Skip to content

Releases: osovv/vv-opencode

v0.35.24

Choose a tag to compare

@github-actions github-actions released this 18 Jun 21:25
91ad85a

0.35.24 (2026-06-18)

Summary

This release restores the 1.05M context limit for the openai patch-preset after a brief adjustment to 400K for ChatGPT Pro plan compliance, and introduces explicit intent review rounds to the WorkflowPlugin: work items now require mode (implementation or review_only) and requiredReviewers (spec, code, or both), reviewers are launched and tracked in parallel, results are collected into a full review round before deciding the next lifecycle state, and review-only mode treats reviewer FAIL as a completed finding without routing to the implementer—giving users more deterministic and flexible pipeline control.

  • fix: restore 1.05M openai patch-preset limits (9d5df8a)
  • fix: set openai patch-preset context to 400K for ChatGPT Pro plan compliance (0177f03)
  • fix: use 400K context limit for openai patch-preset (ChatGPT Pro plan) (84c6029)
  • feat(workflow): add explicit intent review rounds (f02e784)

v0.35.23

Choose a tag to compare

@github-actions github-actions released this 18 Jun 20:24
9893df3

0.35.23 (2026-06-18)

Summary

This release improves compatibility and model configuration by adding explicit 1.05M context and 128k output limits to OpenAI patch-preset models (GPT-5.4 and GPT-5.5), ensuring those models operate at their intended capacity, and by switching the sync, read, and guardian config paths to use lenient config parsing so that older 4-role configurations without the reviewer role are gracefully upgraded with defaults instead of causing errors. This prevents upgrade breakage for users with legacy configs and ensures the new reviewer role is automatically populated.

  • fix: add 1.05M context / 128k output limits to openai patch-preset models (8c5b48c)
  • fix: use lenient config parsing in sync/read paths to handle old 4-role configs (ae11b70)

v0.35.22

Choose a tag to compare

@github-actions github-actions released this 18 Jun 17:00
a3f8e3f

0.35.22 (2026-06-18)

Summary

This release removes the orchestrator role, simplifying the built-in role system to five roles with a clean smart and reviewer split. The vv-controller agent has been re-bound to the smart role, and all built-in presets have been updated to drop the orchestrator entry. Additionally, the vv-zai preset now correctly assigns the GLM-5.1 model to the reviewer role and the GLM-5-turbo model to the orchestrator role before its removal, ensuring proper review model selection. This reduces configuration complexity and clarifies the separation between primary smart agents and dedicated reviewers.

  • refactor: remove orchestrator role, keep only smart + reviewer split (86779bd)
  • fix: swap zai reviewer/orchestrator models — glm-5.1 for review, glm-5-turbo for orchestration (877a57b)
  • feat: add reviewer and orchestrator roles, split smart role bindings (42baa72)

v0.35.21

Choose a tag to compare

@github-actions github-actions released this 16 Jun 15:11
10060c9

0.35.21 (2026-06-16)

Summary

This maintenance release removes the dead .vvoc/plans/ directory path and stale vv-analyst and vv-architect agent references, as planning artifacts now live exclusively under the .vvoc/specs/<id>/ layout. The GRACE documentation (development plan, verification plan, knowledge graph, and requirements) has been updated to reflect the current architecture, and several module contracts, maps, and change summaries have been corrected or added. These changes reduce code surface, eliminate confusion from outdated references, and ensure that project documentation accurately describes the managed skills and spec-driven planning flow.

  • chore: refresh GRACE artifacts after full integrity scan (e0df404)
  • refactor: remove dead .vvoc/plans/ code path and stale vv-analyst/vv-architect references (f4076e1)

v0.35.20

Choose a tag to compare

@github-actions github-actions released this 16 Jun 11:48
b3b6fc0

0.35.20 (2026-06-16)

Summary

This release introduces layered project-scope configuration, allowing users to isolate vv-opencode setup to individual projects with vvoc install --scope project, while the new vvoc launch command launches OpenCode with deterministic config sources for sandboxed testing. Complex spec sessions now support an optional design-context.xml companion that preserves decision rationale and rejected alternatives for planners and reviewers without expanding the normative spec, and the skill sync system gains config-safety rules that prevent silent overwrites of user-owned reference files. Documentation and templates have been cleaned up by removing stale references to legacy sub-agents and folder layouts, consolidating the spec package directory as the canonical organizational unit and making the user-facing workflow documentation more accurate.

  • docs: remove legacy folder references from docs and templates (10879d4)
  • docs: remove stale vv-analyst/vv-architect references (30e9bab)
  • feat(config): support layered project scope (beaaa4b)
  • feat(skills): add spec package design context (5181cdf)

v0.35.19

Choose a tag to compare

@github-actions github-actions released this 15 Jun 10:00
7435e4f

0.35.19 (2026-06-15)

Summary

The vv-reflect skill now synthesizes generalized lessons and reusable procedures instead of session recaps, capturing durable domain knowledge, business context, and product intent so that future agents can apply insights to similar-but-not-identical tasks rather than replaying what happened in a single session.

  • feat(skills): improve vv-reflect lesson synthesis (d458c15)

v0.35.18

Choose a tag to compare

@github-actions github-actions released this 15 Jun 09:31
a7fce07

0.35.18 (2026-06-15)

Summary

Updated the vv-osovv preset's smart agent to use openai/vv-gpt-5.5-xhigh, replacing the previous DeepSeek-based model, to deliver higher-quality responses for complex reasoning tasks while continuing to use existing fast, vision, and default agents for other workloads.

  • feat(preset): update vv-osovv smart model to openai/vv-gpt-5.5-xhigh (89752e2)

v0.35.17

Choose a tag to compare

@github-actions github-actions released this 14 Jun 15:41
31fd0da

0.35.17 (2026-06-14)

Summary

This release improves the accuracy of automatically generated changelog summaries by feeding the full textual diff of each commit into the summary generation prompt, so the model can ground its output in the actual file changes rather than relying solely on commit titles and metadata. This means release notes are now more faithful to what was really modified, reducing the risk of invented or misleading descriptions in the changelog.

  • fix(release): include commit diffs in summaries (f1c930c)

v0.35.16

Choose a tag to compare

@github-actions github-actions released this 14 Jun 14:35
6b2e7c4

0.35.16 (2026-06-14)

Summary

This release introduces lifecycle statuses for skills specs and plans, giving users clearer visibility into the state of their skill workflows—whether a spec is being drafted, reviewed, or finalized, and whether a plan is in progress, completed, or blocked—making it easier to track progress and identify next steps in skills-based automation.

  • feat(skills): add spec and plan lifecycle statuses (5c7c095)
  • chore: add typecheck to lefthook pre-commit (cca0f03)

v0.35.15

Choose a tag to compare

@github-actions github-actions released this 14 Jun 09:41
65acfb7

0.35.15 (2026-06-14)

Summary

This release fixes a test issue by adding a missing id field to an inline type in the patch-provider test, ensuring test accuracy and preventing potential false failures during validation.

  • fix: add missing id field to inline type in patch-provider test (6f6e33c)