From Dev10x
Orchestrates adversarial evaluation of architectural options using domain-specific architect agents to produce ranked ADR drafts. For multi-approach design decisions and trade-off analysis.
npx claudepluginhub dev10x-guru/dev10x-claude --plugin Dev10xThis skill is limited to using the following tools:
Orchestrate adversarial discussion between architect agents to
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Orchestrate adversarial discussion between architect agents to evaluate architectural options and produce ranked ADR drafts.
The user provides:
| Agent | Domain |
|---|---|
architect-api | API design, schema, real-time |
architect-db | Database, schema, multi-tenant |
architect-domain | DDD, aggregates, bounded contexts |
architect-frontend | Components, state, SSR, testing |
architect-infra | Deployment, CI/CD, Docker, observability |
For each option, launch the assigned architect agent as a parallel subagent:
Agent(subagent_type="general-purpose",
model="opus", # Design tier — architecture evaluation
description="Advocate for Option N",
prompt="""You are advocating for: [Option N — description]
Context:
- [Topic description]
- [Relevant existing ADRs]
- [Current code state summary]
Your task:
1. Read all relevant source files in the codebase
2. Find evidence supporting your assigned option
3. Count concrete metrics (models, files, LOC, patterns)
4. Identify 3-5 arguments WITH file:line citations
5. Identify risks of NOT choosing your option
6. Be honest about weaknesses of your position
Output format:
## Arguments for [Option N]
1. [Argument with file:path evidence]
## Metrics
- [Concrete counts and measurements]
## Risks of Alternatives
- [What goes wrong if NOT chosen]
## Acknowledged Weaknesses
- [Honest downsides]""",
run_in_background=true)
After all advocates return, launch the adr-reviewer agent with
all collected arguments. The reviewer:
Reviewer agent skip (acceptable): When the main agent has
already accumulated sufficient codebase context from prior
research and advocate agents, it may perform the synthesis
inline without launching the adr-reviewer sub-agent. This
avoids an unnecessary round-trip while preserving the option
for complex evaluations where independent fact-checking adds
value.
Present the draft ADR to the user with:
REQUIRED: Call AskUserQuestion (do NOT use plain text).
Options:
/Dev10x:adr-evaluate
Topic: State management approach for the dashboard
ADR Number: 025
Options to evaluate:
1. Server-side state via load functions
2. Client-side reactive store
3. Hybrid with server-first, client cache
Agent assignments:
- architect-frontend (Option 1): SSR patterns
- architect-frontend (Option 2): reactivity patterns
- architect-api (Option 3): data loading