By lebobo88
Hydra — Enterprise Agent Mesh supervisor (routes work across executive / engineering / creative / legal / healthcare / sales / research / customer-support squads).
Scaffold a new squad pack: creates squads/<slug>/squad.yaml from a template + registers keyword fingerprints.
Approve a paused HITL gate and resume the supervisor graph.
Show rolling budget consumption across workflows; optionally set a new cap.
Multi-squad campaign template: routes executive + creative + engineering in one workflow with explicit dependency wiring.
Replay a past workflow from checkpoint (LangGraph checkpoint → optionally pinned to original model/CLI versions).
Budget tripwire. Re-checks workflow cost against the configured budget after every squad return; downgrades model tier at 80%, blocks at 100% pending HITL. Drives on top of the CFO agent from the executive squad.
Owns the human-in-the-loop interrupt boundary. Renders pending HITL_REQUEST envelopes to the operator, captures the decision, and resumes the supervisor graph.
Decomposes a routed goal into typed cross-squad envelopes. When the executive squad is in play, defers to ExecutiveSuite's boardroom for strategic framing first.
Routes work to the correct squad pack. Uses keyword + industry tags first, then falls back to LLM classification when ambiguous. Used by hydra-supervisor and as the engine behind /hydra:run intake.
Top-level Hydra supervisor. Owns the workflow lifecycle: intake → planning → approval → dispatch → executing → synthesis → postcheck. Drives the LangGraph state machine in hydra_core/supervisor.py.
Per-workflow budget enforcement: tripwire at 80% (downgrade model tier), HITL at 100%. Telemetry pattern for cost-per-task tracking.
How to construct, validate, and route Hydra's typed cross-squad message envelopes. Use whenever an agent emits or consumes a CSuiteDecisionPacket, PRD, ArchRFC, DevTask, CreativeBrief, ShotList, AssetJob, HITLRequest, DecisionRecord, or Handoff.
When and how Hydra pauses for human input. Defines HITL gate placement, render format, resume contracts, expiry behavior, and override semantics.
Memory fabric usage — ephemeral / episodic / semantic. Read this before reading from or writing to Hydra memory.
How Hydra discovers, resolves, and routes to squad packs. Read this before adding/removing/modifying a squad.
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Many heads. One Spirit. The covenant holds.
A LangGraph supervisor that routes work across specialized AI agent squads — executive strategy, software engineering, creative production — under constitutional governance with human-in-the-loop gates, budget enforcement, and cross-model judging.
Hydra is a Claude Code plugin + Python LangGraph supervisor that routes work across a constellation of heterogeneous AI agent squads — Executive (C-suite strategy), Forge (engineering), Garland (creative), plus marketing operations and a roster of governance-ready stubs. It is Pentecost, not Legion: many distinct agents under one cryptographically-anchored covenant (CONSTITUTION.md), routed via typed envelopes, audited by AgentSmith, and tagged with the eight-cell I Ching vocabulary backed by the optional TheEights memory daemon as a shared cross-project substrate.
Hydra does not re-implement the squads' work. It sits above them: classifying goals, decomposing them into typed cross-squad envelopes, dispatching subgraphs in parallel, synthesizing the result into a single decision record, and enforcing the constitutional rule of faith on every step.
The Pentecost flower sigil — central immortal head, three crown petals, eight-cell substrate. See docs/constellation/ for the full layered presentation, the reveal.js cinematic deck, and executive memos.
Hydra is one node in a mesh of nine sibling AI systems bound together by a tenth layer, AgentMesh. Each system is independently shippable; together they form a unified AI operating system. Several of the siblings below — ExecutiveSuite, RLM-Creative, MarketBliss, Senate, and Xenia — are also Hydra squads (see Squad topology); they appear here as ecosystem members in their own right.
graph TD
subgraph MESH["🕸️ AgentMesh — binding control plane (the tenth layer)"]
direction TB
HYDRA["🐍 Hydra<br/><em>Supervisor · Router · Synthesizer</em>"]
ES["👔 ExecutiveSuite<br/><em>20 C-suite agents · 4 orchestrators</em>"]
PP["⚙️ pair-programmer<br/><em>Best-of-N harness · 39 sub-agents</em>"]
RLM["🎨 RLM-Creative<br/><em>8 Garland Heads · Helios sub-crew</em>"]
MB["📣 MarketBliss<br/><em>5 marketing squads</em>"]
SENATE["⚖️ Senate<br/><em>12-jurist Curia · legal-compliance</em>"]
XENIA["🔥 Xenia<br/><em>11-agent Hearth · customer-support</em>"]
SMITH["🛡️ AgentSmith<br/><em>Meta-governance · N1–N10 invariants</em>"]
EIGHTS["∞ TheEights<br/><em>Root of trust · memory · self-evolution</em>"]
end
HYDRA -->|"CSuiteDecisionPacket"| ES
HYDRA -->|"PRD · DevTask"| PP
HYDRA -->|"CreativeBrief · ShotList"| RLM
HYDRA -->|"CreativeBrief"| MB
HYDRA -->|"LegalReview"| SENATE
HYDRA -->|"SupportTask"| XENIA
ES -->|"DecisionRecord"| HYDRA
PP -->|"DecisionRecord"| HYDRA
RLM -->|"DecisionRecord"| HYDRA
SENATE -->|"DecisionRecord"| HYDRA
XENIA -->|"DecisionRecord"| HYDRA
SMITH -.->|"validates"| HYDRA
SMITH -.->|"validates"| ES
SMITH -.->|"validates"| PP
SMITH -.->|"validates"| RLM
EIGHTS -.->|"root of trust · MemoryRef"| SMITH
EIGHTS -.->|"MemoryRef handles"| HYDRA
EIGHTS -.->|"memory"| ES
EIGHTS -.->|"memory"| PP
EIGHTS -.->|"memory"| RLM
MESH -.->|"enroll · route · observe (no governance of its own)"| EIGHTS
RLM-Gaming — the Arcade crown. A senior multi-agent game studio (vision, design, production, runtime-AI/netcode direction, art/audio direction, QA/balance, live-ops, ratings/cert, fair-play security). Orchestrates + delegates: code -> engineering (pair-programmer), assets -> garland (RLM-Creative).
npx claudepluginhub lebobo88/hydra --plugin hydraMulti-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Curate auto-memory, promote learnings to CLAUDE.md and rules, extract proven patterns into reusable skills.
Memory compression system for Claude Code - persist context across sessions
Standalone image generation plugin using Nano Banana MCP server. Generates and edits images, icons, diagrams, patterns, and visual assets via Gemini image models. No Gemini CLI dependency required.