Changelog

Notable changes to the vibe-engineer CLI and the workflow it ships. Releases use semantic versioning: minor bumps add features and stay backwards compatible; major bumps would break the chunk schema or the runtime contract.

0.2.0

Feature release. Fully backwards compatible. The headline change is the intent-ownership shift: chunks now have a canonical statement of what they're for, a sharpened five-status taxonomy, and tooling to migrate older corpora into the new standard. Substantial body of work outside the headline too: wiki-based entity memory, the leader-board cross-agent messaging layer, orchestrator reliability improvements.

Intent ownership

• docs/trunk/CHUNKS.md: canonical statement of the four principles. Code owns implementation; chunks own intent; chunks exist only for intent-bearing work; goals describe intent in present tense; status answers a single ownership question.
• Five-status taxonomy: FUTURE, IMPLEMENTING, ACTIVE, COMPOSITE (new), HISTORICAL. COMPOSITE captures chunks that share intent ownership with peers.
• SUPERSEDED deprecated. The runtime keeps parsing it for backward compatibility, but the ACTIVE → SUPERSEDED transition is removed (no new chunk becomes SUPERSEDED) and the SUPERSEDED → HISTORICAL off-ramp stays open so the existing set drains organically. Parsing a SUPERSEDED chunk emits a DeprecationWarning.
• /audit-intent skill: full-corpus migration tool. Walks every ACTIVE chunk, applies one of six actions per chunk (rewrite tense, log over-claim, fix broken code_paths, historicalize, log cross-artifact inconsistency, leave clean), and produces a structured report. Designed for parallel sub-agent fan-out at 5 chunks per agent.
• Migration guide covering the operator-led-to-full-auto rollout pattern. Reference numbers from the self-migration of this project's own corpus (341 chunks, 11 waves, ~180 rewrites, 7 historicalizations, 62 inconsistency entries logged for triage).
• docs/trunk/INCONSISTENCIES/ log: concurrent-safe record of mismatches between what chunks claim and what code delivers. Designed to outlive any individual audit.

Entity system v2 (wiki-based memory)

• Entities now keep their long-running memory as a structured wiki rather than a flat memory tier. Wiki schema with identity, knowledge domains, and structured pages.
• ve entity create, attach, detach, fork, merge, push, pull: lifecycle commands for entities backed by git submodules.
• ve entity startup / shutdown: load identity + recent memory; consolidate session work back into the wiki at session end.
• ve entity ingest-transcript, from-transcript: bootstrap an entity from prior session transcripts.
• ve entity touch + episodic-memory commands for retrieval-as-reinforcement and on-demand search across past sessions.

Leader board (cross-agent messaging)

• End-to-end-encrypted append-only log channels between agents across machines. Cloudflare Durable Objects backend with a local-server adapter for dev/test.
• ve board send, watch, watch-multi, ack, channels: cursor-based at-least-once delivery with manual ack semantics.
• Steward skills (/steward-setup, /steward-watch, /steward-send, /steward-changelog): persistent project-watching agents that triage inbound work via the board.
• Swarm-level monitoring via /swarm-monitor and request-response patterns via /swarm-request-response.

Orchestrator

• Mandatory review phase between implement and complete with a structured ReviewDecision tool. Review feedback file (REVIEW_FEEDBACK.md) drives a bounded review ↔ implement loop (max 3 iterations on the orchestrator path).
• Worktree retain mode (ve orch inject --retain) preserves the worktree for inspection after completion.
• Phase-aware crash recovery, optimistic locking on state transactions, deferred worktree creation, finalization recovery, session auto-resume on Anthropic API rate limits.
• Conflict oracle: detects whether two in-flight chunks touch compatible code regions; supports parallelize / serialize operator decisions.
• Sandbox enforcement: blocks agents from running git commands outside the chunk's worktree.
• Dashboard live-tail for streaming agent logs over WebSocket; ve orch tail for terminal access to the same.

Workflow / chunk lifecycle

• /chunk-execute with built-in review loop. /chunk-review as a standalone skill that spawns a fresh subagent for an independent read of intent vs. implementation.
• Reviewer infrastructure: per-reviewer trust levels, decision file format, auto-curation hooks, decision template rendering.
• /narrative-execute walks a narrative's proposed chunks in dependency order, dispatching parallel agents.
• /validate-fix iteratively fixes validation errors until clean, gated by max-iteration and error-classification caps.

CLI / templates / refactors

• src/ve.py reduced to a thin entry point; operational logic lives in src/cli/ per artifact type.
• src/models.py split into the src/models/ package, one module per artifact.
• src/orchestrator/api.py split into the src/orchestrator/api/ package by endpoint family.
• Chunk and subsystem directory naming dropped the NNNN- sequence prefix in favor of pure short-name directories. Causal ordering moved to the created_after frontmatter field.
• ve chunk start renamed to ve chunk create (with start kept as an alias).
• Investigation artifact type added; subsystem artifact type gained richer status transitions and CLI surface.
• Friction log workflow: capture pain points, accumulate them, promote to chunks when patterns emerge.
• Task directories: cross-repository workflows where a single task coordinates artifacts across multiple project repositories via external.yaml pointers.

Backwards compatibility

The runtime continues to parse status: SUPERSEDED on existing chunks. Older chunks that haven't been migrated keep working; they'll just emit a deprecation warning when read. Projects can upgrade ve at any time and migrate at their own pace using /audit-intent.

0.1.0

Initial release. Core chunk lifecycle (create, plan, implement, complete), trunk documentation scaffolding, project init, template rendering, basic CLI surface for chunks and narratives.