EstreGenesis
A starter kit for building software with AI coding assistants.
A reusable seed of agentic-coding patterns for Claude Code, Codex, Cursor, etc.
A public seed of agentic-coding patterns for Claude Code.
Install — Register the marketplace, then install plugins
Install — Marketplace registration + plugin install
Installation — Marketplace registration and per-plugin admission
EstreGenesis ships five optional modules as Claude Code plugins. Two steps:
EstreGenesis ships as five independent Claude Code plugins published through a single marketplace. Two-step install:
EstreGenesis is distributed as five orthogonal Claude Code plugins admitted via a single marketplace manifest. Installation is a two-stage operation:
(A) register the marketplace once, then (B) install the plugins you want from the cards below.
(A) one-time marketplace registration (idempotent — re-running is safe), then (B) per-plugin install commands. Plugins are independent — install any subset.
Stage (A) is idempotent marketplace registration; stage (B) admits individual plugins via fully-qualified references. Zero inter-plugin dependencies — freely chosen admission set.
Run this once in Claude Code to tell it where the EstreGenesis plugins live.
Run once in Claude Code to register the marketplace source. The shorthand resolves to the repo's marketplace manifest.
Persists a marketplace source descriptor binding the `estregenesis-plugins` namespace to the manifest. Idempotent re-issuance is well-defined.
/plugin marketplace add SoliEstre/EstreGenesisPick any of the five cards below and click Copy on the install command. You can install just one, all five, or any mix.
Copy any install command from the five cards below. Each plugin is independent — any subset is a valid install set.
Each install invocation admits an individual plugin from the registered namespace; admission graph is flat, subset unconstrained.
Available plugins
Superscalar
Rules for running many AI agents in parallel without losing consistency.
Execution-scheduling discipline for parallel sub-agent dispatch: issue_width bound + worktree isolation + in-order retire + consistency gate.
Out-of-order speculative dispatch discipline: issue_width + worktree-isolated reorder buffer + admission gate + opt-in speculation. Entry 06 A/B: +118% grounded refs, −40% speculative claims.
/plugin install superscalar@estregenesis-plugins
No extra libraries needed
Hyperbrief
When an AI asks you to decide, it gives you a one-page summary in a fixed shape (what · why · options · recommendation).
Decision-delegation gating: 4-score trigger rubric + JSON IR + deterministic Node renderer → MD + interactive HTML. 9-section structure, 3-axis × 5-level tone profile, byte-identical output.
Decision-delegation gating discipline. 4-score escalation admits to schema-strict IR; deterministic renderer yields byte-identical output. Integrated with Constellation §13.16.9 + Superscalar §3.1.
/plugin install hyperbrief@estregenesis-plugins
Needs: ajv library (one of the standard schema-checking libraries for Node)
Greatpractice
Small obligations you keep forgetting (update the docs · check before sending · sync paired files) get turned into auto-enforced rules.
Memory-triggered practice codification. feedback_*.md entries pass a 5-axis maturation gate, graduate into a 3-tier hierarchy, enforced via 7-event lifecycle hooks.
Memory-triggered codification distilled from 9-axis synthesis. 5-stage pipeline + 3-backbone architecture. phronesis_boundary protects un-codifiable work.
/plugin install greatpractice@estregenesis-plugins
No extra libraries needed · 🐶 Early dogfooding
Ultrasafe
Before each release, several AI helpers act like attackers and look for weak spots — at the same time, from different angles.
Pre-release simulated pentest. 8-agent parallel red-team fan-out + GTA/DSP cross-cutting. ≥3 iteration loop + 4-condition AND clean-signal gate. 3-layer synthesis (OSCAL + Hyperbrief IR + Greatpractice).
Pre-release attestation distilled from 17-axis synthesis. 3-backbone architecture. Advisory v0.1.x → blocking v0.2.x. Schneier + Herley + trust-erosion grounding.
/plugin install ultrasafe@estregenesis-plugins
No extra libraries needed · 🐶 Early dogfooding
Constellation
Live workspace where multiple AI agents talk to each other in real time.
Live multi-agent orchestration: WebSocket + A2A + dashboard + MCP. 24h dogfood: 5 e2e PRs, 0 violations.
Live multi-agent orchestration. §13.x codifies attachment transport + relay + §13.9.1/.2 + §13.16.9 Hyperbrief intents. Baseline: 24h dogfood + bundle 007.
/plugin install constellation@estregenesis-plugins
Needs: ws library (one of the standard WebSocket libraries for Node)
All five plugins work the same way in both Claude Code terminal (CLI) and Claude Code VS Code extension. The next section shows where to start in each environment.
All five plugins behave identically across Claude Code CLI and VS Code extension — same slash commands, skills, hooks. Next section covers entry-point differences.
Plugin runtime semantics are host-uniform across Claude Code CLI and VS Code extension. Subsequent section enumerates entry-point differences only.
What's in the seed
Think of it as a battle-tested checklist for working with AI coding tools.
A portable bootstrap for AI-assisted development. Six seed tiers cover Phase 0 setup, AGENTS.md scaffolding, coordination patterns, references to two optional modules.
EstreGenesis distills agentic-coding patterns refined across two real AI-native projects into a portable bootstrap.
Modules
Project Seed
A one-page starter that tells your AI assistant how to behave on your project.
Six-tier bootstrap. Phase 0 setup + AGENTS.md scaffold + coordination patterns.
Six-tier bootstrap. Phase 0 setup → AGENTS.md scaffold → any AI coding service.
README →Superscalar
Parallel AI work that stays consistent. Measured: 2.2× more grounded · 40% less guesswork.
Discipline for parallel sub-agent dispatch. issue_width + worktree isolation + retire + consistency gate.
Execution-scheduling discipline for parallel sub-agent dispatch.
Read the A/B report →Hyperbrief
A structured brief on every AI-to-user decision request: what / how / why / consequences / decision tree / recommendation. One-click re-render in the reader's plain language (button auto-localizes to user's language).
Decision-delegation gating discipline. 4-score escalation + JSON IR + deterministic renderer + 3-axis × 5-level tone profile. Byte-identical output verified. Integrated with Constellation + Superscalar.
Decision-delegation gating discipline distilled from a 6-axis deep-research synthesis. Schema-enforced JSON IR + deterministic renderer. Two self-applied dogfood entries + adopter bundle 008. Integrated with Constellation §13.16.9 + Superscalar §3.1.
Read the discipline →Greatpractice
Memory notes are the trigger; recurring patterns get promoted into auto-enforced practices. Backed by 9-axis cross-domain research.
Memory feedback → 5-axis maturation gate → ratified entry in 3-tier tree, enforced by 7-event lifecycle hooks. phronesis_boundary protects un-codifiable work.
Memory-triggered codification discipline distilled from a 9-axis cross-domain deep-research synthesis. 5-stage pipeline + 3-backbone architecture. Frontmatter governance SSoT. Normative grounding via Distributed Cognition + Organizational Anthropology.
Read the discipline →Ultrasafe
Several AI helpers find weak spots before release, fix-and-retry at least three rounds until clean.
Pre-release simulated penetration testing. 8-agent fan-out + ≥3 iteration AND-gate + 3-layer synthesis. Advisory→blocking.
Pre-release security attestation distilled from 17-axis cross-domain synthesis. 3-backbone architecture. Tier A patches absorbed. Schneier asymmetry grounding.
Read the discipline →Constellation
Multiple AI agents talking in real time + shared dashboard + safe handoffs. Five real PRs in 24h, zero mistakes.
Live multi-agent orchestration: WebSocket + A2A + dashboard + MCP. 24h dogfood: 5 e2e PRs, 0 violations. Hyperbrief A2A integration shipped at v2.3.20.
Live multi-agent orchestration: WebSocket + A2A + dashboard + MCP. §13.9.1/.2 + §13.16.9 Hyperbrief intents + bundle 007 absorbed.
Read the protocol →How you'll use this
How to use it — per environment
Invocation surfaces by host environment
Claude Code runs in two environments: the terminal (CLI) and VS Code. All EstreGenesis plugins work the same way in both. The entry point is what changes.
Claude Code ships in two host environments — terminal CLI and VS Code extension. Plugin behaviour is identical; only entry-point + chat-surface affordances differ.
Claude Code exposes two host surfaces sharing an identical plugin runtime; they differ exclusively in session bootstrap and chat-affordance rendering.
Claude Code CLI
Run Claude Code directly from the terminal.
Direct-from-terminal session. Best fit for headless workflows, CI integration, tmux-based multi-agent setups.
Terminal-resident host with stdin/stdout REPL and full shell integration. Preferred for scripted invocation and shell-managed orchestration.
- Start Claude Code in the terminal:
claudeorclaude code - Register the marketplace:
/plugin marketplace add SoliEstre/EstreGenesis - Install your chosen plugin:
/plugin install constellation@estregenesis-plugins(or any other module name) - Use slash commands, skills, and hooks in your session.
- Launch a CLI session:
claude(orclaude code) from your project root. - Register the marketplace (one-time, idempotent):
/plugin marketplace add SoliEstre/EstreGenesis - Install per-plugin:
/plugin install <name>@estregenesis-plugins— any subset of the five modules. - Slash commands, skills, and hooks are now active — plugin hooks fire on lifecycle events (SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop, PostCompact).
- Instantiate a CLI host process:
claudeorclaude code— the process inherits the invoking shell's cwd as the workspace root. - Persist the marketplace source descriptor (idempotent across re-issuance):
/plugin marketplace add SoliEstre/EstreGenesis - Admit individual plugins via fully-qualified namespace reference:
/plugin install <name>@estregenesis-plugins— admission graph is flat (zero cross-plugin dependency edges). - The 7-event lifecycle drives slash dispatch, skill registry, and hook execution; semantics are host-uniform with the VS Code surface.
Claude Code VS Code Extension
Use Claude Code as a panel inside VS Code.
VS Code extension surface — sidebar chat panel + integrated editor diff view. Best fit for IDE-resident workflows where the agent operates alongside open editor buffers.
VS Code extension host backed by the LSP-adjacent extension API. Plugin runtime semantics are bit-identical to the CLI host; only chat-surface rendering and session bootstrap differ.
- Install the Claude Code extension: search "Claude Code" in the VS Code extension marketplace (or visit https://claude.com/code)
- Open Claude Code: use the sidebar panel or press
Ctrl+Shift+P(Cmd+Shift+P on Mac) and search for "Claude Code: Open Chat" - Register the marketplace: type
/plugin marketplace add SoliEstre/EstreGenesisin the chat window - Install plugins: use
/plugin install <name>@estregenesis-plugins, then access modules from the sidebar.
- Install the Claude Code extension from the VS Code marketplace (search "Claude Code") or https://claude.com/code.
- Open the chat surface: sidebar panel or
Ctrl+Shift+P(Cmd+Shift+Pon Mac) → "Claude Code: Open Chat". - Register marketplace in the chat input:
/plugin marketplace add SoliEstre/EstreGenesis(one-time, idempotent). - Install per-plugin via
/plugin install <name>@estregenesis-plugins; skills and slash commands surface in the sidebar and accept the same lifecycle hooks as the CLI host.
- Install the Claude Code VS Code extension from the official VS Code extension marketplace (canonical search query: "Claude Code") or via the upstream distribution at https://claude.com/code.
- Invoke the chat-surface command: activate the sidebar panel or issue the command palette query
Ctrl+Shift+P/Cmd+Shift+P→ "Claude Code: Open Chat". - Persist the marketplace source descriptor through the chat input:
/plugin marketplace add SoliEstre/EstreGenesis— semantically identical to the CLI surface; idempotent re-issuance is well-defined. - Admit individual plugins via
/plugin install <name>@estregenesis-plugins; the resulting slash-command dispatch, skill registry, and 7-event lifecycle hook execution are bit-identical to the terminal host surface, with rendering routed through the VS Code chat panel.
Either way, the same EstreGenesis plugins work with the same behavior in both environments.
Either entry path lands on the same plugin runtime — identical slash-command dispatch, skill registry, and lifecycle hook execution across CLI and VS Code.
Both invocation surfaces converge onto a bit-identical plugin runtime; rendering substrate is the sole observable axis of variation.
What we made in the first 24 hours
Initial 24h dogfood — what shipped
Initial 24-hour dogfood — release artefacts
First day after public release, before the modules below were folded in. Numbers have grown since — see below.
Historical snapshot from the first 24h (around v2.5.15). Predates the Hyperbrief module ship + adopter bundles absorbed since.
Historical metrics from the initial 24h window (≈ v2.5.15). Antedates Hyperbrief ship + adopter bundles.
Since then
Cumulative since the 24h baseline
Cumulative state since the initial baseline
Pace settled but didn't stop. Third optional module (Hyperbrief) joined, two adopter reports folded back in.
Cumulative tags now at v2.5.36. Hyperbrief module shipped v2.5.23 + Constellation/Superscalar wiring v2.5.27–29 + 4 adopter-bundle absorption cuts (v2.5.31 / 33 / 34 / 35).
Cumulative tag count ≈ v2.5.36. Post-baseline = Hyperbrief module + cross-module wiring + 3 adopter-bundle absorption cuts.
The headline finding
Running AI agents in parallel doesn't automatically give good results. Speed without discipline ≠ quality.
Parallelism is not sufficient; orchestration discipline produces consistency. Wall-clock cost is phase serialisation, not parallel inefficiency.
Parallelism is not sufficient; orchestration discipline produces consistency.