CAVRA Productization Report

CAVRA Productization Report

Executive summary

The repository has been transformed into CAVRA, Controlled Agentic Verification & Runtime Authority. CAVRA is positioned as a runtime governance and authority layer for AI coding agents, with pre-action enforcement, policy-as-code, evidence, PR attestation, MCP governance, Claude Code setup, and an interactive sandbox.

New repository URL: https://github.com/Huzefaaa2/cavra

Branch name: productize-cavra-pr

PR URL: https://github.com/Huzefaaa2/cavra/pull/1

History-preserving migration branch also exists at productize-cavra, but GitHub could not open a PR from it because the target repository main branch has unrelated history. The reviewable PR branch was created from origin/main with the same CAVRA final tree.

Product identity

Visible identity is CAVRA. Tagline: Before the agent acts, CAVRA decides. CAVRA is documented as a runtime authority layer for autonomous engineering actions across code, tools, cloud, CI/CD, and governed enterprise workflows.

Implemented and preserved features

Preserved: policy registry, YAML policy packs, File Guard, Command Guard, Git Guard, MCP policy concepts, session audit, PR attestation, webhook export, Typer CLI, tests, and regulated policy examples.

Added: CAVRA package path, cavra CLI, cavra-mcp-server, Claude Code initializer, richer decision response format, MCP tool checks, sandbox data model, FastAPI API contract, policy commands, flagship demo, schemas, protobuf contract, Docker assets, enterprise docs, and CAVRA policy packs.

Documentation and diagrams

Created or updated README, architecture, dual-plane architecture, threat model, product strategy, market positioning, competitive landscape, enterprise adoption, roadmap, partner strategy, monetization, demos, compliance mapping, control catalog, deployment, quickstart, CLI, API, integrations, policy docs, evidence format, approval workflows, MCP governance, agent identity, security model, air-gapped deployment, procurement readiness, Claude Code integration, sandbox docs, and enterprise readiness files.

Added production implementation governance docs: docs/production-roadmap.md, docs/implementation-plan.md, docs/user-stories.md, and docs/enterprise-challenges.md.

Added wiki-ready pages under docs/wiki/, including Home, White Paper, Production Roadmap, Implementation Plan, User Stories, Enterprise Challenges, Diagrams, and Phase Completion Log.

Added C4 and runtime diagrams under docs/diagrams/, including Mermaid diagram sources and SVG image assets. The C4 container diagram now has a more elaborate grouped architecture view and a user-friendly SVG asset.

Added GitHub repository readiness controls and documentation: protected main, Dependabot, CodeQL, PR template, CODEOWNERS, issue templates, VS Code recommendations, release documentation policy, and repository readiness guide.

Added high-quality user-facing diagram images for architecture, runtime flow, evidence hub, policy lifecycle, developer journey, OIDC/RBAC deployment, immutable evidence storage, and Go parity plus sandbox deployment.

Added CAVRA brand assets: SVG logo mark, horizontal logo, stacked logo, favicon, product thumbnail, GitHub social preview, and PNG exports for README, documentation, dashboard, and release surfaces.

Added transparent CAVRA engineering-agent methodology: declarative agent manifests, agent task issue template, label catalog, conservative GitHub Actions orchestrator scaffold, cavra-agentic-delivery policy pack, architecture documentation, wiki pages, and the user-facing agent orchestration diagram. The methodology explicitly requires bot identities and prohibits fake human developer identities.

Published the GitHub Wiki at https://github.com/Huzefaaa2/cavra/wiki with the white paper, roadmap, user stories, challenge mapping, C4 diagram pages, SVG diagram assets, Phase 2 policy engine hardening page, Phase 3 evidence hub page, evidence key management page, evidence metadata migrations page, GitHub repository readiness page, release documentation policy, transparent agent methodology, and agent orchestration architecture. Wiki commits: 784a847, 9b24196, e584f14, 92a14ab, 983dc54, 66cd075, 4683941, c26f7af, and 235c30c.

Claude Code and MCP status

cavra init claude-code creates .mcp.json, .cavra/policy.yaml, and .cavra/session/. Documented one-line install path: claude mcp add cavra -- cavra-mcp-server.

cavra-mcp-server exposes CAVRA governance tools and validates file, command, Git, MCP, attestation, evidence, policy, and session workflows.

Interactive sandbox status

apps/sandbox-ui/ implements the Before the Agent Acts sandbox and first evidence console slice with a security-console layout, persona and policy-mode controls, action stream, decision stream, evidence viewer, evidence metadata search, PR attestation verification, operational readiness status, and Claude Code install CTA. It uses the same CAVRA decision outcomes as the runtime and is runnable locally with Python static serving or Docker Compose. .github/workflows/deploy-sandbox.yml now publishes the static sandbox to GitHub Pages from main after JavaScript validation.

Go enforcement-plane status

go/cavra-runtime/ now contains the first Go enforcement-plane parity scaffold. It includes a Go module, runtime evaluator, command-line entrypoint, compiled-policy JSON loader, generated enforcement contracts, typed release-governance evidence contract payloads, daemon and CI runner examples, Unix-socket daemon transport, reusable daemon client helper, CLI --daemon mode, daemon lifecycle start/status/stop, request/response evidence hooks, shared parity cases, and Go unit tests for critical file, command, Git, MCP, and release governance record decisions plus typed release-governance contract payloads. The Python integration layer now includes an explicitly opt-in Go backend path with disabled, shadow, enforce, and promoted modes, CLI/API readiness reports, production-readiness evidence, audited fallback to Python on runtime failure or parity mismatch, deployment readiness checks for CI runner bundles, workstation channels, and updater policy, promotion readiness checks requiring approved audited parity evidence, rollback readiness checks requiring an approved path back to Python-only mode, rollback rehearsal evidence checks with console visibility, and fresh rollback drill history, active rollback drill scheduling, and stale notification delivery before Go is selected as an optional backend. Python remains authoritative by default; the scaffold is intentionally bounded while acknowledgement escalation and binary release operations continue to mature.

Validation commands run

• python3 -m pytest -q -> 17 passed.
• PYTHONPATH=src python3 -m cavra.cli --help -> passed.
• PYTHONPATH=src python3 -m cavra.cli policy test -> passed.
• PYTHONPATH=src python3 -m cavra.cli evaluate read_file .env --json -> passed.
• PYTHONPATH=src python3 -m cavra.mcp_server --list-tools -> passed.
• PYTHONPATH=src python3 -m cavra.mcp_server --check-command 'terraform apply -auto-approve' -> passed.
• PYTHONPATH=src python3 -m cavra.cli demo before-the-agent-acts --output /tmp/cavra-demo -> passed.
• PYTHONPATH=/Users/huzefahusain/Projects/cavra/src python3 -m cavra.cli init claude-code from /tmp -> passed.
• FastAPI dependency was installed locally with python3 -m pip install --user 'fastapi>=0.110' 'uvicorn>=0.27'.
• PYTHONPATH=src python3 - <<'PY' ... create_app() -> passed after dependency install.
• docker build -t cavra:local . -> passed.
• docker run --rm cavra:local --help -> passed.
• docker run --rm --entrypoint cavra-mcp-server cavra:local --list-tools -> passed.
• docker run --rm cavra:local policy test -> passed after fixing installed-package policy discovery.
• docker compose up -d --build -> passed.
• curl -fsS http://127.0.0.1:8000/health -> passed.
• curl -fsS http://127.0.0.1:8000/version -> passed.
• curl -I -fsS http://127.0.0.1:5173 -> passed.
• docker compose down -> passed.
• for p in policies/*; do PYTHONPATH=src python3 -m cavra.cli policy validate "$p"; done -> passed.
• PYTHONPATH=src python3 -m cavra.cli policy compile --policy-pack cavra-ai-agent-baseline -> passed.
• PYTHONPATH=src python3 -m cavra.cli policy diff policies/cavra-ai-agent-baseline policies/cavra-banking-baseline -> passed.
• PYTHONPATH=src python3 -m cavra.cli policy sign ... --key secret and policy verify ... --key secret -> passed.
• docker build -t cavra:local . with packaged schemas -> passed.
• docker run --rm cavra:local policy validate policies/cavra-ai-agent-baseline -> passed.
• docker run --rm cavra:local policy test -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence bundle --output /tmp/cavra-evidence --key secret -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence verify /tmp/cavra-evidence --key secret -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence siem-event /tmp/cavra-evidence -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence export-siem /tmp/cavra-phase3-bundle --output /tmp/cavra-phase3-siem -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence storage-plan /tmp/cavra-phase3-bundle --output /tmp/cavra-phase3-storage --retention-days 365 -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence generate-keypair ... -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence verify /tmp/cavra-phase3-signed --public-key ... --minimum-retention-days 365 -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence retention-policy /tmp/cavra-phase3-signed --output /tmp/cavra-phase3-retention --retention-days 365 -> passed.
• PYTHONPATH=src python3 -m cavra.cli evidence index /tmp/cavra-phase3-signed --store /tmp/cavra-evidence-metadata.json -> passed.
• node --check apps/sandbox-ui/sandbox.js -> passed.
• python3 -m pytest -q -> 36 passed.
• HTML parser smoke check for apps/sandbox-ui/index.html -> passed.
• curl -I -fsS http://127.0.0.1:5173 for the hosted evidence console -> passed.
• docker run --rm cavra:local evidence bundle --output /tmp/cavra-evidence --key secret -> passed.
• docker run --rm -v cavra-evidence-check:/tmp/evidence cavra:local evidence verify /tmp/evidence --key secret -> passed.

GitHub repository readiness

Repository Huzefaaa2/cavra is accessible from Codex through GitHub CLI with admin permission and from local developer tools through https://github.com/Huzefaaa2/cavra.git.

main is protected with one required approving review, stale review dismissal, required conversation resolution, force-push protection, and branch deletion protection.

Enabled repository security and quality features include secret scanning, secret scanning push protection, Dependabot alerts, Dependabot security updates, Issues, Wiki, auto-merge, update branch, and delete branch on merge. Repository config files now include Dependabot, CodeQL, PR template, CODEOWNERS, issue templates, and VS Code recommendations.

Brand validation

Brand search for old visible product names returned no matches after cleanup. Remaining generic CAVRA policy IDs are expected.

Roadmap backlog

Phase 2, Policy Engine Hardening, is now implemented. Added strict JSON Schema policy validation, policy inheritance resolver, normalized compile output, semantic policy diff, policy signature metadata, signature verification, and tests for bundled policy validation, inheritance, diff, and tamper detection.

Phase 3, Evidence Hub and Attestation, is near complete. Added evidence bundle manifest generation, checksum verification, optional HMAC manifest signature, Ed25519 manifest signatures, key IDs, trust-root verification, rotation guidance, retention policy artifacts, PR attestation output and verification reports, compliance mapping output, SIEM event output, provider-specific SIEM export payloads, immutable storage reference plans, evidence metadata indexing, SQLite search with filters and pagination, hosted console views, initial SQLite migration, API metadata persistence, CLI evidence commands, and evidence tests.

Phase 4, Approval Router, is complete for the current production-readiness slice. Added JSON and SQLite approval persistence, API and CLI approval lifecycle, break-glass evidence, repository-specific routing, signed OIDC/JWKS validation, repository RBAC policy files, live approval provider delivery, console approval actions, console break-glass creation, and approval audit detail views.

Next recommended implementation work: implement private hosted policy registry readiness evidence in cavra-enterprise for the post-onboarding SaaS operating readiness slice. The latest completed commercialization batches delivered public trial-to-pilot intake, public licensing hardening, public SaaS Control Plane contracts, public tenant onboarding, public entitlement status, public hosted policy registry readiness, public tenant audit-store operating contracts, public billing/subscription boundary documentation, private trial package readiness, private customer pilot handoff, private tenant onboarding readiness, private entitlement/license handoff, private paid-pilot promotion, private customer rollout closeout, and public-safe documentation syncs while preserving open-core boundaries.

Later roadmap backlog: full persistent API storage for every domain, expanded parity tests, production Enterprise connectors, trial distribution, and SaaS control-plane services.

Market, partner, and commercialization summary

CAVRA is positioned as the enterprise runtime authority layer for AI-assisted engineering. Partner strategy is to show AI coding vendors and enterprise workflow platforms that CAVRA increases adoption by solving runtime governance, evidence, audit, and approval. Commercialization paths include community, team, enterprise, compliance packs, MCP Trust Registry, OEM, certification, support, hosted SaaS, self-hosted, and air-gapped enterprise.