The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external sources such as specifications, documentation, and tickets to generate its task breakdowns.
Ingestion points: Requirements are ingested from conversation context, files in the docs/ directory, and external tickets.
Boundary markers: The skill uses a structured task template for its output but does not define explicit delimiters or instructions to ignore embedded commands within the input data.
Capability inventory: The skill writes files to the local file system (.agents/tasks/) and executes CLI commands to create issues on remote trackers.
Sanitization: No explicit sanitization of the input content is performed before interpolation into tasks or commands.
Mitigation: A mandatory "Pre-Mutation Gate" requires the agent to present a summary and halt for human approval before performing any write operations or network activity.
[COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh issue create) to publish tasks to remote trackers. This is a standard functional tool for a task management skill, and its use is governed by the mandatory human approval step described in the protocol.

to-tasks