Add files via upload

updated roadmap

Signed-off-by: Marcus H <148545173+mackeh@users.noreply.github.com>

Author: mackeh

aegisclaw-roadmap.md

@@ -0,0 +1,155 @@
+# AegisClaw Roadmap
+
+> Last updated: February 2026
+
+---
+
+## Completed Phases
+
+### ✅ v0.1.x — Foundations (Complete)
+
+- Go-based CLI (`aegisclaw init`, `secrets`, `sandbox`, `logs`)
+- Policy engine with granular scopes (`files.read`, `shell.exec`, `net.outbound`)
+- TUI-based human-in-the-loop approval for high-risk actions
+- Hardened Docker sandbox (non-root, read-only rootfs, dropped capabilities, seccomp)
+- `age`-based secret encryption for API keys
+- Tamper-evident, hash-chained audit logging
+- OpenClaw adapter for agent runtime integration
+- Egress proxy for network control
+- Signed skill verification (ed25519)
+
+### ✅ v0.2.x — Policy & Runtimes (Complete)
+
+- OPA (Rego) policy engine integration
+- gVisor (`sandbox_runtime`) support for stronger isolation
+- Skill manifest format (`skills/*.yaml`) with scope declarations
+- Adapter health monitoring and connection status
+
+### ✅ v0.3.x — Observability & UX (Complete)
+
+- Modern web dashboard (dark mode, Security Operations Center view)
+- Real-time terminal streaming with live log output
+- Prometheus metrics endpoint
+- OpenTelemetry tracing
+- Active secret redaction in logs and console output
+- Emergency lockdown / panic button
+- Security envelope visualisation (sandbox status indicator)
+- Explainable audit tooltips (why an action was allowed/denied)
+- Skill store with remote registry browsing and one-click install
+
+---
+
+## Upcoming Phases
+
+### 🔜 v0.4.x — Usability & Developer Experience (Q2 2026)
+
+#### Installation & Onboarding
+
+- **Package manager distribution**: `brew install aegisclaw`, `go install`, pre-built binaries via GoReleaser for Linux/macOS/Windows (amd64 + arm64)
+- **Interactive `init` wizard**: Guided setup that detects Docker/gVisor availability, configures default policies, and walks through first secret and skill registration
+- **Starter skill packs**: Curated bundles of safe, pre-signed skills (file organiser, web search, code runner) so users have something useful immediately after install
+- **`aegisclaw doctor`**: Diagnostic command that checks Docker, gVisor, secrets, adapter connectivity, and policy health in one go — outputs a clear pass/fail checklist
+
+#### Day-to-Day Workflow
+
+- **`docker-compose` skill orchestration**: Support multi-container skills (e.g., agent + database + cache) with coordinated sandboxing and shared network policies
+- **Policy templates**: Pre-built Rego policy profiles — `strict` (deny-by-default, approve everything), `standard` (allow known-safe, approve high-risk), `permissive` (allow most, log everything) — selectable during init
+- **Scope autosuggestion**: When a skill requests scopes beyond its manifest, suggest the minimal scopes needed based on observed behaviour rather than requiring manual YAML editing
+- **Dashboard mobile responsiveness**: Responsive web UI for monitoring agent activity from a phone or tablet
+- **Notification system**: Webhook, Slack, and email notifications for pending approvals, denied actions, and emergency lockdowns
+
+#### CLI Enhancements
+
+- **`aegisclaw replay <log-id>`**: Replay an audit log entry in dry-run mode to understand what happened and what would happen if re-executed
+- **`aegisclaw diff <policy-a> <policy-b>`**: Compare two Rego policies side-by-side with highlighted permission differences
+- **Shell completions**: Bash, Zsh, Fish, and PowerShell autocompletions generated from CLI metadata
+
+---
+
+### 🛡️ v0.5.x — Advanced Security (Q3 2026)
+
+#### Runtime Hardening
+
+- **Kata Containers / Firecracker support**: MicroVM-based isolation for workloads that need stronger-than-Docker boundaries
+- **Nix/bubblewrap sandbox**: Lightweight, non-Docker sandbox option for environments where Docker isn't available or desired
+- **Runtime behaviour profiling**: Learn normal syscall and network patterns per skill, flag anomalies in real-time (e.g., a file-organiser skill suddenly making network requests)
+- **Resource quotas**: CPU, memory, disk I/O, and network bandwidth limits per skill — prevent runaway agents from consuming host resources
+
+#### Secret Management
+
+- **`sops` integration**: Support Mozilla SOPS-encrypted files alongside `age`
+- **Pluggable vault backends**: HashiCorp Vault, Infisical, Bitwarden, and AWS Secrets Manager as secret sources — secrets are never written to disk unencrypted
+- **Secret rotation**: Automatic key rotation with configurable schedules and notification when skills need re-authentication
+- **Ephemeral secrets**: Short-lived credentials injected into sandboxes that auto-expire after execution
+
+#### LLM Safety
+
+- **NeMo Guardrails integration**: LLM prompt protection layer — detect and block prompt injection, jailbreaks, and off-topic steering before prompts reach the model
+- **Prompt/response audit trail**: Log every LLM interaction (prompt + response) with optional PII redaction, creating a full chain of accountability for agent decisions
+- **Token budget enforcement**: Per-skill and per-session token limits to prevent cost runaway from agent loops
+- **Output content filtering**: Configurable filters that flag or block agent outputs containing sensitive data, harmful content, or policy violations
+
+#### Auth & Access Control
+
+- **Tailscale/WireGuard integration**: Private mesh networking so the dashboard and API are only accessible over encrypted tunnels
+- **Authelia/Keycloak SSO**: Web UI identity provider integration for team deployments — RBAC with admin, operator, and viewer roles
+- **mTLS for adapter communication**: Mutual TLS between AegisClaw and OpenClaw endpoints to prevent man-in-the-middle attacks
+- **API key scoping**: Per-key permissions so different integrations (CI, dashboard, CLI) have minimal required access
+
+---
+
+### ✨ v0.6.x — Woo Factor & Ecosystem (Q4 2026)
+
+#### Visual & Interactive
+
+- **Live threat map**: Real-time animated dashboard view showing agent actions as they happen — skill executions pulse, denied actions flash red, approvals glow green — a "mission control" feel for your AI agents
+- **"Agent X-Ray" mode**: Click any running skill to see a live breakdown: active syscalls, open file handles, network connections, memory usage, and current scope consumption — full transparency into what the agent is actually doing inside the sandbox
+- **Security posture score**: Embeddable badge and dashboard widget (`AegisClaw: A+`) scoring your configuration across sandboxing, secret management, policy strictness, and audit integrity — gamification that rewards good security hygiene
+- **Approval UX overhaul**: Rich approval cards (web + Slack + mobile push) showing exactly what the agent wants to do, with context (which skill, what scope, risk level), diff of proposed changes, and one-tap approve/deny
+
+#### Skills Ecosystem
+
+- **Git-based skill distribution**: `aegisclaw skill install github.com/org/skill` — pull skills directly from Git repos with hash-chained provenance verification
+- **Skill marketplace**: Community registry with ratings, verified publishers, security audit badges, and automated vulnerability scanning of skill images
+- **Skill sandboxing profiles**: Per-skill seccomp and AppArmor profiles auto-generated from observed behaviour during a "learning" phase, then locked down for production
+- **Skill composition**: Chain multiple skills into workflows with data passing between sandbox boundaries — each step isolated, full audit trail across the pipeline
+
+#### Developer Experience
+
+- **VS Code extension**: Sidebar panel showing AegisClaw status, live audit stream, one-click approvals, and Rego policy linting with inline diagnostics
+- **`aegisclaw simulate`**: Dry-run mode that predicts what a skill would do (file access, network calls, resource usage) without actually executing it — like a flight simulator for agent actions
+- **Policy playground**: Browser-based Rego editor with live evaluation against sample skill manifests and audit scenarios — test policies before deploying them
+- **Terraform/Pulumi provider**: Infrastructure-as-code resources for provisioning AegisClaw instances, policies, and skill registries in team/org deployments
+
+#### Integrations
+
+- **MCP (Model Context Protocol) server**: Expose AegisClaw as an MCP tool server so any MCP-compatible AI assistant can run sandboxed skills through AegisClaw's security envelope
+- **GitHub Actions integration**: `aegisclaw/action@v1` that runs skills in CI with the same sandbox guarantees as local execution — consistent security in dev and CI
+- **Webhook-driven automation**: IFTTT-style triggers — "when a skill is denied 3 times, notify the team and auto-escalate to admin"
+
+---
+
+## Long-Term Vision (2027+)
+
+- **Multi-node orchestration**: Distribute agent workloads across multiple machines with centralised policy management and unified audit logs
+- **Federated skill trust**: Cross-organisation skill sharing with cryptographic trust chains — org A's signed skills are verifiable by org B without a central authority
+- **eBPF-based runtime monitoring**: Kernel-level observability without modifying the sandbox — trace syscalls, network flows, and file access at near-zero overhead
+- **AI-powered policy generation**: Analyse a skill's code/manifest and automatically suggest the minimal Rego policy — "this skill only needs `files.read:/tmp` and `net.outbound:api.openai.com`"
+- **Compliance frameworks**: Pre-built policy packs for SOC 2, HIPAA, GDPR, and NIST — one command to apply a compliance baseline
+- **AegisClaw Cloud**: Hosted SaaS with org management, centralised dashboards, SSO, and managed skill registries for teams that don't want to self-host
+
+---
+
+## How to Contribute
+
+We welcome contributions! See [CONTRIBUTING.md](CONTRIBUTING.md) for details.
+
+**High-impact areas right now:**
+
+- 🐳 Adding Kata Containers / Firecracker runtime support
+- 🔐 Pluggable vault backend implementations (Vault, Infisical, Bitwarden)
+- 📝 Writing and publishing community skills with signed manifests
+- 🧪 Security testing and fuzzing of the sandbox boundary
+- 📚 Documentation improvements and tutorials
+
+Report bugs or request features via [GitHub Issues](https://github.com/mackeh/AegisClaw/issues).