spike-executor¶
Execute RHOAI SPIKE investigations with human-in-the-loop approval gates on OpenShift. Provides a 9-step lifecycle covering intake, plan generation, Jira ticket creation and sync, AI-powered research enrichment with hallucination detection, pytest test suite generation and execution on OpenShift clusters, rubric-based feasibility scoring with a security gate, and RFE document generation.
The skill orchestrates via a CLI (spike-executor) that handles each step as a
separate command, writing artifacts to an artifacts/ directory with a consistent
Plugin Details
- Version: 0.2.0
- Author: IKRedHat
- License: Apache-2.0
- Category: Product Planning
- Repository: IKRedHat/SPIKE-executor
- Tags: spike assessment jira research scoring rfe openshift rhoai feasibility
Skills¶
| Skill | Description | Invocable |
|---|---|---|
/SPIKE-executor |
Execute RHOAI SPIKE investigations with human-in-the-loop approval gates |
Installation¶
Claude Code
/plugin install spike-executor@opendatahub-skills
OpenAI Codex — add the marketplace, then enable spike-executor from the /plugins browser:
codex plugin marketplace add opendatahub-io/skills-registry
Architecture¶
Single-skill plugin with a 9-step CLI-driven workflow. Each step maps to a spike-executor subcommand that reads/writes from the artifacts/ directory. Breakpoints after each step pause for human review and approval before proceeding.
Key modules: plan_generator.py (Jinja2 template rendering), jira_sync.py (Jira REST API with retry, labels, comments), test_generator.py (Phase 2 plan parser + pytest suite generation), test_executor.py (pytest runner + JUnit XML parsing), scorer.py (rubric-based 0-3 scoring with domain weights and security gate), state.py (workflow persistence), cli.py (Typer CLI).
Step 5 (AI Research) is the most complex: generates a research scaffold via templates, then the LLM performs a comprehensive web research deep dive covering community health, license, architecture, performance, UBI feasibility, operator integration, security (CVE, FIPS, rootless, air-gap, supply chain), and hardware/MLOps. A validation step (validate-research) checks for hallucination via URL reachability, evidence-backed claims, TBD audit, and GitHub metrics cross-referencing.
Scoring uses a 4-domain rubric (UBI 25%, Operator 20%, Security 30%, Hardware 25%) with 14 checks, each scored 0-3. Decision bands: GO >= 80, PIVOT 55-79, NO-GO < 55. A security critical failure gate blocks GO if any security check scores 0.
Three hooks handle artifact safety: backup-artifact.sh (PreToolUse on Write), validate-artifact.sh and auto-validate-research.sh (PostToolUse on Write).