Release: v0.22.8

This release improves Opus 5 workflow efficiency and strengthens execution contracts across agents.

Highlights

• Bounds exploration and implementation to approved scope, Direct MVP, and total change budgets.
• Improves handoffs for quality, integration-test, security, and final verification flows.
• Adds irreversible-risk coverage, shared-mutation route parity, and consumer-observable capability checks.
• Removes model-known and duplicated guidance from shared coding, testing, and frontend skills while preserving broad discovery and repository-specific rules.
• Preserves false-positive test guardrails and gates bundle/rendering fixes on approved scope and measurable evidence.
• Keeps generated plugin metadata synchronized.

Release: v0.22.7

This release refines development workflows to prioritize required value and minimize unnecessary design surface.

Highlights

• Adds evidence-backed Design Convergence through Direct MVP, Failed Items, Adopted Additions, and Rejected Additions.
• Keeps quality fixes within the confirmed task scope and improves review routing for missing inputs.
• Consolidates prompt guidance and synchronizes all plugin distributions.

Release: v0.22.6

What's Changed

• Updated TypeScript and React guidance to follow each project's configured tooling, compatibility targets, and quality signals.
• Made test guidance project-aware across runners, file conventions, network mocks, browser harnesses, and configured quality thresholds.
• Added verification-aware execution for TDD, behavior-preserving refactors, blocked bug reproductions, and non-executable deliverables.
• Improved integration and E2E test selection with explicit lanes, deduplication, scoring criteria, per-feature budgets, and existing-harness reuse.
• Refined implementation guidance for fallback observability, impact analysis, cleanup scope, authentication and authorization review, and targeted test execution.
• Reduced unnecessary external research and alternative generation while preserving meaningful implementation strategy decisions.
• Aligned task analysis, planning, execution, and review handoffs, including Work Plan review gates, retry safeguards, and affected-verifier reruns.
• Consolidated external-resource discovery into a single declaration after the required domain context is collected.

Release: v0.22.5

What's Changed

• Improved Design Doc reviews by validating adopted designs against confirmed requirements and component contracts.
• Kept confirmed scope and analysis context consistent across design and review workflows.

Release: v0.22.4

Changed

• Agent definitions now declare their skills frontmatter as a YAML list instead of a comma-separated string, matching the format documented for subagent frontmatter. Skill names are unchanged, so agent behavior is identical. The tools field remains comma-separated, per its documented form.

Release: v0.22.3

Patch release adding an evidence-backed assumption check to the design-doc workflow.

Changes

• Behavioral assumption verification: Design docs often rely on unstated behavioral claims — a framework/library default, a capability assumed already provided, or a feature assumed already implemented — that flow into implementation unverified. The existing dependency check only confirmed that a component exists, not that it behaves as assumed. Each relied-upon behavioral claim now attaches evidence (codebase file:line, command result, or authoritative doc) at design time, or is recorded as an unverified assumption bound to a follow-up.
• Generator (technical-designer, backend + frontend): a Behavioral Claim Verification step records each claim with Confirmed: Yes and its evidence, or Confirmed: No plus a matching Risks and Mitigation row naming how it will be verified during implementation (command, test, or code-inspection point) or guarded by a fallback.
• Template (design-template): an Assumed Behaviors slot in the Agreement Checklist, reusing the existing Evidence/Confirmed pattern; the restated claim links the slot row to its Risks and Mitigation row.
• Reviewer (document-reviewer): a behavioral-claim evidence check flags, as an important feasibility issue, any relied-upon claim that is absent from the slot, Confirmed: Yes without evidence, or Confirmed: No without a matching Risks and Mitigation row.

Release: v0.22.2

Fixed

• Subagents orchestration guide: enforce fresh subagent invocation. The Execution Method section now states that each subagent invocation is a fresh Agent tool call that isolates each phase's context. A SendMessage resume reuses the prior agent's context and breaks that isolation, so it is no longer an implicit option. This prevents the orchestrator from continuing a completed subagent and silently disrupting the workflow.

Release: v0.22.1

Patch release hardening the task-executor agents.

Changes

• Unimplemented dependency handling: The executor now treats a missing or unimplemented dependency as a stop condition only when it prevents preserving the required contract and no local, reversible construct (local slice or contract-preserving stub/adapter) can satisfy it. Otherwise it proceeds with the local approach and records the integration handoff; genuinely undecidable architectural trade-offs escalate.
• Escalation detail completeness: Design-compliance escalations now populate every field the escalation schema requires, with the Design Doc requirement mapped to design_doc_expectation and the actual situation (absent dependency / undecided decision) mapped to actual_situation.

Release: v0.22.0

Highlights

• Added the new llm-friendly-context skill for clearer LLM-facing prompts, handoffs, generated artifacts, and instructions.
• Wired llm-friendly-context into artifact-producing agents and recipe prompt/artifact creation points.
• Added document-reviewer coverage for LLM-facing artifact clarity, including ambiguous downstream execution and non-executable downstream work.
• Strengthened task-decomposer self-checks to reduce ambiguous task files and preserve consistency with the Work Plan and related design artifacts.
• Added llm-friendly-context to the task analyzer skill index and README skill listings.
• Bumped package and plugin versions to 0.22.0.

Release: v0.21.3

Strengthens failure-mode and spec-completeness coverage across the planning agents so a feature's combinatorial shape is forced into a verifiable contract instead of slipping through planning, task decomposition, and review.

Improvements

• New failure-mode category — missing-sort-key ordering added to the domain-independent Failure Mode Checklist (ordering when some elements lack the sort key). The work plan now enumerates nine categories, and the document reviewer flags it when applicable.
• Mode × branch combination expansion — when a change adds a mode, flag, or variant that overlays an existing selection / ordering / filtering / display branch, acceptance-criteria drafting now requires expanding the combination of the new value with each existing branch value. Catches modes that take effect on one branch while silently no-opping on the others.
• Reset/clear lifecycle as a contract — the Design Doc Client State Design table gains a Reset/Clear Behavior column, and the work planner records "state returns to its unused/default value on reset" as a state-lifecycle negative so it is verified rather than assumed.