dmjcustomizations
divyamohan1993/dmjcustomizations
Parallel-first agentic engineering skills for Claude Code: team-based workflows, adversarial verification, security and performance gates from line 1. A modern rebuild of superpowers.
Instalação
npx convez-ai divyamohan1993/dmjcustomizations1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
dmjcustomizations
Parallel-first engineering skills for Claude Code. A ground-up rebuild of superpowers for the agent-team era: terse, date-agnostic, adversarially verified, secure from line 1.
Why
Superpowers encoded the right discipline on old assumptions: one agent, one chat thread, expensive code, a human watching every step. All four flipped. dmjcustomizations keeps the gates and rebuilds the mechanics.
- Parallel between gates, serial at gates. Agent teams fan out; the user approves at a few hard checkpoints. Nothing else blocks.
- Evidence over claims. Fresh-context teammates adversarially verify specs, code, and every "done". Self-review is not review.
- Demonstrate, don't describe. Competing approaches become disposable worktree spikes with benchmarks, not paragraphs.
- Dynamic by design. No hardcoded models, versions, or dates. Skills probe for the strongest model and newest stable tooling at invocation time.
- Security and performance from line 1. Threat models at design time, quantum-safe crypto defaults, O(1)-first thinking, budgets the skills require projects to enforce in CI.
- Terse. Every skill is context-budgeted: under 500 words, and the always-loaded meta-skill under 300.
Skills
| Skill | One-liner |
|---|---|
| using-dmj | Meta-skill, injected each session: how skills are found, routed, and prioritized |
| brainstorming | Idea to approved spec: parallel context sweep, batched questions, evidence-based options |
| writing-plans | Spec to implementation plan with per-task dependencies, parallelizability, acceptance criteria |
| executing-plans | Team executes a plan: tasks claimed concurrently, worktree isolation, review gates |
| team-driven-development | Plan execution in the current session with implementer and reviewer teammates |
| dispatching-parallel-teams | Any 2+ independent tasks: fan out a team, coordinate, synthesize |
| test-driven-development | Iron Law TDD plus an extreme edge-case taxonomy (adversarial, concurrency, boundaries) |
| systematic-debugging | Root cause before fixes; parallel hypothesis investigation |
| verification-before-completion | Evidence before claims; independent fresh-context verification |
| requesting-code-review | Multi-lens parallel review panel: correctness, security, performance, simplicity |
| receiving-code-review | Rigor over performative agreement; verify before implementing feedback |
| using-git-worktrees | Isolation for parallel work and disposable spikes |
| finishing-a-development-branch | Verified finish: merge, PR, or cleanup, with team and worktree teardown |
| writing-skills | TDD for documentation: baseline, write, close loopholes, team-tested |
| defending-in-depth | Security from line 1: threat model, OWASP, zero trust, quantum-safe defaults |
| enforcing-performance-budgets | O(1)-first, measured budgets, cache-first, regressions block merges |
| researching-deeply | Parallel research with adversarial source verification and dated citations |
| exploring-codebases | Five-lens parallel codebase mapping with an anti-redundancy gate: reuse before rebuild |
| explore | Parallel slice-by-slice tracing of how a codebase really works, explained in chat, no artifacts |
| harnessing-claude | Capability routing: strongest native Claude feature for every job, never hand-rolled substitutes |
| crafting-experiences | Experience supremacy: Jobs test, first-second hook, cinematic with purpose, zero user burden |
| landing-sessions | Session teardown: learnings to memory, state to git, resources to zero, threads surfaced |
| selling-the-vision | Launch persuasion: one-second hook, keynote arc, one message, one CTA, honest claims |
Install
The plugin installs as dmj, so skills fire as dmj:brainstorming, dmj:test-driven-development, etc.
From GitHub:
/plugin marketplace add divyamohan1993/dmjcustomizations
/plugin install dmj@dmj
Or from a local clone:
/plugin marketplace add D:\dmjcustomizations
/plugin install dmj@dmj
Then disable superpowers so the two rule systems do not compete:
/plugin uninstall superpowers
Provenance
Forked from obra/superpowers 5.1.0 by Jesse Vincent (MIT). Rebuilt 2026-06-10 by Divya Mohan. MIT licensed.