Inspiration

Release and change reviews rely on the Jira dependency graph visible when a reviewer checked it. A day later, a prerequisite may be reopened, a new blocking link may be added, two issues may form a dependency cycle, or a linked security ticket may be hidden behind permissions—yet a single issue status badge does not show what drifted since the review.

Dependency Evidence for Jira freezes the visible dependency context behind a review into a scope-bound snapshot, then rechecks current Jira state against that baseline before a team acts.

What it does

  • Explicit prerequisite scope: Inside a native Forge jira:issuePanel, the reviewer selects an actual Jira issue-link type (such as Blocks) and confirms whether inwardIssue or outwardIssue is the prerequisite using the site's own link labels (is blocked by vs blocks). Related or duplicate links are never silently guessed as blockers.
  • Bounded caller-scoped capture: Collects reachable prerequisites via api.asUser().requestJira (read:jira-work only) in bounded pages (up to 60 issues, 6 levels, 12 pages), producing a JSON record bound to site_id, root_key, link_type_ids, direction, and a SHA-256 edit-detection digest.
  • Field-level drift recheck: Comparing current Jira evidence against a saved baseline reports changed fields (status, summary, updated, dependencies), newly reachable blockers, strongly connected dependency cycles, and scope mismatches.
  • Conservative unknown-state handling: 403 forbidden, 404 missing/hidden, 429 rate-limited, or pagination-stalled items stay explicit unknown records—they never become done or authorize a release. No snapshots are stored on a server.

How we built it

  • Atlassian Forge (nodejs24.x, Runs on Atlassian): Native jira:issuePanel module built with @forge/react (UI Kit 2), @forge/bridge, @forge/resolver, and @forge/api (api.asUser().requestJira + route). Deployed to production and verified live on Jira Cloud (REL-1..REL-4).
  • Deterministic graph & diff engine: Independently implemented JavaScript traversal, Tarjan-style cycle grouping, baseline compatibility checks, and escaped self-contained offline HTML reports (src/cli.js), backed by 19 behavior tests, TypeScript UI Kit checks, and esbuild bundles (npm run check + forge lint).

Challenges we ran into

  • Permission & visibility boundaries: Jira does not distinguish a deleted key from an issue hidden by permissions (404), and multi-page JQL searches are not atomic. Designing the data contract so missing, forbidden, or truncated items remain unknown rather than disappearing prevented false "all clear" verdicts.
  • Site-specific link semantics: Different Jira sites orient inwardIssue and outwardIssue differently across custom link types, so the UI surfaces live /rest/api/3/issueLinkType descriptions before capture.

Accomplishments & What we learned

Separating frozen review evidence from release sign-off keeps human accountability clear: a Jira done status category is a workflow field, not an organization-wide proof of delivery, and making post-review drift inspectable turns dependency checks into repeatable engineering hygiene.

Disclosures

  • AI assistance: Codex assisted with architecture, implementation, tests, and documentation. Runtime behavior is 100% deterministic—it makes zero LLM or Rovo calls.
  • Open source & licensing: Original code is MIT (https://github.com/misakamikotoghr-5205/dependency-evidence-forge). Official @forge/* packages remain governed by Atlassian Developer Terms (see THIRD_PARTY_NOTICES.md and docs/landscape.md).

Built With

Share this project:

Updates

Submission history