v1.6.0

Behaviour changes in exec

exec now decides success from the exit code. Check any caller that relied on the old rule, which treated stderr output as the failure signal.

• A non-zero exit with empty stderr now fails. It used to return as a success with its stdout. This affects grep with no match, test -f on a missing file, and diff with differences.
• An exit of 0 with stderr output now succeeds. It used to fail. The stderr follows stdout under a [stderr] marker.
• A failure now reads exit <code>:, then stderr, then stdout under [stdout]. A command that a signal ended reads exit 137 (signal KILL).

Most of the pull requests listed below change ssh-fleet-mcp-server, which has its own releases.

What's Changed

• fix(ci): pin .nvmrc to 24.21.0 for reliable OIDC publishing by @jordanburke in #27
• docs(spec): fleet governance layer by @jordanburke in #26
• feat(orchestrator): fleet governance phase 1 — audit log by @jordanburke in #28
• feat(orchestrator): fleet governance phase 2 — per-host policy by @jordanburke in #29
• feat: fleet governance phase 3 — CA certificates by @jordanburke in #30
• feat(orchestrator): first-run setup — starter config, fleet_setup tool, check subcommand by @jordanburke in #31
• fix(core): find Homebrew tmux; treat a missing tmux socket as no sessions by @jordanburke in #41
• fix: both servers install and start from npm by @jordanburke in #45
• docs: running the fleet server on several machines by @jordanburke in #49
• fix(orchestrator): exec exit-code signal, hosts_list error state, config-edit notice, build stamp by @jordanburke in #57
• chore: somamcp 2.4.0, with a real label on every fleet tool error by @jordanburke in #62
• fix: Mac mini's reports — signal exit, connector stderr, exec timeout_ms, audit client label by @jordanburke in #63
• fix: PR #63 review follow-ups — range-check [pool], tmux timeouts, no-exit-status closes by @jordanburke in #64
• feat: somamcp 2.5.0, and audit records the client's handshake name by @jordanburke in #65
• feat: publish path for ssh-fleet-mcp-server, and a packaging check on the real artifact by @jordanburke in #66
• chore: connector 1.6.0, and a 30-minute wait for the fleet tarball by @jordanburke in #67

Full Changelog: v1.5.1...v1.6.0