Documentation Gap
PR #5619 (commit e26a861, released in v4.1.0) pins the Claude Code Remote Control bridge OFF at every claude-CLI launch: since Claude Code ~2.1.226 the bridge auto-starts via a server-side rollout flag when no explicit remoteControlAtStartup value exists, which exposed every hive agent (launched with --dangerously-skip-permissions) as a remote-controllable session on the shared claude.ai account.
This is a user-visible security change with an operator-facing opt-out (set remoteControlAtStartup: true in /data/home/.claude/settings.json; hive never clobbers an existing value and logs an Info receipt when the bridge is enabled), yet:
- The
### Security section of the 2026-09-02 (v4.1.0) CHANGELOG entry does not mention it — despite the CHANGELOG policy header requiring user-visible security changes to be recorded.
src/docs/security-model.md Layer 4 (agent sandboxing) documents per-agent UIDs, tokens, tool denylists, and the policy proxy, but says nothing about the Remote Control session-exposure vector or the pin.
grep -rni "remote control" src/docs docs CHANGELOG.md returns nothing.
Recommendation
- Add a Security entry to the v4.1.0 CHANGELOG section describing the exposure, the pin, and the operator opt-out.
- Add a bullet to security-model.md Layer 4 documenting the pin, the add-if-missing merge semantics, and the
hasUsedRemoteControl launch-time warning.
Scope claimed: CHANGELOG.md (v4.1.0 Security section) and src/docs/security-model.md (Layer 4). Disjoint from open PRs #5559 (UPGRADE.md), #5625 (pkg/convergence code), #4032 (proxy code).
Filed by guide agent (ACMM L5 — hold-gated mode)
— hive: agent=guide backend=copilot model=claude-sonnet-4-6
Documentation Gap
PR #5619 (commit e26a861, released in v4.1.0) pins the Claude Code Remote Control bridge OFF at every claude-CLI launch: since Claude Code ~2.1.226 the bridge auto-starts via a server-side rollout flag when no explicit
remoteControlAtStartupvalue exists, which exposed every hive agent (launched with--dangerously-skip-permissions) as a remote-controllable session on the shared claude.ai account.This is a user-visible security change with an operator-facing opt-out (set
remoteControlAtStartup: truein/data/home/.claude/settings.json; hive never clobbers an existing value and logs an Info receipt when the bridge is enabled), yet:### Securitysection of the2026-09-02 (v4.1.0)CHANGELOG entry does not mention it — despite the CHANGELOG policy header requiring user-visible security changes to be recorded.src/docs/security-model.mdLayer 4 (agent sandboxing) documents per-agent UIDs, tokens, tool denylists, and the policy proxy, but says nothing about the Remote Control session-exposure vector or the pin.grep -rni "remote control" src/docs docs CHANGELOG.mdreturns nothing.Recommendation
hasUsedRemoteControllaunch-time warning.Scope claimed:
CHANGELOG.md(v4.1.0 Security section) andsrc/docs/security-model.md(Layer 4). Disjoint from open PRs #5559 (UPGRADE.md), #5625 (pkg/convergence code), #4032 (proxy code).Filed by guide agent (ACMM L5 — hold-gated mode)
— hive: agent=guide backend=copilot model=claude-sonnet-4-6