Skip to content

odh-ai-helpers

A comprehensive suite of developer productivity tools for Python packaging, CI/CD debugging, upstream maintenance, and workflow automation. The plugin spans three functional domains: (1) a Python packaging toolchain that can analyze build complexity, resolve full dependency trees, locate source repos, find and check licenses against Red Hat redistribution policy, discover build environment variables, and surface known packaging bugs; (2) a vLLM backport triage pipeline that fetches upstream bugfix PRs, classifies them, checks for existing cherry-picks, scores and ranks candidates, auto-cherry-picks clean fixes, pushes reports, compares requirements across versions, and summarizes Slack channel activity; and (3) general-purpose skills for ADR review panels, GitLab CI debugging, Jira chat-log uploads, and Git shallow-cloning.

The Python packaging skills are designed to work both independently and as a coordinated pipeline through the python-packaging-investigator agent, which orchestrates all packaging skills in parallel and produces a standardized build analysis report. The vLLM backport skills form a linear pipeline from PR fetching through classification, deduplication, scoring, and cherry-picking, with each stage producing JSON artifacts consumed by the next.

Plugin Details

Pipeline

odh-ai-helpers pipeline

Skills

Skill Description Invocable
/adr-review Review an Architectural Decision Record (ADR) using a team of specialist reviewer subagents and produce a consolidated report
/gitlab-pipeline-debugger Debug and monitor GitLab CI/CD pipelines for merge requests, check pipeline status, view job logs, and troubleshoot CI failures
/git-shallow-clone Perform a shallow clone of a Git repository to a temporary location
/jira-upload-chat-log Export and upload the current chat conversation as a markdown file attachment to a Jira ticket
/python-full-deps Resolve the full install-time dependency tree for a Python package with environment markers
/python-packaging-bug-finder Find known packaging bugs, fixes, and workarounds for Python projects by searching GitHub issues
/python-packaging-complexity Analyze Python package build complexity by inspecting PyPI metadata, compilation requirements, and distribution types
/python-packaging-env-finder Investigate environment variables that can be set when building Python wheels for a given project
/python-packaging-license-checker Check whether a Python package license is compatible with redistribution in Red Hat products
/python-packaging-license-finder Deterministically find license information for Python packages by checking PyPI metadata and Git repository LICENSE files
/python-packaging-source-finder Locate source code repositories for Python packages by analyzing PyPI metadata and project URLs
/vllm-backport-fetch-prs Fetch merged bugfix PRs from upstream vLLM within a configurable date window using GitHub CLI
/vllm-backport-classify Classify PRs by backport relevance using labels, title patterns, and file-existence heuristics
/vllm-backport-check-backported Check if PRs are already cherry-picked in a downstream release branch via SHA and title matching
/vllm-backport-score-rank Score and rank backport candidates by severity, scope, and risk using a deterministic composite score
/vllm-backport-push-report Push triage report to a GitHub repository with timestamped directory structure
/vllm-backport-cherry-pick Attempt automatic cherry-pick of clean backport candidates to a downstream release branch
/vllm-compare-reqs Compare Python requirements between upstream vLLM and a downstream fork to identify version mismatches and missing packages
/vllm-slack-summary Generate a concise Slack-formatted summary of vLLM backport triage results

Agents

Agent Description
python-packaging-investigator Investigates Python package repositories to analyze build systems, dependencies, and packaging complexity

Installation

/plugin install odh-ai-helpers@opendatahub-skills

Architecture

The plugin follows a modular architecture with three distinct skill families:

Python Packaging Toolchain -- Seven skills that share a common pattern of wrapping helper scripts (Python/Bash) and chaining through skill invocations. The source-finder locates repositories, the shallow-clone skill provides local access, and downstream skills (complexity, license-finder, license-checker, env-finder, bug-finder) analyze different facets. The investigator agent orchestrates all of them in parallel sub-agents.

vLLM Backport Pipeline -- Eight skills forming a linear data pipeline. Each produces a JSON artifact (raw-prs.json -> filtered.json -> candidates.json -> analyzed.json -> ranked.json -> cherry-pick-result.json) consumed by the next stage. The pipeline combines deterministic script-based steps with agent-driven semantic analysis for unclear classifications.

Utility Skills -- Standalone skills (adr-review, gitlab-pipeline-debugger, jira-upload-chat-log, git-shallow-clone) that operate independently with no inter-skill dependencies. The adr-review skill uses six parallel reviewer sub-agents with human-in-the-loop correction before synthesis.