Wukong Code v0.0.16

Wukong Code 0.0.16 — Converge Safely

Released: 2026-07-21

Distribution: free/BYOK native binaries for macOS, Linux, and Windows.

0.0.16 makes the existing Loop converge around a fixed user-owned target. It
does not add payment, hosted execution, managed model credits, PR publishing,
Team features, or a multi-reviewer quorum.

Loop convergence

• A new Loop freezes the user objective and optional completion criterion in an
  append-only local contract. Writer, reviewer, and strategist cannot rewrite it.
• Each review receives the previous blocking findings and must mark them
  resolved, still_open, or replaced_by before it can pass.
• Blocking findings must map to a frozen criterion. Stable keys and fingerprints
  exclude prose, absolute paths, and line numbers.
• Two consecutive identical blocker sets trigger one fresh-context, read-only
  strategist. If the same blocker remains afterward, Loop pauses with
  NEEDS_WORK/no_progress instead of spending the remaining iterations blindly.
• Invalid structured reviewer output retries once on the same model. An explicit
  reviewer model that remains invalid or unavailable falls back to the writer
  model once. Cancellation never triggers fallback.

Permission Mode v2

• Manual remains the interactive confirmation mode.
• Auto no longer asks questions or silently grants them. It allows ordinary
  autonomous work but blocks workspace-external writes, sensitive targets,
  high-risk Bash, and unclassified custom/MCP tools. An active Loop pauses
  with NEEDS_WORK/permission_required.
• YOLO is an explicit dangerous approval bypass, not an unlimited capability.
  User deny rules, safety hooks, Plan guard, profile allowlists, and read-only
  child-agent ceilings still apply.
• Headless prompts default to guarded Auto. Dangerous headless execution requires
  --yolo --yes; --yes by itself no longer selects YOLO. The legacy
  --auto-approve flag emits a migration warning.

Resume, Roles, and TUI

• Welcome may show one opt-outable current-workspace suggestion from Codex,
  Claude Code, or Cursor, bounded to 30 days and 50 candidates per source.
• The Wukong session picker can search user prompts from the latest 100 sessions
  using bounded, cancellable local reads. System, tool, and synthetic content is
  not indexed.
• Role Profiles can declare input, output, and capability contracts. These
  contracts only narrow the base profile and tool set.
• Planner, reviewer, strategist, and read-only explorer built-ins cannot inherit
  Bash or mutating tools from a more permissive parent.
• The footer uses one compact Task Plane count; /tasks remains the place for
  status, duration, cancellation, and detailed output.

Compatibility and privacy

• Existing Goal/session/Role records, LoopResult states, exit codes, completion
  outbox, and lifecycle APIs remain compatible.
• Legacy strategy records are normalized in memory without rewriting history.
• New convergence, permission, capability-ceiling, and Resume-suggestion metrics
  contain only bounded categories and counts. They exclude objectives, blocker
  keys, paths, commands, custom tool names, session IDs, prompts, and transcripts.

Validation completed so far

• Agent-core full suite: 3,307 tests passed, one todo, one skipped at the recorded
  development checkpoint.
• CLI full suite: 2,367 tests passed, three skipped at the recorded development
  checkpoint.
• Real Node, Python, Rust, and Go fixtures each cover a failing project check,
  an actual file fix, and a subsequent PASS.
• Codex, Claude Code, and Cursor each cover suggestion, source picker, editable
  Loop objective, Goal creation, and terminal PASS in the TUI integration suite.
• The bundled CLI, docs site, and local macOS arm64 SEA binary build and pass
  their smoke checks.
• The public release contains six native ZIP files and six matching SHA-256
  files. A fresh 0.0.16 install and a real native 0.0.15 → 0.0.16
  wukong upgrade both verify the checksum and run as 0.0.16.
• The public update manifest, version-targeted announcement, download page,
  changelog, Today article, and production health endpoint are live and report
  0.0.16.

The release remains free and BYOK. Payment, managed model credits, PR Gate
publishing, Team features, hosted execution, and multi-reviewer quorum remain
outside 0.0.16.

Wukong Code v0.0.15

What's Changed

Recoverable Loops

• Resume an active Loop after a CLI restart, inspect its details, or end it as cancelled without inventing a Gate result.
• Keep private local run records and retry failed lifecycle completion through a bounded local outbox.
• Re-read the workspace, Git state, project checks, and review policy before resumed work continues.

Faster First Success

• Run objective, provider, model, workspace, Goal, and project-check preflight before reserving Loop quota.
• Show a shorter first-run surface and a compact terminal result with Gate, iterations, the primary blocker, and the next action.
• Treat projects with no checks as NEEDS_WORK with setup guidance; unavailable tools are not reported as failed code.

Real Lifecycle Status

• Distinguish active, stale, completed, cancelled, and abandoned Loop runs in Dashboard and protected metrics.
• Preserve compatibility with older completion payloads while making terminal retries idempotent.

Install or Upgrade

curl -fsSL https://wukong.today/install.sh | sh
wukong upgrade

Native builds and SHA-256 checksums are available for macOS, Linux, and Windows on x64 and arm64.

Wukong Code 0.0.15 remains free and bring-your-own-key. Payment, GitHub PR automation, Hosted Report, managed model credits, and new Agent sources are not part of this release.

Wukong Code v0.0.14

What's Changed

Wukong Code v0.0.14 makes unfinished AI coding work easier to continue and makes the Loop result easier to understand.

Cross-Agent Resume

• Add /resume codex, /resume claude, and /resume cursor in the TUI.
• Discover local sessions from Codex, Claude Code, and Cursor without modifying the source session.
• Keep bare /resume focused on Wukong sessions.
• Let users choose Continue in Wukong or Start /loop with an editable objective.
• Never replay historical tool calls, permissions, or old test results.

Clearer Loop Gate

• Replace the ambiguous Verify ✗ · Proof BLOCKED footer with one actionable Gate result.
• Distinguish failed checks, checks that did not run, missing project checks, risk-review requirements, and reviewer blockers.
• Preserve compact alignment and truncation in narrow terminals.

Real Project Checks

• Expand nested workspace and monorepo check discovery.
• Add broader Python, Rust, and Go project-check discovery.
• Report missing tools as unavailable instead of treating them as failed code.
• Keep legacy verify-pass, scan-clean, and judge-pass inputs compatible while mapping them to the unified Gate.

Product Cleanup

• Keep verify, scan, and proof as advanced local Loop diagnostics rather than separate products or quotas.
• Remove Hosted Report upload, hosted history, hosted-report quota, and public report sharing.
• Keep v0.0.14 free and bring-your-own-key. Payments and real Pro entitlements remain planned for v0.1.0 or later.

Install or Upgrade

curl -fsSL https://wukong.today/install.sh | sh

Existing v0.0.13 native installations on macOS and Linux can run:

wukong upgrade

Native binaries are included for macOS, Linux, and Windows on x64 and ARM64.

Wukong Code v0.0.13

Wukong Code 0.0.13 — Review until ready

Status: released as v0.0.13. The release tag triggers native builds for the six supported platform targets.

Upgrade from 0.0.12

Native 0.0.12 binaries still contain the old manual-upgrade behavior. Upgrade once through the verified installer:

curl -fsSL https://wukong.today/install.sh | sh

The installer selects the matching 0.0.13 release asset, verifies its SHA-256 file, and replaces the binary in the existing install directory. Starting with 0.0.13, macOS and Linux native installs can use wukong upgrade for later releases.

What's Changed

0.0.13 makes the review step useful inside the Loop. Wukong now returns multiple structured findings, separates blocking defects from non-blocking advice, and feeds concrete blockers into the next coding iteration.

Structured review findings

• Each finding has a stable ID, severity, category, evidence, file location, next action, confidence, and validation status.
• Only blocking findings produce NEEDS_WORK; important and note findings remain visible without creating an endless repair loop.
• Invalid reviewer output is retried once, counted locally, and becomes ERROR when no trustworthy result can be produced.
• Legacy 0.0.12 single-blocker reviewer output remains readable.

Project review policy

Run:

wukong review init

This creates .wukong/review-policy.md. The policy becomes part of the workspace fingerprint and reviewer input, so a verdict cannot be reused after the policy changes. Project policy is treated as untrusted review criteria and never grants the reviewer write or shell permissions.

Local feedback and review statistics

Record whether a finding was useful:

wukong review feedback <finding-id> accept
wukong review feedback <finding-id> false-positive
wukong review feedback <finding-id> fixed
wukong review feedback <finding-id> unclear
wukong review stats

The same actions are available in the TUI through /review. Feedback, optional reasons, reviewer-cycle counts, and invalid-output counts stay in ~/.wukong/review/ by default. Wukong hides quality percentages until at least 10 findings have feedback.

Independent reviewer model

Use another configured model for the read-only reviewer:

wukong loop "finish the API change" --review-model reviewer

The writer model remains unchanged. 0.0.13 stays free and bring-your-own-key; no managed reviewer credits or paid entitlement are introduced.

Loop-first TUI and Web copy

• Idle TUI chrome now points users to /loop <goal> instead of showing zero-value Today, verify, scan, and proof counters.
• Loop phases use Write, Check, Review, and Fix language.
• The website, download flow, Pricing, Dashboard, Onboarding, terminal preview, SEO, and email copy use the same Goal → Write → Check → Review → Fix model.
• Existing hosted reports remain a compatibility surface, not the primary 0.0.13 workflow.

Deliberately deferred

• Payment, Checkout, subscription management, and real Pro entitlements
• Automatic policy mutation or cloud reviewer improvers
• GitHub PR comments and status checks
• Team policy synchronization and managed reviewer credits

These remain 0.1.0+ work. Pro and Team stay Coming Soon until real Loop activation and review value are validated.

Wukong Code v0.0.12

Wukong Code 0.0.12 — Loop until verified

What's Changed

0.0.12 turns Wukong Code from a collection of verification commands into one Loop-first workflow. Give Wukong an objective; it writes, runs the repository's real checks, scans the change, asks a fresh-context read-only reviewer, and continues only when the result is NEEDS_WORK.

One result users can act on

• PASS — the deterministic gate and reviewer passed; process exit code 0.
• NEEDS_WORK — the change has actionable blockers or reached its iteration limit; exit code 1.
• ERROR — Wukong could not produce a trustworthy conclusion; exit code 2.
• Authentication or quota rejection uses exit code 3; user interruption uses 130.

Each result carries a reason code, compact check results, structured blockers, and a next action. TUI, headless JSON output, SDK snapshots, and persisted Goal records use the same contract.

Stronger delivery gate

• A single workspace snapshot is shared by verify, scan, and proof.
• If checks mutate the workspace, Wukong retries once and then returns workspace_changed instead of presenting stale proof.
• Project-authorized commands in .wukong/checks.json take precedence.
• Nested npm, pnpm, yarn, and bun packages are discovered and executed in stable order.
• Source changes without a discovered test/typecheck command return tests_missing; documentation-only changes may continue to review.
• The reviewer runs with a fresh context and a read-only profile. It cannot override deterministic blockers or edit the workspace.

Free Loop usage

• Guest: one local Loop trial, up to two iterations.
• Signed-in Free: 10 Loop sessions per calendar month, up to five iterations per session.
• A soft Pro-coming-soon message starts with the third monthly session; the hard limit applies after the tenth.
• Only the top-level Loop counts. Internal verify, scan, and proof checks are never counted separately.
• Start requests are idempotent, completion records preserve the final three-state result, and temporary quota-service failures fall back to a local five-iteration safety limit.

Role Profiles

Role Profiles are experimental in 0.0.12. Enable experimental.role_profiles, then use wukong roles list, wukong --role <name>, or /transform <name>. A role may narrow tools and change the writer prompt/model; it cannot expand the base permission boundary. Role marketplaces, team synchronization, and a separately selectable reviewer model are not part of this release.

Compatibility

0.0.11 Goal records remain readable. Legacy verify-pass, scan-clean, and judge-pass records and flags are accepted and migrated in memory to proof-pass; original record files are not rewritten. No legacy wire enum is removed in 0.0.12.

Deliberately deferred

• Hosted Evidence schema v2 and Loop evidence upload
• --review-model and managed reviewer credits
• Background/cloud Loop, PR gates, team policies, and billing lifecycle

These remain 0.0.13+ work so 0.0.12 can validate whether users repeatedly use the core Loop before Wukong adds more product surface.

Wukong Code v0.0.11

What's Changed

Wukong Code 0.0.11 closes the first hosted-evidence UX loop: generate proof locally, keep it private in hosted history by default, and create a public reviewer link only when explicitly requested.

Added

• Non-blocking TUI announcements fetched from https://wukong.today/cli/announcements.json. Messages can target client versions and UTC time windows, use once / cooldown / always frequency, and link only to approved Wukong destinations.
• Explicit public hosted-report sharing through wukong verify --upload --public, wukong scan --upload --public, wukong proof --upload --public, and wukong report upload <path> --public.
• A hosted-evidence next step after wukong proof, showing the private-history and public-share commands instead of leaving upload undiscoverable.
• Hosted-report plan, usage, quota, and Dashboard destination in logged-in wukong today output.

Changed

• Hosted report uploads are private by default. A public reviewer link is created only with the explicit --public flag, and --public without --upload returns a usage error.
• Refined TUI welcome, loop status, proof actions, status tracks, and footer alignment for narrow and wide terminals.
• Reworked the Wukong Dashboard around real report history, evidence status, hosted quota, and task-oriented next actions instead of generic SaaS metrics.
• Added public documentation for software updates, automatic-update controls, announcement privacy, operator JSON workflow, and version compatibility.

Fixed

• Hosted report visibility is now explicit end to end, so a public upload reliably returns a share link while ordinary uploads stay private.
• Hosted usage and announcement network failures remain fail-open and never block the local briefing or TUI startup.
• The public announcement endpoint now ships as cacheable static JSON with a five-minute cache policy and no authentication requirement.

Install

curl -fsSL https://wukong.today/install.sh | sh

Full documentation: https://docs.wukong.today

Wukong Code v0.0.10

What's Changed

Fixed

• OAuth device-code login now targets https://wukong.today (the previous auth.wukong.today host had no DNS).
• Accept legacy oauth/ credential key prefixes so persisted oauth_host is not dropped during login.
• Startup no longer warns when refreshing managed:wukong models while logged out.
• CLI --help documentation link points to https://docs.wukong.today.

Full Changelog: v0.0.9...v0.0.10

Wukong Code v0.0.9

What's Changed

Added

• wukong today Daily Proof Briefing: local verify/scan/proof stats, streak, focus, and next actions. Run wukong today or /today.
• M1 Hosted Reports: upload redacted verify/scan/proof reports to wukong.today with wukong verify --upload, wukong scan --upload, wukong proof --upload, or wukong report upload <path>.
• Hosted report list and share page at https://wukong.today/r/<id>.
• Per-tier monthly hosted report quota (founder/pro) with usage tracking.
• Loop-driven TUI chrome: Claude Code-style borderless editor, loop-phase heartbeat, and verify/proof conclusion tracks.

Fixed

• Chrome-state no longer treats goal blocked as a verify failure; hydrate keeps the newest stats events; warning/judge mapping matches CLI semantics.
• Nix/SEA Node 26.4.0 build and verification workspace wiring.

Full Changelog: https://github.com/mutnpc/wukong-code/blob/main/CHANGELOG.md

Wukong Code v0.0.8

What's Changed

Added

• Milestone 0.5 AI Coding Loop control plane: /loop for local interval pacing and /schedule for cloud cron jobs.
• /loop --every cron-style scheduler pacing for recurring agent loops.
• Verification layer merge proof and stop conditions for verify-skill.

Changed

• Point CLI default OAuth host and API base URL to wukong.today.
• Regenerate web favicon from the Wukong logo.

Fixed

• Clear all lint warnings and fix latent bugs uncovered by lint (verification double-resolve, jitter hash safety, oauth Set typing).
• Fix Vitest #/cli/* alias resolution so the CLI test suite runs green.

Full Changelog: https://github.com/mutnpc/wukong-code/blob/main/CHANGELOG.md

Wukong Code v0.0.7

What's Changed

• chore(release): bump @wukong.today/code to v0.0.7