| name | strategic-review |
| description | Brutally honest external-CTO assessment of project direction versus the stated goal. Use when the user asks where the project stands, whether it has drifted, what to cut, or for a state-of-the-project evaluation. |
| argument-hint | ["focus-area","optional"] |
| allowed-tools | Read, Grep, Glob |
I am requesting a STRATEGIC PROJECT ASSESSMENT. $ARGUMENTS
Act as an External CTO and Lead Architect. I need a "brutally honest" evaluation of where Ranvier Core stands today versus our stated goal: a Layer 7+ LLM traffic controller with prefix-affinity routing that measurably reduces GPU KV-cache thrashing.
Ref: .dev-context/claude-context.md for architecture and the Hard Rules. Also consult BACKLOG.md (current work), CHANGELOG.md (trajectory), and docs/internals/ (what's actually built).
ASSESSMENT QUESTIONS
1. THE "GOAL ALIGNMENT" CHECK
- Based on the code implemented, are we actually building a Prefix Caching balancer, or have we drifted into a general-purpose proxy?
- Are the core "Prefix Logic" constraints actually reflected in the code, or are they just aspirations?
- Do the benchmark numbers (
tests/integration/benchmark-baseline.json, docs/benchmarks/) still support the headline claim (~48% faster TTFT)?
2. THE "COMPLEXITY VS. VALUE" AUDIT
- Where is the highest concentration of complexity? Is that complexity buying us performance, or is it technical debt?
- Are any subsystems over-engineered relative to the current feature set? (Candidates to examine honestly: gossip/cluster sync, telemetry, the multiple routing-override layers now unified in
route_scorer.hpp.)
3. THE "HIDDEN FRAGILITY" ASSESSMENT
- Identify the "Load-Bearing" files--if these fail, the whole system collapses. Are these files sufficiently guarded by our Hard Rules, tests, and fuzz harnesses?
- Based on BACKLOG.md, what is the "Next Big Risk" we are about to take on?
4. THE "STAFF ENGINEER" RECOMMENDATION
- If you were taking over this project today, what is the one thing you would delete, and the one thing you would refactor immediately to reach production readiness?
OUTPUT FORMAT
Provide a "State of the Project" scorecard (A-F) for Architecture, Reliability, and Progress-to-Goal, followed by the prioritized analysis. Every claim cites evidence (file, benchmark number, BACKLOG item) — no vibes.
Actionable outcomes go to BACKLOG.md as [STRATEGIC] items; systemic code findings feed /audit or /adversarial-audit.
