v3.1.0: Epic-Swarm Reliability Overhaul

Epic-Swarm Reliability Overhaul (Post-Mortem Fixes)

Based on the first production run of /epic-swarm (PRO-330), this release addresses all issues from the post-mortem report.

Critical Fixes (P0)

• Sequential write dispatch — Write phases now dispatch agents one at a time with explicit cd to each worktree, guaranteeing isolation. Read-only phases remain parallel. Configurable via SWARM_PARALLEL_WRITES.
• Working directory enforcement — Agent prompts include absolute path instructions and explicit file manifests for read-only phases.
• Worktree integrity verification — Mandatory check after every agent dispatch catches cross-contamination before it propagates.

High Priority (P1)

• Approval gates — User confirmation required before merge-to-main, push-to-remote, and ticket closure. Gated on SWARM_AUTO_MERGE config.

Medium Priority (P2-P3)

• Phase-skip policy — Orchestrator cannot skip phases autonomously; must present rationale and wait for user approval.
• Blocking baseline tests — Tests must pass before wave execution begins.
• Swarm state persistence — Field-level schema with per-ticket phase tracking, updated after every significant event.

Improvements (P4-P5)

• Graceful degradation — Works without /planning metadata via heuristic dependency analysis.
• Token budget strategy — Prescriptive docs always included in full; contextual docs truncated with pointers when needed.
• Dual security review — Pre-merge per-ticket scan + post-merge comprehensive review on integrated codebase.

Execute-Ticket Alignment

• Post-dispatch verification step after every agent
• Worktree mode security note clarifying pre/post-merge review split
• Phase-skip approval requirement

Documentation

• New /epic-swarm section in TECHNICAL_REFERENCE.md
• Updated README, QUICK_REFERENCE, and CHANGELOG

See CHANGELOG.md for full details.