The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were identified in the task selection logic.
[DATA_EXPOSURE]: The skill interacts with local project files such as prd.json and prd_backlog.json to manage task states. This behavior is restricted to the project context and does not involve sensitive system files or credentials.
[COMMAND_EXECUTION]: While the skill contains JavaScript snippets, these are presented as instructional logic for the agent to follow. There are no calls to external shells or unsafe execution of user-provided strings.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from task descriptions in the PRD. While this presents an attack surface if task content is sourced from untrusted external users, the skill's logic is focused on structural metadata (categories, dependencies, status) rather than executing instructions found within the data.
Ingestion points: Task items loaded from prd.json and prd_backlog.json (SKILL.md).
Boundary markers: None identified for individual task fields.
Capability inventory: File writes to prd.json and prd_completed.json, and task assignment operations (SKILL.md).
Sanitization: Not explicitly defined in the logic snippets.

pm-organization-task-selection