The Agent Skills Directory

[SAFE]: The skill implementation focuses on local task persistence and observability. The use of a bundled script for file operations follows best practices for atomicity and concurrency management.
[PROMPT_INJECTION]: No attempts to override safety filters or agent instructions were found. The instructional tags are used to define operational logic for the checkpointing mechanism.
[DATA_EXFILTRATION]: Analysis of the bundled script and instructions confirms no network operations or access to sensitive file paths. Data interaction is confined to the specific run-record directory.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote scripts. It utilizes a zero-dependency local Node.js script for utility functions, which is standard for specialized agent skills.
[COMMAND_EXECUTION]: The skill uses local shell commands (such as jq and node) to interact with its own state files. These operations do not involve untrusted input execution or privilege escalation.

checkpoint-resume