Bone Framework – Kiroween Edition

Turn chaotic operational reality into reusable, AI-powered rituals. One undead SOP skeleton. Two digital daemons. Zero forgotten rituals.

Comment

Inspiration

Bone Framework – Kiroween Edition was born from a familiar pain: teams keep dealing with the same incidents and onboarding problems, but the “solution” usually lives in Slack threads, emails, or someone’s memory. At the same time, the Kiroween Skeleton Crew brief called for a lean template that could power multiple apps and showcase Kiro’s specs, hooks, and steering. That fit perfectly with the “spine / ribs / bone marrow” metaphor: one SOP skeleton powering different “digital daemons” for incident response and onboarding, wrapped in a Halloween spellbook vibe.

What it does

Bone Framework is a shared SOP engine (“the skeleton”) that powers two apps:

  • Incident Exorcist – Paste a messy incident thread (PagerDuty + Slack style), click Summon SOP, and get a structured runbook with:
    • context, triggers, steps, owners, guardrails
    • X-Ray “bone health” status
    • “Ghosts of Incidents Past” to see previous versions
  • Onboarding Ritual – Describe a new hire’s role, tools, and goals once; the framework generates a 30/60/90 onboarding plan, plus a Ritual Simulator that walks through each step with the artifacts required.

When an LLM key is provided, the “bone marrow” calls a model (via callLLMForSOP) to synthesize SOPs from text and marks them as [LLM]. If the key is missing or quota is exhausted, it automatically falls back to deterministic [STUB] SOPs so the experience never breaks.

How we built it

We used Kiro as the primary IDE and treated almost everything as a conversation:

  • Defined a spec-driven skeleton in .kiro/specs for the SOP schema and pipeline, then let Kiro generate a TypeScript monorepo: packages/bone-framework, packages/shared, and two apps under apps/.
  • Used vibe coding to refine the pipeline (Classifier → Synthesizer → Validator), the APIs, and the React frontends for Incident Exorcist and Onboarding Ritual.
  • Added steering docs to lock in the Halloween metaphor (spine, ribs, bone marrow, daemons), so every Kiro session stayed on-theme.
  • Implemented agent hooks to keep docs/ARCHITECTURE.md in sync with pipeline.ts, and wired a filesystem MCP config so Kiro can read actual project files as live context.
  • Integrated an LLM client that generates SOP JSON matching the shared types, with a try/catch fallback to stub logic.

Challenges we ran into

  • Designing a reusable SOP schema that works for both high-stress incidents and softer onboarding flows without becoming bloated.
  • Balancing Halloween style vs professional UX: we wanted ghosts, skulls, glowing summoning circles, and animations, but the UI still needed to feel like a serious internal tool, not a game.
  • Getting LLM integration right under hackathon constraints: handling quotas, JSON-only outputs, and ensuring failures degraded gracefully to [STUB] instead of breaking the demo.
  • Using many Kiro features in one project (specs, steering, hooks, MCP) without letting the meta tooling overshadow the actual product.

Accomplishments that we're proud of

  • Delivering a true Skeleton Crew architecture: one Bone Framework skeleton and two fully working, distinct apps sharing the same SOP engine.
  • Shipping an AI-powered SOP pipeline that can genuinely synthesize incident runbooks and onboarding plans from raw text, while staying robust in stub mode.
  • Building a coherent Halloween experience end-to-end: skull logo, summoning circles, floating ghosts, X-Ray bone density, candles for 30/60/90 phases, and themed microcopy that still makes sense to an SRE manager.
  • Using Kiro “for real”: specs to define the skeleton, steering to lock the metaphor, agent hooks to keep docs updated, MCP to extend file awareness – not just a single vibe-coding session.
  • Creating a demo that tells a clear story: chaos → skeleton → ritual → digital daemon.

What we learned

  • Spec-first with Kiro is extremely powerful for hackathons: once the skeleton was clearly written down, generating and refactoring code for two apps was much easier than starting from scratch.
  • Good fallback design for AI matters: treating the LLM as “bone marrow” instead of the whole body meant the product stayed usable even when keys or quotas failed.
  • The Kiroween judging criteria (value, implementation of Kiro, quality/design) push you to think in three dimensions: real usefulness, deep Kiro usage, and polished experience all need to be present, not just one.
  • Investing in a strong narrative (skeleton + daemons + rituals) makes the project easier to explain and remember in a 3-minute demo.

What's next for Bone Framework – Kiroween Edition

  • Automatic ingestion instead of copy-paste: connect to Slack incident channels, ops email inboxes, PDFs, and wikis so new incidents and onboarding packets become RawInput without manual work, powered by MCP-style connectors.
  • Add more “ribs” on the same spine: change management rituals, compliance checks, maintenance playbooks – all as additional modules on the Bone Framework skeleton.
  • Evolve the digital daemons from passive guides into safe tool-callers that can open tickets, attach metrics, and propose updates to existing SOPs after each run.
  • Turn the hackathon prototype into a multi-tenant SaaS skeleton where teams can host their own grimoires, share rituals across squads, and use Kiro to keep the skeleton and documentation in sync as their systems grow.

Built With

Share this project:

Updates

Submission history