Aladdin V2 is the engineering method distilled from MRRC: IT veteran judgment awakened by vibe coding and agentic engineering, then disciplined by FDE, SDD, harness, and product leverage.
A software engineering method for turning veteran experience into reusable agentic engineering assets.
The seven-layer model collapses to one operational loop. Each layer answers a single question, carries one constraint, and buys one effect.
Who: veteran + field truth
Decide what to do, what not to do, what must hold — before any code.
Who: AI / agent, inside the boundary
Vibe coding for speed; the FDE rhythm (Echo→Delta→Product) for direction.
Who: human + AI together
SDD, harness, ADR, postmortem — the carriers that outlive this cycle.
Constrain, execute, capture — then the next boundary is cheaper to draw.
Radios, USB audio devices, ATR-1000 tuner, serial control, RTL-SDR, TLS, and local system dependencies prevent purely imaginary solutions.
PTT timing, TX/RX audio queues, AudioWorklet buffering, Opus frame alignment, and WebSocket scheduling expose real engineering risk.
FDE retrospectives, SDD 14 chapters, DESIGN docs, postmortems, AGENTS, and agentic plans show how knowledge becomes reusable.
The methodology is operationalized as an OpenCode skill: .opencode/skills/aladdin-v2/SKILL.md.
L1Write the boundary before the code. No boundary, no agent.
L1Field truth outranks opinion; speed lives only inside the boundary.
L2Delta code is disposable; the captured asset is what you keep.
L3Every cycle must make the next one cheaper — or it isn't finished.
∞A method must mark its own range. Outside it, bring your own equipment.
A method is only trustworthy once it marks its own boundaries. Here is where Aladdin V2 is strong, and where it has never been pressure-tested.
py_compile checks, expected output, even pre-written commits.AGENTS.md guardrails.Pick what you need. Lowest barrier: just use the Boundary Card. The rest scale with team maturity.
Under 10 lines, written before anyone (or any agent) starts: goal, what's out of scope, what must hold, the 3 things most likely to be guessed wrong, how failure is detected. Highest ROI, lowest migration cost.
The original six (Context / Spec / Plan / Verification / Runtime / Memory) plus a new Safety/Cost harness: what an agent may touch, the budget ceiling, the human-confirm points before production.
Replace qualitative scores with falsifiable trend lines: handoff exploration time, repeat-incident count, marginal cost of a similar feature. Leverage is real only if these curves bend down.
At each key decision, one lightweight challenge: a peer's reverse argument, an "if the veteran is wrong" stress test, or an evidence-source audit. A falsification entry for the single judgment source.
Mark applicability before adopting: stable domain, concentrated judgment, long feedback cycle = high fit. High-churn discovery or regulation-over-field domains need the FDE rhythm rewritten.
Correction loop ↔ single-hero gap · Range annotation ↔ n=1 overfit · Hard metrics ↔ qualitative self-comfort · Safety/Cost harness ↔ the cost & safety silence.
Aladdin V2 is maintained in the MRRC repository under docs/aldv2/, together with the OpenCode skill definition.