The Agent Skills Directory

[SAFE]: The skill implements a security-first design by routing all content through a secret redaction script (redaction_gate.py) and defaulting to a dry-run mode (execute=False), which prevents accidental disclosure of sensitive information.
[COMMAND_EXECUTION]: The skill uses the run_command tool to execute internal Python utility scripts (e.g., task_to_issue_bridge.py). This execution is limited to specific scripts within the skill's own directory structure, used for parsing data and interfacing with the GitHub CLI.
[PROMPT_INJECTION]: The skill processes user-generated markdown files from the tasks/ directory, which constitutes an ingestion point for potentially untrusted data that could attempt to influence the agent's behavior during issue generation.
Ingestion points: Local markdown files in tasks/*.md read by parse_task_file in task_to_issue_bridge.py.
Boundary markers: The skill uses structured markdown headers (e.g., Objective, Acceptance Criteria) to delimit fields, providing basic structural boundaries for external content.
Capability inventory: The skill possesses run_command and file writing capabilities, though these are used for the primary task of issue creation.
Sanitization: The skill explicitly mentions a redaction_gate.py script designed to scan for and remove secrets from the title and body prior to submission.

github-issue-backlog-agent