Wukong Code Changelog

• -Preserve Loop revisions, resumed usage, provider imports, and cross-workspace session boundaries without losing user configuration
• -Make automatic updates notification-only until the user explicitly runs wukong upgrade
• -Require HTTPS and bounded plugin archives, support SHA-256 verification, install disabled first, and roll back atomically on failure
• -Discover checks only from real root or workspace-member declarations and retain explainable provenance
• -Improve local Web reconnects, permission controls, keyboard navigation, workspace persistence, dialogs, and side chats
• -Keep the release free, local-first, and BYOK without adding commands, paid plans, hosted inference, or managed model credits

Why v0.1.1 shipped this way

• -Add one editable Preflight for the Goal, Finish Line, checks or versioned no-check decision, provider destination, permissions, workspace state, and runtime limits
• -Enforce provider-call and optional token ceilings and stop blind retry when a provider outcome is unknown
• -Require complete and fresh Gate evidence; incomplete, stale, conflicting, interrupted, or corrupt evidence cannot become PASS
• -Show deterministic terminal results, decisive evidence, file attribution, checks, usage, blocker, recovery, and next action
• -Preserve direct upgrade and local-state compatibility from v0.0.22
• -Publish six immutable native ZIPs with adjacent SHA-256 files and exact-source provenance; macOS binaries are intentionally unsigned and unnotarized

Why v0.1.0 shipped this way

• -Fall back from modelpricelab.com to models.dev when the primary provider catalog fails; the published ZIPs still require one remote catalog because the intended embedded snapshot was omitted
• -Include common GLM and Kimi provider entries in the source-tree fallback catalog
• -Apply --skills-dir to interactive terminal sessions as well as headless runs
• -Present -r, --resume as the canonical session-resume option while retaining -S, --session as a hidden compatibility alias
• -Explain guarded Auto, YOLO, Plan, headless prompts, --yes confirmation, and additional workspace access more clearly
• -Describe judge as a deterministic local merge gate rather than a guarantee that a workspace is safe to merge
• -Publish six native ZIP files and six matching SHA-256 files for macOS, Linux, and Windows

Why v0.0.22 shipped this way

• -Derive built-in risk evidence identity from the complete occurrence: kind, severity, message, file, line, and evidence
• -Preserve distinct occurrences of the same risk kind instead of rejecting later findings as duplicate broker input
• -Keep identical occurrences deterministic so exact duplicate input can still be deduplicated
• -Pass the complete bounded risk evidence set to the Loop reviewer without weakening deterministic Gate failures
• -Publish the same focused patch across six native targets with matching ZIP and SHA-256 assets
• -Exclude isolated worktrees, Safe Promote, Windows FFI, hosted execution, payments, quota, source upload, and telemetry backend

Why v0.0.21 shipped this way

• -Discover strict verification criteria from existing personal, project, monorepo package, trusted local, and built-in Skill sources
• -Keep deterministic suggestions unselected until the user confirms exact source, full provenance digest, criteria, and order
• -Use one isolated fresh reviewer with bounded evidence broker tools instead of arbitrary workspace, shell, Skill, MCP, or web access
• -Require separate local broker trust and per-run confirmation of the exact BYOK provider destination and allowed input origins
• -Ship optional API compatibility, migration safety, log security, and frontend design reference recipes
• -Publish six native ZIP files and six SHA-256 files, including a verified Intel macOS SEA compatibility fallback
• -Keep the release free, local-first, and BYOK with no Skill command runtime, hosted inference, payments, quota, cloud sync, or telemetry metrics backend

Why v0.0.20 shipped this way

• -Confirm Goal, optional Done when, Must not constraints, real checks, warnings, and the run limit before creating a Loop
• -Preview the same Finish Line with headless --dry-run and require exact Gate, warning, and workspace trust confirmation
• -Freeze required-check command, source locator, definition, order, policy, workspace, and approval digests
• -Reject known inline credentials and recognizable destructive or upload commands before persistence or spawn
• -Join checks, scan, reviewer, aggregate, evidence, attempt completion, and terminal results with durable local IDs and digests
• -Create a new traceable continuation run for terminal work and a revised Goal/contract when the Finish Line changes
• -Keep the release free, local-first, and BYOK with no hosted inference, payments, quota, cloud sync, or telemetry storage backend

Why v0.0.19 shipped this way

• -Keep the default TUI command surface focused on Resume → Loop → Gate, with advanced tools under /help advanced
• -Resume local Kimi Code and Grok sessions through bounded, read-only imports
• -Confirm Device Login on the Wukong website with Google or GitHub, while allowing an honest empty model catalog
• -Preserve the user’s BYOK default model when no Wukong-hosted models are available
• -Accept explicit text-only /feedback without login, attachments, logs, prompts, source code, or local evidence
• -Retain feedback for at most 90 days and apply edge rate limiting without storing IP or device identifiers
• -Keep hosted inference, payments, monthly Loop quotas, Hosted Terminal, cloud sync, and Web/D1 product telemetry out of this release

Why v0.0.18 shipped this way

• -Released as the current public download for macOS, Linux, and Windows
• -Make /resume the primary TUI command while retaining /sessions as a compatibility alias
• -Keep /login account-only and move every model API-key workflow to /provider
• -Remove Guest and Free Loop allowances; local BYOK Loops have no Wukong monthly limit
• -Migrate local Loop recovery records from quota-era lifecycle v1 to content-free lifecycle v2
• -Remove Hosted Terminal, legacy Companion routes, managed model claims, Checkout, and preview plans from the website
• -Keep existing opt-out-aware CLI telemetry without adding a Web or D1 product-metrics backend

Why v0.0.17 shipped this way

• -Released as the current public download for macOS, Linux, and Windows
• -Keep each Loop focused on the goal and finish condition you gave it
• -Remember earlier blockers so review cannot silently move the target
• -Stop repeated no-progress cycles with a clear result and next step
• -Make Auto guarded and non-interactive while keeping YOLO explicit and bounded by hard safety rules
• -Add bounded Resume suggestions, Role contracts, compact task status, and local session prompt search

Why v0.0.16 shipped this way

• -Released as the current public download for macOS, Linux, and Windows
• -Recover an active Loop after restart, or end it as cancelled without fabricating a Gate result
• -Run objective, model, workspace, Goal, and project-check preflight before reserving Loop quota
• -Retry failed lifecycle completion through a private local outbox with bounded backoff
• -Show shorter first-run and terminal result surfaces with an actionable primary blocker
• -Expose real active, stale, completed, cancelled, and abandoned lifecycle states in Dashboard and protected metrics

Why v0.0.15 shipped this way

• -Released as the current public download for macOS, Linux, and Windows
• -Resume unfinished Codex, Claude, and Cursor sessions from the TUI with /resume
• -Continue a resumed task directly or turn it into an editable /loop objective
• -Show clear Gate stages and reasons instead of ambiguous Verify / Proof labels
• -Discover real Python, Rust, and Go project checks and report unavailable tools explicitly
• -Keep Hosted Report and report-upload workflows retired

Why v0.0.14 shipped this way

• -Return multiple structured review findings and only block the Loop on blocking defects
• -Add project-specific review criteria through .wukong/review-policy.md
• -Add local finding feedback, review-cycle counts, and invalid reviewer-output diagnostics
• -Add an optional independent BYOK reviewer model with --review-model
• -Unify TUI and Web copy around Goal → Write → Check → Review → Fix

Why v0.0.13 shipped this way

• -Make Loop the primary workflow with real project checks and a fresh-context read-only reviewer
• -Return durable PASS, NEEDS_WORK, or ERROR results with structured blockers and stable exit codes
• -Add one Guest trial and 10 signed-in Free Loop sessions per month without charging internal checks
• -Discover nested monorepo checks, share one workspace snapshot, and reject stale gate conclusions
• -Add experimental Role Profiles and preserve legacy Goal records and --until flags

• -Add non-blocking, version-targeted TUI announcements with safe action links
• -Refine local verification output and proof diagnostics
• -Reorganize Dashboard around account and Loop status
• -Refine TUI welcome, proof actions, loop status, and footer alignment across terminal widths
• -Publish the versioned announcement endpoint, operator JSON contract, and five-minute cache policy

• -Sync with Wukong Code CLI v0.0.10: OAuth host fix and quieter startup refresh
• -Canonical docs redirect: wukong.today/docs → docs.wukong.today
• -Marketing copy aligned to proof/loop/today narrative

• -Sync with Wukong Code CLI v0.0.9: wukong today and loop-driven TUI chrome
• -Add authenticated Loop usage and account status endpoints
• -Expose CDN version endpoints /latest and /latest.json for CLI upgrade checks
• -Health API reads app version from package.json
• -Expand Vitest coverage for marketing pages and API routes

• -Sync with Wukong Code CLI v0.0.8: OAuth device-code flow and managed platform API endpoints
• -Refresh homepage copy around the AI coding + verification workflow and proof-layer positioning
• -Update public assets, favicon, and pixel mascot for the Wukong Code brand
• -Simplify dashboard with reports, subscription, and settings tabs
• -Add CLI OAuth approval page at /auth/device for device-code login

• -Point the CLI default OAuth host to wukong.today so wukong login uses the website device-code flow

• -Rebrand the TUI welcome card to Wukong Code with a compact startup layout
• -Simplify the login platform selector to OAuth only, removing API-key platform options

• -Refreshed Wukong Code homepage around the AI coding + verification workflow
• -Added the animated pixel Wukong mascot and matching favicon/social assets
• -Simplified the download page around one-line install, verify, and start commands
• -Updated footer navigation with the Today blog link
• -Tightened landing copy, terminal demo, and public launch polish

• -Rebrand the CLI as Wukong Code: npm package @wukong.today/code and public release repository mutnpc/wukong-code
• -Add TUI /verify, /scan, and /guard commands and a verification agent

• -Terminal-native AI coding agent
• -Built-in verification suite: /verify, /scan, /guard
• -Multi-model provider support (DeepSeek, Moonshot, OpenAI-compatible)
• -CLI web UI via `wukong web`
• -Native binary distribution for macOS, Linux, and Windows
• -Cloudflare Pages deployment