The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the treeherder-cli tool from a GitHub repository (github.com/padenot/treeherder-cli) and uses uvx to download and run lumberjackth from PyPI.
[REMOTE_CODE_EXECUTION]: Instructions involve installing and executing third-party tools using cargo install and the uvx package runner.
[COMMAND_EXECUTION]: The skill utilizes several CLI commands (treeherder-cli, lj, uvx, cargo, curl, jq) to query CI status and manage data.
[DATA_EXFILTRATION]: Network operations are performed to communicate with treeherder.mozilla.org for fetching logs and build artifacts.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it retrieves and analyzes untrusted log data from an external CI service.
Ingestion points: CI logs and test failure descriptions retrieved from treeherder.mozilla.org (referenced in SKILL.md and references/cli-reference.md).
Boundary markers: No explicit delimiters or instructions are used to distinguish log data from agent instructions.
Capability inventory: The skill allows for subprocess execution, file system access (artifacts/caching), and network requests.
Sanitization: Log content is processed without filtering or sanitization for potential injection patterns.

treeherder