v0.6.1: progressive disclosure, launch polish

The skill is now a short conductor checklist (SKILL.md) with four references that load only when that part of a campaign is active: codex-dispatch, fleet-operations, squads, review-gates. Bootstrap templates ship in the skill's assets/campaign-hq/ and worker reports collect into out/.

Also: honest plan-usage sizing for Codex waves, restored multi_agent_v1 native-subagent doctrine with worked squad-lead examples, four-line codex convergence in the patterns poster, a Codex-led squad in the README diagram, and examples matched to the new templates.

v0.4.0 — permissions doctrine, powerful defaults

• Permissions section: default worker envelope is workspace-write with network on and approvals off for Codex (both verified against the CLI), edits pre-accepted for Claude workers; tighten only on user request
• Strongest-model default: no autonomous downshifting; per-task downgrades require a recorded user preference
• Session-resume steering, CI as a second verification layer
• README: fleet topology and wave diagrams (vertical, render-verified), usage section, worked brief and report schema in examples

v0.3.0 — squads, steering, review gates

First tagged release of campaign-conductor.

• Conductor doctrine: bootstrap, worker routing with persistent preferences, campaign sizing
• Parallel fleets: worktree-per-worker waves, fleet table with session ids, stall detection, speculative dispatch
• Squads: depth-capped nested delegation with namespace isolation and per-leaf evidence
• Worker reports and review gates: fixed-schema reports, cross-model review, bake-offs
• Codex capabilities doctrine: web search scouts, image input/generation, review mode, session resume steering
• Strongest-model default; downshifts only on recorded user preference
• Examples: mid-campaign state files, worked brief, worker-result schema