cd /news/ai-agents/i-submitted-bastion-to-codegeist-liv… · home › topics › ai-agents › article
[ARTICLE · art-173732] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

I submitted Bastion to Codegeist: live CVE triage inside Jira

A developer built Bastion, a Forge app for the Codegeist 2026 hackathon that brings live CVE triage into Jira through a Rovo agent, MCP tools, and an issue panel. Every CVE answer is pulled from the NVD API at request time, and risk scores come from deterministic, severity-weighted rules rather than model guesses; a daily sweep only adds labels and never edits ticket text. The app is deployed as Bastion v2.2.0 on a Jira site, but the developer says it has no production users yet and is seeking feedback on the false-confidence failure mode, where a clean scan means only that no obvious CVE or package signal was found in a ticket.

recommendation

Ground agent answers in live API lookups and keep scoring deterministic and auditable; when no record exists, return silence rather than a fabricated result.

more in playbook: Designing agent systems
by read2 min views1 publishedNov 21, 2026

Security findings usually live in scanners, feeds, and chat threads. Engineering work lives in Jira. Bastion is my attempt to close that gap for the Codegeist 2026 hackathon: a Rovo agent + MCP tools + Jira issue panel that answers security questions with live NVD data instead of guesses.

Submission is live on Devpost: https://devpost.com/software/bastion-om3npy A ticket says "log4shell" or "CVE-2024-3094" and then nothing happens. Someone has to context-switch to a vulnerability feed, check whether the affected product is actually in the stack, decide severity, and file work. By the time that happens, the signal is stale.

Bastion keeps the security context next to the work item:

Live data or silence. Every CVE answer comes from the NVD API at request time. If NVD has no record, Bastion says so — it does not fabricate CVSS scores or affected products.

Deterministic scoring. The risk score is severity-weighted and explainable. That matters for security UX: an agent can summarize, but the number should come from rules a reviewer can audit.

No content rewriting. The daily sweep only adds labels. It never edits ticket text, which keeps the automation boring enough to trust.

This is a hackathon submission, not a revenue claim. The Forge app is deployed as Bastion v2.2.0 on a Jira site, the Devpost page shows it submitted to Codegeist 2026, and the demo video is linked. What I do not have yet is production users telling me where the workflow breaks.

If you work on Jira-heavy teams, the failure mode I want feedback on is false confidence: a clean scan should mean "no obvious CVE/package signal in this ticket," not "this system is secure." I tried to make that distinction explicit in the UI copy, but real tickets will teach me more than my own testing. The same security-triage instinct shows up in the NEXUS store as smaller digital kits: Sigma detection rules, an incident-response runbook, and a forensics evidence-collection checklist. Store: https://e74eb3fab8202e.app.lhr.life

Free starter checklist: https://e74eb3fab8202e.app.lhr.life/free If your team tracks security work in Jira, where does the process break first: detection, triage, or follow-through?

── more in #ai-agents 4 stories · sorted by recency
── more on @bastion 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/i-submitted-bastion-…] indexed:0 read:2min 2026-11-21 · —