AgentRun 0.4.0 - Docker Compose and Make

AgentRun now manages local Docker Compose stacks through their container identities and runs approved targets from existing Makefiles. No wrapper or project-file rewrite is required.

• Native Core adapter shared by CLI and stdio MCP: create, inspect, logs, stop and restart registered containers.
• Engine identity, full container IDs and Compose labels verified before destructive operations. No automatic adoption, volume deletion or Docker cleanup.
• Explicit make profiles and optional prepareMake for Compose; existing preparation targets can generate env files. Optional build/pull permissions remain in user policy.
• Six stack MCP tools, agentrun stack subcommands and agentrun start --profile. Inventory and stop-all include stacks.
• Updated setup guidance and 38 initial profiles. Existing policies are preserved.

Validation: 53 Rust tests, including real Linux process groups, Make, Docker Compose and MCP integration. Static musl binaries are built and tested by CI on native x86_64 and ARM64 runners; installation is exercised in common Linux distributions.

After upgrading, rerun agentrun-setup codex and open a new session. Add appropriate Make/Compose profiles explicitly to existing policies. See docs/service-managers.md for configuration and limits.

Compatibility: older process registries remain readable. Once stacks are recorded, older AgentRun binaries reject the extended registry. Stack restart reuses existing containers; it does not rebuild/recreate. Fixed container names, adoption, replication, watch, systemd and remote Docker are unsupported. This is not a sandbox for untrusted code.