| name | ci-recovery |
| description | Gemini-native organization-wide CI recovery — scans all open PRs, classifies failures, and applies remediation with 2M+ token grounding |
| capability | skill.ci_recovery |
| node_name | node_skill_ci_recovery_orchestrator |
| dot_color | 208 |
CI Recovery Skill (Gemini Edition)
Organization-wide CI recovery pipeline that polls all open PRs across OmniNode-ai repositories, leveraging Gemini's massive context window to accurately classify failures and orchestrate automated remediation.
Workflow
- Scan: Gemini lists all open PRs and extracts CI failure logs for those with failing status checks.
- Classification: Categorizes failures using Gemini's deep log analysis:
flaky_test: Identified through known-flaky history or recent re-run pass.infra_issue: Runner timeouts, network glitches, or connection refusals.config_error: Missing dependencies, lock file mismatches, or env var drift.real_failure: Genuine regressions requiring code changes.
- Remediation:
- Rerun: Triggers
gh run rerun for flaky or infra issues.
- Fix Dispatch: Dispatches
ci-fix-pipeline for configuration errors or real failures.
- Reporting: Generates structured per-cycle reports and emits results to the ONEX event bus.
Gemini Advantages
- Whole-Organization Failure Analysis: Gemini can simultaneously identify that a "real_failure" in repo A is actually a side effect of a "config_error" in repo B.
- Superior Log Triage: More accurately distinguishes between transient infra noise and structural configuration errors by understanding the technical intent of the logs.
- Intelligent Fix Orchestration: Better at batching remediation actions to avoid redundant CI runs across multiple PRs.
Arguments
--repos: Comma-separated repo names to scan.--dry-run: Classify failures without applying fixes.--max-fixes: Hard cap on fix attempts per cycle.
