| name | aladdin-v2 |
| description | Use when the user mentions 阿拉丁 V2, 老灯新灵, FDE, SDD, harness, agentic engineering, vibe coding, software architecture methodology, or wants to plan/review/execute an engineering cycle using veteran judgment plus AI agents. |
Aladdin V2 Skill
Use this skill to apply the 阿拉丁 V2 / 老灯新灵 method: IT veteran judgment plus vibe coding / agentic engineering, driven by FDE cycles, structured by SDD, and constrained by harness.
Core Model
The model is three layers, one loop. Each layer answers one question, carries one constraint, buys one effect.
L1 Constrain → who: veteran + field truth → constraint: write the Boundary Card first
→ effect: turns AI's fast GUESSING into fast EXECUTION
L2 Execute → who: AI/agent inside the boundary, FDE rhythm (Echo→Delta→Product)
→ constraint: validate biggest uncertainty first, converge in 3–5 rounds
→ effect: the speed the boundary just bought
L3 Capture → who: human + AI → carriers: SDD, harness, ADR, postmortem
→ constraint: every cycle leaves ≥1 reusable asset, or it isn't done
→ effect: next time is cheaper = leverage
Loop: Boundary → Speed → Asset → (cheaper) Boundary
The earlier seven concepts are NOT gone — they are roles/carriers inside the three layers:
Aladdin V2 = 老灯新灵 + FDE Cycle + SDD Spine + Agentic Harness + Product Leverage
老灯新灵 = Veteran Judgment + Vibe Coding + Field Truth
- L1 Constrain: 老灯 (veteran judgment) + Field Truth define what to do / not do / must-hold.
- L2 Execute: 新灵 (vibe/agentic coding) for speed + FDE Cycle (Echo→Delta→Product) for direction.
- L3 Capture: SDD Spine (architecture memory) + Harness (context/specs/plans/tests/guardrails/postmortems) + Product Leverage (lower next-cycle cost).
One sentence: no boundary, no agent; rough Delta is fine, but every loop must leave an asset that makes the next boundary cheaper to draw.
When To Use
Use this skill for:
- Creating or refining engineering methodology documents.
- Planning a feature or refactor as an FDE cycle.
- Turning a vague request into Echo / Delta / Product work.
- Creating agentic plans, checklists, specs, ADRs, SDD updates, or guardrails.
- Reviewing whether a proposed implementation has enough harness and Product leverage.
- Updating
AGENTS.md, SDD, FDE docs, or repo-local OpenCode skills around engineering practice.
- Diagnosing why vibe coding produced fast but unsafe code.
Do not use this skill for small one-line edits or purely informational answers unless the user asks for methodology, planning, architecture, FDE, SDD, harness, or agentic workflow.
Always Start With Phase Detection
Before acting, classify the user's request into one phase:
Discovery: user is asking what the problem/method should be.
Echo: define field truth, demo, scope, and success metrics.
Delta: implement or plan a fast prototype to validate one uncertainty.
Product: abstract, document, add harness, update SDD/ADR/AGENTS.
Review: inspect whether work respects Aladdin V2 guardrails.
Skill-System: create or update reusable OpenCode skills/agents/plans.
If the user asks to implement, proceed; do not stop at theory. If the user asks for a methodology or skill, create the docs/skill files directly.
Required Output Shape For Cycle Work
For any non-trivial cycle, produce or update a document/plan with these sections:
# Aladdin V2 Cycle: [Name]
## 老灯: Veteran Judgment
- Engineering risk:
- Architecture tradeoff:
- Things AI is likely to guess wrong:
- Quality line that must not be crossed:
## Field Truth
- Scenario:
- Pain:
- Evidence:
- Current workaround:
## Echo
- One-sentence problem:
- Demo script:
- Success metrics:
- Out of scope:
- SDD impact:
## Delta
- Biggest uncertainty:
- Prototype strategy:
- Iteration plan:
- Verification commands:
## Product
- Reusable abstraction:
- ADR/SDD updates:
- Harness updates:
- Postmortem/docs:
## Product Leverage
- What becomes easier next time:
- Remaining risks:
Phase Rules
Discovery
Goal: turn fuzzy intent into a concrete Aladdin V2 direction.
Actions:
- Read high-value docs before inventing concepts.
- Extract old-lamp veteran judgment from existing failures, docs, scripts, and architecture decisions.
- Identify which hidden experience must be made explicit.
- Define what the new-spirit/vibe-coding layer can accelerate.
- Recommend a concrete next artifact: methodology doc, cycle plan, SDD update, ADR, AGENTS guardrail, or skill.
Echo
Goal: define a valuable demo, not a large requirements document.
Checklist:
- Problem is grounded in field truth or production evidence.
- Demo is concrete and emotionally/operationally meaningful.
- Metrics are measurable.
- Scope is explicitly bounded.
- SDD impact is known, usually Ch4/Ch5/Ch6.
Avoid:
- Large speculative requirements.
- Feature lists without demo.
- Starting implementation before identifying the biggest uncertainty.
Delta
Goal: validate the biggest uncertainty quickly.
Rules:
- First prototype may hardcode, log, print, or use narrow assumptions.
- Record assumptions that must be removed in Product.
- Demo every 2-3 days or after each small iteration.
- If 5 iterations do not converge, question the approach.
- Do not create broad abstractions before the path is proven.
Verification:
- Prefer focused commands over whole-repo assumptions.
- For MRRC, remember hardware/audio tests can require devices and dependencies.
- Use repo-specific guardrails from
AGENTS.md before touching PTT/audio/FT8/deployment.
Product
Goal: turn learning into reusable leverage.
Required outputs:
- Reusable module/pattern or explicit decision not to abstract.
- ADR or architecture decision update for major choices.
- SDD updates for changed context, NFR, services, components, operations, or version history.
- Harness updates: commands, tests, plans,
AGENTS.md, postmortem, troubleshooting doc, or skill.
- Product leverage statement: what future work is now cheaper.
Do not accept “it works” as done if no knowledge was captured.
Review
Review against these questions:
- Did old-lamp veteran judgment define risks and boundaries?
- Is there field truth, or is this speculation?
- Did vibe coding accelerate work within constraints, or bypass constraints?
- Are NFRs measurable?
- Is there a harness for future agents?
- Did the work create Product leverage?
- Are SDD/ADR/AGENTS/docs updated where needed?
Findings should focus on missing guardrails, unverifiable claims, premature abstraction, skipped Product work, and repeated known risks.
Skill-System
When creating or updating OpenCode skills:
- Put project skills under
.opencode/skills/<name>/SKILL.md.
- Include frontmatter with
name and a concrete trigger-heavy description.
- Keep the skill actionable: phase detection, checklists, templates, references.
- Avoid vague motivational prose in skills; put long explanation in
docs/.
- Remind the user to restart opencode after skill/config changes.
MRRC-Specific References
Use these as source material in this repository:
docs/aldv2/Aladdin_V2_Methodology.md: top-level methodology.
docs/aldv2/Laodeng_Xinling_Methodology.md: core old-lamp/new-spirit model.
FDE.md: MRRC FDE retrospective.
docs/FDE_Practice_Guide.md: operational FDE guide.
SDD/: IBM TeamSD 14-chapter architecture spine.
DESIGN/MRRC_SDD.md: compact SDD overview.
docs/superpowers/specs/: agentic specs examples.
docs/superpowers/plans/: executable agentic plans examples.
AGENTS.md: repo-specific OpenCode guardrails.
docs/PTT_Audio_Postmortem_and_Best_Practices.md: example of memory harness.
Templates
Agentic Plan
# [Task] Implementation Plan
> For agentic workers: execute step-by-step. Do not skip verification.
Goal:
Architecture:
Scope:
Out of Scope:
Guardrails:
## Task 1: [smallest safe step]
- Files:
- Steps:
- Verification:
- Expected result:
## Task 2: [next step]
- Files:
- Steps:
- Verification:
- Expected result:
## Productization
- ADR/SDD:
- Harness:
- Docs:
- Product leverage:
ADR Prompt
## AD-[ID]: [Decision]
Problem:
Context:
Decision:
Alternatives:
Rationale:
Impact:
Verification:
Follow-up harness:
Postmortem Prompt
# [Incident/Problem] Postmortem
What happened:
User/system impact:
Root cause:
Why existing harness missed it:
Fix:
Guardrail added:
Verification:
Product leverage:
Final Response Guidance
When completing work under this skill, summarize:
- Which Aladdin V2 phase was applied.
- What artifacts were created or updated.
- What harness or guardrail was added.
- What future work is now easier.
- Any verification not run and why.