The framework
A collaboration framework for agentic workflows with a focus on governance, research, and execution. Multi-project federation, behavioral agent definitions, a 9-backend dispatch pipeline, semantic knowledge indexing, and an initiative lifecycle with MCP-exposed governance — infrastructure that lives in your codebase and versions with git.
Seven pillars
Each pillar is a working system with real data behind it — not a roadmap item.
Behavioral Agent System
30 agent roles defined through behavioral constraints — disposition, quality bar, fail triggers, domain scope. Not identity claims. Research-validated against Anthropic and academic findings. 8 task types route work to eligible roles automatically.
Governance Engine
72 initiatives tracked from proposal through integration with directoturtle directory structure. Dependencies, genealogy, and cross-initiative intelligence flows. External agents can propose and approve work via 7 MCP governance tools with policy-gated lifecycle transitions.
Execution Pipeline
9 dispatch backends — 5 CLI, 3 API, 1 remote gateway. Task-type routing with overnight restrictions. Authority leases with monotonic fencing tokens for multi-agent coordination. Real-time agent narrative streaming for mission visibility.
Studio Application
18 routes, 118 React components. Multi-project federation — one Studio instance governs multiple codebases. Task boards, initiative trees, workforce management, workflow canvas, research dashboards, and mission control. Zero-downtime blue-green deploys.
Executable Conventions
17 skills with full step-by-step protocols — recursive research, integration review, stress-testing, design, task planning. 9 rules auto-loaded by glob pattern. Self-documenting system with provenance tracking and staleness detection across every maintained document.
Config-as-Code
sherpa.json as canonical config with environment variable interpolation. Three-directory model: .sherpa/ (tool state), .claude/ (agent config), docs/ (human prose). 9 config domains including dispatch routing, knowledge backends, governance policy, and vocabulary. Type-safe with JSON schema and plugin architecture.
Convention Sync CLI
sherpa init scaffolds governance structure into any project. sherpa sync keeps conventions current across codebases. Provenance frontmatter tracks authorship, review state, and verification date for every maintained document.
Cross-cutting infrastructure
Capabilities that span multiple pillars and make the framework more than conventions in a filesystem.
Multi-Project Federation
Project registry with cross-project data merging. Aggregate views across task boards, research feeds, and initiative trees. One governance substrate, many codebases.
Semantic Knowledge Engine
Pluggable backend architecture with zero-dependency algorithmic default. TF-IDF indexing, extractive summaries, agglomerative clustering. Queryable search over governance artifacts via MCP.
SQLite State Layer
Three embedded databases — coordination (authority leases, sessions), events (audit trail), knowledge (semantic index). Markdown stays canonical; SQLite provides derived queryable state. WAL mode for concurrent read safety.
MCP Coordination
HTTP Streamable transport replacing stdio. Multi-client session manager with authority enforcement. Fencing tokens prevent concurrent mutation. TTL reaper for expired leases. Governance scoped to autonomous agents — never constrains human+AI sessions.
Packages
@sherpa/studio-coreDomain logic, config schema, lifecycle engine, dispatch routing, knowledge backend interface@sherpa/studio-ui118 React components for governance visualization@sherpa/studio-mcpMCP server with authority leases, session management, and initiative governance tools@sherpa/studio-cliConvention scaffolding, sync, and provenance trackingBuilt by a practitioner
This framework emerged from building agentic workflows and discovering that the governance infrastructure didn't exist. Every convention, every skill, every agent role was built because it was needed. The system governs its own development — the same initiative lifecycle, the same dispatch pipeline, the same behavioral constraints you'd adopt.