Pipeline Panic

A node-connecting puzzle built to answer one question: how much complexity can a guided hint system absorb before it starts to feel like documentation? CI/CD is the backdrop. The tutorial engine is the real work.

The board: typed nodes wired into a DAG that runs on a real DFS topological scheduler (each node blocks until every upstream input succeeds), with live CPU/memory budgets, objectives, and a terminal that logs each step.

Security splits into four scan types: SAST (Semgrep/CodeQL), SCA (Snyk/npm audit), Secret (Gitleaks), and Container (Trivy/Grype), each with its own findings and terminal output, plus severity-tiered gates that block on CRITICAL/HIGH and warn on the rest.

Deploy strategies are functional, not cosmetic: rolling, blue-green, and canary differ in timing, rollback speed, and blast radius, each with a configurable auto-rollback path.

Honest failure: a deploy that fails with rollback enabled lands in a distinct rolled-back state (recovered, but no new version shipped), reported as PARTIAL. Re-run for a clean deploy.

Info

I built this primarily to practice designing guided tutorial systems for a complex UI. The hard part was the hint system: it has to detect where a player is stuck and surface guidance calibrated to the chosen difficulty, enough to unblock them and no more.

Real DFS topological scheduler: a downstream node blocks until every upstream input succeeds, matching how GitHub Actions dependency chains and GitLab DAGs order work in production
Topology-detecting progressive hint system: reads the live node/connection graph to find where a player is stuck and calibrates guidance to difficulty: enough to unblock, no more. That gap is the problem I built this to solve
Shift-left security model: SecurityScan splits into SAST/SCA/Secret/Container with real tooling (semgrep, snyk, gitleaks, trivy), severity-tiered gates (CRITICAL/HIGH block, MEDIUM/LOW warn), enforced coverage thresholds, and a two-different-tools compliance gate
Honest simulation: a failed deploy rolls back to a recovered-but-not-shipped state (an accurately partial outcome, reported as such). Rolling, blue-green, and canary strategies differ in timing, rollback speed, and blast radius, and flaky tests are mitigated with real retry config. Every mechanic teaches a correct intuition
~22 spec-covered services and 6 child components on signal-driven OnPush with component-scoped DI; WCAG 2.5.5, focus-trap modals, QuotaExceededError-aware save banner; exports a valid GitHub Actions YAML skeleton mapping the visual DAG to needs: chains

About this project

Game DevTutorial DesignDevSecOpsAngular Architecture

Real DFS topological scheduler: a downstream node blocks until every upstream input succeeds, matching how GitHub Actions dependency chains and GitLab DAGs order work in production
Topology-detecting progressive hint system: reads the live node/connection graph to find where a player is stuck and calibrates guidance to difficulty: enough to unblock, no more. That gap is the problem I built this to solve
Shift-left security model: SecurityScan splits into SAST/SCA/Secret/Container with real tooling (semgrep, snyk, gitleaks, trivy), severity-tiered gates (CRITICAL/HIGH block, MEDIUM/LOW warn), enforced coverage thresholds, and a two-different-tools compliance gate
Honest simulation: a failed deploy rolls back to a recovered-but-not-shipped state (an accurately partial outcome, reported as such). Rolling, blue-green, and canary strategies differ in timing, rollback speed, and blast radius, and flaky tests are mitigated with real retry config. Every mechanic teaches a correct intuition
~22 spec-covered services and 6 child components on signal-driven OnPush with component-scoped DI; WCAG 2.5.5, focus-trap modals, QuotaExceededError-aware save banner; exports a valid GitHub Actions YAML skeleton mapping the visual DAG to needs: chains