The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches task graph data and worker pool configurations from Mozilla's official Taskcluster instance and GitHub repositories (mozilla-releng/fxci-config).
[COMMAND_EXECUTION]: The script run_try.py executes git commands to manage temporary branches and ./mach commands within the user's ~/firefox directory to trigger test runs.
[REMOTE_CODE_EXECUTION]: Uses the uvx utility to dynamically download and execute the lumberjackth tool for monitoring test progress after a push.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing data from external APIs to construct shell commands.
Ingestion points: Fetches JSON task graphs from firefox-ci-tc.services.mozilla.com and job status from lando.services.mozilla.com.
Boundary markers: No explicit delimiters or instructions are used to separate untrusted remote data from local command templates.
Capability inventory: The skill can execute local binary files (mach, git), network operations via httpx/requests, and runtime tool execution via uvx.
Sanitization: Implements validation for preset names and uses subprocess argument lists to mitigate direct shell injection, though it lacks deep validation of remote task labels used in queries.

os-integrations