The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/merge_lock.py calls local agent utilities such as manage-status and platform_runtime using subprocess.run. These calls are managed as a list of arguments without shell invocation, following best practices to prevent command injection. These operations are limited to updating internal session metadata and terminal display icons.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill interacts with local state files (e.g., merge.lock, build-queue.json) and logging files within the .plan directory to maintain synchronization between different plan execution sessions. The data processed consists of plan identifiers and timestamps, with no exposure of sensitive user credentials or exfiltration to external domains.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection via ingestion points in merge-queue.json and build-queue.json. Boundary markers are absent as the skill is designed for deterministic script-based coordination rather than direct LLM processing of the data. Capability inventory includes file-system manipulation and local command execution. Sanitization is present through the use of standard JSON parsing and argument validation.

manage-locks