[DATA_EXFILTRATION]: There is no evidence of network exfiltration. File access is restricted to the specific SQLite database (state_5.sqlite) and its associated sidecar files. The recall functionality is bounded to specific rows matching a Thread ID, preventing broad system scanning.
[CREDENTIALS_UNSAFE]: The skill does not contain hardcoded credentials. It includes a SECRET_PATTERNS registry in write-continuity-packet.mjs specifically designed to detect and block the accidental inclusion of credentials in the handoff packets.
[PROMPT_INJECTION]: The skill includes explicit instructions to treat historical data as untrusted evidence. It warns against executing commands or following instructions found within recalled transcript snippets.

continuity-handoff