v2.3.0

What's New

Closes all 15 identified gaps between the spec documents and the implementation. 776 tests passing.

Core Algorithms

• Critical path analysis — longest-path DP on topological sort for identifying bottleneck chains
• Post-hoc sizing validation — per-task token budget checks with warnings for over-budget tasks
• Spec sharding — automatic splitting of >150K token specs into manageable shards

Dispatch & Triggers

• --api flag — direct Anthropic API invocation for spec decomposition (requires anthropic package)
• on_rework trigger — detects rework tasks via discovered-from dependencies

Hole Lifecycle

• Resolution propagation — when a hole is resolved, blocked tasks are automatically updated
• Progressive refinement — narrow hole scope incrementally with partial answers
• Parallel conflict detection — warns when parallel tasks produce the same files

Report Enrichment

• Duration and token columns in completed task tables
• Key Decisions, Warnings, and Risks sections in progress reports
• Report JSON now includes report_number, epic_title, trigger, holes, and ready_work fields

Data Model Fixes

• state.yaml now includes description, produces, consumes, acceptance_criteria fields
• JSON status uses not_started instead of open per schemas.md
• Orchestration verification gates check per-task status and run acceptance tests

Bug Fixes

• --dir flag now respected for beads output
• Model IDs updated to claude-opus-4-6 / claude-sonnet-4-6
• write_beads_output() creates output directory when needed

Testing

• 776 tests (up from 312 at v2.2.0 baseline)
• 7 new Hypothesis property-based tests for algorithms
• 8 new integration tests for cross-module verification
• Full E2E validation of all CLI flows

Full Changelog: v2.2.0...v2.3.0

What's Changed

• feat: address all 15 spec-implementation gaps (v2.3.0) by @rand in #3

Full Changelog: v2.2.0...v2.3.0