kenz1117/dsh-ui-usage-billing

English · 中文 · Quick start · Compatibility

---

Demo GIF

The eight dashboard tabs in turn (Overview / Queue / Billing / Usage / Trends / Agents / Rates / Settings):

✨ Why this plugin

Most billing plugins stop at "token count × unit price". dsh-ui-usage-billing turns billing into a reconcilable ledger pipeline — real usage, live prices, and peak/off-peak awareness that follows the model.

More capability, one place

Cost attribution	Automation & entry points
Drill down through cache hits, official / third-party spend, workspaces, Sessions and relays; agent chains, TTFT, generation speed and cost spikes stay traceable.	Queue composer drafts or JSON batches for off-peak hours; jump from a Session row to the host, or open the dashboard with /usage.
Pricing & models	Experience & privacy
Uncatalogued models are explicit, with aliases and custom prices to fill the gaps; the optional read-only usage_stats tool is off by default.	Independent Chinese / English and ¥ / $ / € controls; no chart library or external CDN, and no system-prompt injection by default.

📊 Dashboard

Sidebar and live feedback

• Trigger card: today / week / month headline, 7-day trend and daily gauge; choose a full card, compact icon or hidden entry, with a quick look on hover.
• Cost bar and alerts: live turn / Session cost, model-aware peak and quota warnings, plus optional system notifications for tier switches and Session completion.

Eight sections, one consistent ledger

Overview / Queue / Billing / Usage / Trends / Agents / Rates / Settings. The hero and KPIs cover today, 7 days, week, month, all time and custom dates; trends, heatmaps, budgets, rates and searchable settings all read the same real ledger.

• Queue: price-band countdown and queued / running / needs-review / completed tasks; up to 50 JSON items per batch, validated atomically. Locks and leases prevent duplicate submission across instances sharing one DSH_HOME. With “Continue tasks in the source Session” enabled, a task queued from a Session continues inside that Session at the next off-peak window (keeping its context and workspace grouping); the card shows “Will continue in Session X” before the run and links to it. A deleted, busy or archived source falls back to a new Session with the reason on the card. New Sessions land in their workspace instead of being orphaned in the sidebar.
• Agents: a parent conversation → sub-agent tree with own cost and chain total including every descendant.
• Settings: full-text search, global and per-model budgets, peak alerts, custom prices, native host settings and plugin information; a hidden sidebar entry can be restored from host Settings.

💰 Billing engine

• Channel attribution: official, direct, relay and unknown routes stay separate; route and model aliases repair historical attribution.
• Time-based pricing: live models.dev data, 247 built-in models across 25 vendors, segmented peak pricing and live FX; later changes never rewrite history.
• Real-cost overrides: set relay-specific prices for the same model and fill gaps for uncatalogued models.
• Budget controls: global and per-model budgets, 50 / 80 / 100% alerts, balance warnings and last-40-turn spike attribution.

🔌 Subscriptions & balance

• Subscription quotas: detect Kimi, Z.ai, OpenCode Go, MiniMax, Claude, CommandCode and more, with remaining quota and reset times; subscription-channel model cost is 0.
• Official balances: DeepSeek, Kimi, StepFun, SiliconFlow, xAI, Zhipu GLM and Tencent Cloud TokenHub, with days remaining estimated from the last 7 days.
• Relays and reconciliation: merge keys by site, read New API / Sub2API rolling quotas, and cross-check custom endpoints or official balance deltas against the ledger.

Detection, endpoints, credentials and troubleshooting for every channel live in the adapter matrix.

📈 Usage visualizations

• Drill down: inspect Sessions, projects, workspaces and sites; month / half-year / year heatmaps show cost or token intensity.
• Analyze: daily tokens by structure or model, tool-call ranking, TTFT / P50 / P90, generation speed and cost composition.
• Trend: 7 / 30-day cost and calls, the last 40 turns, peak/off-peak share and off-peak savings.
• Export: daily / Session / site CSV, daily-token CSV and full JSON; no chart library or external CDN.

🛡️ Robustness & privacy

• Real aggregation: incremental Session-log reads, per-session corruption tolerance, snapshots and a durable ledger; uncatalogued and estimated prices stay explicit.
• Read-only by default: no system-prompt injection, Session creation or Prompt submission; only explicit enqueueing writes a new Session, still under host permissions.
• Loopback security: peer-socket and Host checks on HTTP, plus Origin, Content-Type and body limits on writes; CSV formula-injection protection.
• Complete experience: independent language and three-currency preferences, keyboard focus, and provider health expressed through shape, color and text.

🚀 Quick start

One codebase currently supports DSH 0.1.2-alpha.1 through 0.2.2-rc.1. Stable and preview releases share the same compatibility matrix: stable releases use npm latest, while previews use alpha. npm latest currently points to the stable 1.8.3 release (host 0.2.2-rc.1 compatibility, the composer-hover flicker fix and the Hunyuan Hy4 GA catalog expansion); alpha points to 1.8.3-alpha.3 (queue tasks continuing inside their source Session), and the next preview starts at 1.8.3-alpha.4.

dsh plugin add requires an explicit --profile. Pinning an exact version is recommended so a package manager's release cooldown cannot resolve a different version. Install the current stable release with:

dsh plugin --profile web add npm:@kenz1117/dsh-ui-usage-billing@1.8.3

Default installations can also use @latest. Check the host with dsh --version or npm ls -g @deepseek-ai/dsh.

Alternatively, add it to the host cordis.patch.yml by hand:

- insert:
    - id: ui-usage-billing
      name: '@kenz1117/dsh-ui-usage-billing'

After the host starts, the billing entry appears above the sidebar Settings. No extra configuration is needed; when sessionPersistence is available it aggregates real usage automatically.

🛒 Install from the DSH Community Market

This plugin also publishes a catalog entry using the Market's Path A standard source. Paste the manifest URL below into Settings → Plugins → Plugin Store → Source management → Add source, then browse and install with one click:

https://kenz1117.github.io/dsh-ui-usage-billing/catalog-source.json

Market-hosted installs resolve an exact registry release from npm (the Host resolves the install target from npm latest; the version listed in the catalog is display-only). The catalog files live in catalog/ in this repository and are validated in CI before being published by GitHub Pages.

📚 Advanced reference

⚙️ How it works

The plugin has a server side and a browser side:

Browser                                   Server (Node)
  │                                        │
  ├─ GET /api/billing/usage-stats ────────▶ ├─ sessionPersistence walks persisted session logs
  │                                        ├─ attributes a call to its preceding request/header model
  │                                        ├─ buckets tokens by cache hit / miss
  │                                        └─ estimates cost (CNY) from the live rate table
  ├─ GET /api/billing/pricing ────────────▶ ├─ models.dev model prices + Tencent Finance FX (tiered fallback)
  ├─ GET /api/billing/balance ────────────▶ ├─ DeepSeek official balance API (credentials seam)
  ├─ llm.models health probe ─────────────▶ └─ returns aggregated stats JSON
  └─ renders the dashboard

• Server (src/index.ts): injects webServer, sessionPersistence and credentials, and registers loopback endpoints for usage, pricing, balances, subscriptions, relays, notification claims, the tool switch and GET/POST /api/billing/task-queue. The aggregator incrementally folds Session logs into a durable usage ledger. The queue probes optional sessionController / workspaceRegistry capabilities through ctx.get() rather than startup injection, so missing capabilities never block the host or dashboard.
• Browser (src/client/): requests the endpoints above to render the dashboard and probes each provider connection via llm.models. Until real data arrives it shows an all-zero empty snapshot, never fabricated samples.

💡 Billing details

The rate table (src/client/pricing.ts) stores each model in its native currency: domestic providers enter CNY directly, overseas providers enter USD. Cost is computed and displayed in CNY uniformly — USD models convert via the live rate, domestic models never pass through a rate. At startup the server fetches the live rate and model prices (src/pricing-fetch.ts): USD→CNY prefers the Tencent Finance quote (keyless, reachable in China), falling back to open.er-api then the built-in default; it then refreshes every 6 hours, and the rate-table modal shows a "today's rate" marker plus live / built-in badge. The display currency follows the user: switching ¥ / $ converts each per-1M-token unit price via convertUnitPrice at the live rate (falling back to the native currency when the rate is unavailable).

cost (CNY) = (missInput × p_input + cacheHit × p_cacheHit + output × p_output) / 10⁶
           —— prices in native currency; USD models convert at the live USD → CNY rate

input in the stats is total input (cacheHit + cacheMiss); estimation splits it into hit / miss to avoid double counting. Models with two-band billing are mixed by DEFAULT_PEAK_SHARE (default 0.5); weekends (Beijing Sat/Sun) are charged at the off-peak rate all day.

Month-to-date forecast: month-to-date spend + last-7-calendar-day pace × the days still left in the month. Only the days that have not happened yet are extrapolated, so a quiet recent week never inflates the number — earlier versions averaged over the month's recorded days and multiplied by the whole month length, which inflated sparsely used months several times over.

Supported models (2026-11 lineup, OpenAI-compatible series, 247 models)

The full catalog lives in the Rates tab and in BUILTIN_MODEL_CATALOG of src/builtin-catalog.ts (246 active + 1 retired model kept for historical repricing); representative models per vendor below.

Provider	Representatives
DeepSeek	V4.1 Flash, V4 Pro (3 models; priced per official change boundary: base tier → peak/off-peak v1 → weekend off-peak)
Zhipu AI	GLM-5.3, GLM-5.2 (17 models)
Aliyun	Qwen3.8 Max, Qwen3-Coder 480B (27 models)
Doubao	Doubao Seed-2.1 Pro, Doubao Seed-2.0 Pro (13 models)
Moonshot	Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, Kimi K3, Kimi K2.7 Code (18 models)
Xiaomi	MiMo V2.6 Pro (6 models)
MiniMax	MiniMax-M3, MiniMax-M2.7 (8 models)
Unisound	U2 Flash, Unisound U2 (3 models)
Baidu	ERNIE-5.1, ERNIE-4.5 300B (2 models)
Tencent	Hunyuan Hy4 (limited 50% promo), Hunyuan Hy4 Preview, Hunyuan Hy3 (9 models)
Anthropic	Claude Opus 5, Claude Sonnet 5 (14 models)
Mistral AI	Mistral Large 3, Mistral Small 4 (30 models)
Cohere	Command A, Command R (7 models)
OpenAI	GPT-6 Astra, GPT-5.6 Sol / Terra / Luna (39 models)
Google	Gemini 3.1 Pro, Gemini 3.6 Flash (25 models)
xAI	Grok 4.7, Grok 4.20 (8 models)
Meta	Llama 4 Maverick, Muse Spark 1.3 (8 models)
Meituan	LongCat 2.0 (estimated)
ModelBest	MiniCPM-V 4.5 (estimated)
Xiaohongshu	Dots3-Note Preview (estimated)
01.AI	Yi-Lightning
StepFun	Step 3.7 Flash, Step 5 Preview (3 models)
iFlytek	Spark 4.0 Ultra (plan-based)¹
SenseTime	SenseNova 6.5 (beta)¹
Baichuan	Baichuan M3-Plus
Other	Unified fallback pricing for uncatalogued models (cost 0)

¹ iFlytek and SenseTime have not published per-token prices — the table shows estimates; cost is 0 when these models go through a subscription channel (coding / token plan / opencode), and recalibrates automatically when official pricing is published. Subscription channels align with pi-ai built-in providers (kimi-coding, zai-coding-cn, opencode, opencode-go, qwen/xiaomi token-plan regional variants), overridable via subscriptionProviders.

To add a model: append an entry to BUILTIN_MODEL_CATALOG in src/builtin-catalog.ts (key aliases live in BUILTIN_MODEL_KEY_ALIASES in the same file; the catalog is built only into the node half and served to the client from the host via /api/billing/pricing).

🔌 HTTP API

The public HTTP endpoints and field definitions are documented in source, including usage, pricing, balances, subscriptions, relay quotas, GET/POST /api/billing/task-queue and GET /api/billing/task-queue-summary (see src/index.ts, src/aggregate.ts, src/relay.ts, src/routes/task-queue.ts). The queue returns host capability, workspaces, official models and durable tasks; mutations are enqueue / enqueueMany, remove-not-started and clear-finished. The summary returns only the price-band state and queued count, never Prompts. All endpoints accept loopback requests only; queue writes additionally require a loopback Origin, JSON Content-Type and a body no larger than 10 MiB.

⚙️ Configuration

Field	Default	Description
statsPath	unset	Absolute path to a fallback .dsh-usage-stats.json (used when sessionPersistence is unavailable)
ledgerPath	<harness home>/.dsh-usage-ledger.json	Independent durable ledger path; stores folded metrics only (no message bodies or session titles), so deletion does not erase recorded usage. The default root follows the host harness home (DSH_HOME env first, falling back to ~/.dsh), so isolated environments never share ledger data (issue #52)
taskQueuePath	<harness home>/.dsh-usage-task-queue.json	Absolute off-peak queue path; stores explicitly submitted Prompts, workspaces, models and execution state. Writes are atomic with mode 0600 and a .bak backup; separate DSH_HOME environments never share a queue
balanceApiKeyEnv	DEEPSEEK_API_KEY	Credential ref for the DeepSeek balance query; only used as a fallback when llm-pi-ai has no apiKeyEnv for deepseek
subscriptionProviders	built-in channel list (incl. tencent-token-plan)	Subscription (coding / token plan) provider id list — tokens counted, cost 0; aligned with the subscription-card recognition. Note it replaces the built-in list rather than appending to it — when customizing, list the built-in entries too, otherwise removed subscription channels fall back to pay-as-you-go
routeAliases	not set	Historical route aliases (old provider route name -> current route): renamed/deleted routes relocate into their channel instead of the "unknown" bucket. Example: { "deepseek-official": "tencent" }
modelKeyAliases	not set	User model aliases (log model id -> billing catalog key, value must be an existing MODEL_CATALOG key): bind uncatalogued ids without waiting for a release. Example: { "hy4-preview": "hunyuan" }
monthlyBudget	unset	Default monthly budget (CNY); sent with usage-stats as the budget bar's initial amount (user UI settings take precedence and persist locally)
lowBalanceThreshold	50	Low-balance alert threshold (CNY); sent with usage-stats, alerts once a day when any provider's CNY balance is below it
trustedHosts	unset	Extra host names allowed through the Host-header check when DSH runs behind a reverse proxy (e.g. ['llm.example.com']); empty by default = behaviour identical to previous releases. Consulted only after the peer-socket loopback check has passed, so it widens only the second, defence-in-depth layer; exact host match, case- and port-insensitive, with no suffix or wildcard semantics
subscriptionPlans	auto-detect	Subscription quota adapter whitelist ({ provider, baseUrl?, region? }); when unset, auto-detects all subscription providers from llm-pi-ai (queries those with a quota API, marks the rest)
declaredEndpoints	unset	Declared endpoints ({ displayName, origin, path, fields?, windows?, raw? }): self-declare balance/quota interfaces for providers absent from the built-in table, writing only dot-paths ("where the number is") with no expressions; the request URL is built from the matched same-origin provider's origin and safety bounds (single-slash absolute path, GET only, reject cross-origin redirects, response-size/timeout caps, credentials only from the matched provider's own apiKeyEnv) are enforced by src/declarative.ts
reconcilePath	<harness home>/.dsh-usage-reconcile.json	Balance-delta reconcile baseline path (default root also follows DSH_HOME / ~/.dsh); cross-checks the official (DeepSeek-direct only) balance change against the local ledger's official-channel cost for the day, and flags a drift above the threshold (0.3 CNY and >15%); top-ups / grants / currency changes reset the baseline instead of alerting
searchCallEstimateCny	0.02	Per-call cost estimate for web-search requests (web/deepseek-search-llm-request, which log no usage events) in CNY; set to 0 to disable the estimate (calls still counted, not billed)

🛠 Development

Requirements: Node.js ^22.19 || >=24, pnpm.

pnpm install
pnpm test          # unit + API-contract mock + UI component tests (vitest + jsdom)
pnpm lint          # discipline guards: types / logging / empty catch / CSS↔TSX class parity
pnpm run build     # tsc -b && tsdown, produces lib/index.js and lib/client.js

For on-machine debugging: dsh plugin --profile web add "link:<absolute path to this repo>" points the host at this working tree — afterwards edit code → pnpm run build → restart the host, no npm publish needed (see docs/DEV-MODE.md for the loading chain). HTTP-contract e2e: pnpm test:e2e (needs a local dev host running). Engineering conventions (two-line release strategy, host-upgrade SOP, acceptance checklists) live in AGENTS.md.

📦 Release

This standalone npm package maintains one main codebase. Stable releases use non-prerelease versions and npm latest; previews use prerelease versions and npm alpha. npm latest currently points to the stable 1.8.3 release, while alpha points to 1.8.3-alpha.3; see COMPATIBILITY.md.

pnpm test && pnpm run build   # full tests + build before release
npm publish --access public
npm dist-tag add @kenz1117/dsh-ui-usage-billing@<version> alpha   # preview
npm dist-tag add @kenz1117/dsh-ui-usage-billing@<version> latest  # stable

The host discovers the browser side automatically via the dsh.client declaration (platform: web) and the exports["./client"] bundle in package.json — no registry registration needed. The compatibility matrix is declared per version from actual verification (dsh.compatibility.dshReleases); run node scripts/check-compat.mjs before and after publishing to re-verify.

🔐 Permissions & Compatibility (DSH STORE)

• Permission level: high: reads durable Session logs, calls vendor / subscription / balance / pricing APIs, reads apiKeyEnv through the credentials seam, and writes the ledger, snapshots and task queue under the harness home. The plugin itself has no command-execution / shell capability; only explicit enqueueing creates a host Session and submits a Prompt, while Agent tools remain governed by the host's existing permissions and approvals.
• Update channel: user-reviewed: with file / network / credential capabilities, DSH STORE requires local manual confirmation on every install; review the repo, pinned commit, lifecycle scripts, and impact scope before installing.
• Compatibility: stable and preview releases from the current main share one matrix covering DSH 0.1.2-alpha.1 through 0.2.2-rc.1, with 34 per-version declarations. The current stable release is 1.8.3 (npm latest; host 0.2.2-rc.1 compatibility, the composer-hover flicker fix and the Hunyuan Hy4 GA catalog expansion); the preview channel is 1.8.3-alpha.3 (npm alpha, queue tasks continuing inside their source Session). Release channels and monitoring are documented in COMPATIBILITY.md. Node.js ^22.19.0 || >=24.0.0.
• Lifecycle: no preinstall / install / postinstall / prepare (ready on install).
• No impersonation: adds only its own entry id ui-usage-billing; @deepseek-ai/dsh-* packages are peerDependencies only (no reinstall / replace / shadowing of official components); the package uses the third-party namespace @kenz1117/*.
• Build artifacts: runtime files lib/* and cordis.patch.yml are committed at the pinned commit and declared in files.
• Source anchor: sources are locked to a pinned commit on the GitHub default branch and traceable. DSH STORE automation re-reads the default-branch HEAD roughly every 8 hours as the new pinned commit and decides re-review by SemVer change.

🤖 Model Experience

None by default. The plugin exposes no tool to the model out of the box: the "inject usage query" setting is off by default, and while it is off the model can neither see a tool nor read any data from this plugin.

After the user explicitly enables that setting, exactly one read-only tool is registered:

Tool	Parameters	Behaviour
usage_stats	range: today / month / session / all / bySite / relay	Queries the usage and estimated cost this plugin already aggregated (CNY, estimated from the official price catalog — not a bill); session requires an agent session context

Boundaries: the tool reads only this plugin's own aggregates — never Session bodies or Prompts — and never writes back to the log or touches Session KV cache. The plugin injects no system prompt. The off-peak queue is not a model tool: it is a separate, explicitly authorized action whose Prompt becomes a user message in a host Session (its own new Session by default, or the source Session when “Continue tasks in the source Session” is on), so it must not be confused with this read-only tool boundary. The tool consumes context on every request, which is why it stays off by default.

⚠️ Known Limitations and Deferred Work

• The off-peak queue is still a preview: it accepts only time-of-day models on the host's built-in deepseek-official route; relays and custom routes are excluded. There is no automatic pause, cancellation or retry. With source-Session reuse on, the task keeps the source Session's current model (the plugin never switches it — that would change your global default model) and only continues there while the Session is idle; otherwise it falls back to a new Session with the reason recorded. After host shutdown, a dropped result stream or lease expiry, the task becomes "Needs review"; open its Session and decide manually whether to enqueue a new task.
• Balance queries cover DeepSeek / Moonshot (Kimi) / StepFun / SiliconFlow / xAI / Zhipu GLM (Z.ai CN region) / Tencent Cloud TokenHub (token-plan quota): the first six use a standard Bearer API key; Tencent Cloud uses the cloud-API key pair (<SecretId>:<SecretKey>, TC3-signed management API). Other providers expose no public balance API or need non-Bearer auth (Xiaomi MiMo via console Cookie, SenseTime via AccessKey signing, MiniMax/Doubao via quota or AK/SK), so they currently show "not configured"; the extension point is src/balance.ts (add a querier per provider balance API).
• Relay quota depends on upstream private schemas: New API / Sub2API interface fields are not public, so an unreadable station is labeled "no quota" rather than fabricating an amount; if a station's response fields differ, extend the parsers in src/relay.ts. An "unknown route" means that route exists neither in the current provider config nor in the host llm registry (deleted config, logs older than the install, …); historical call data is not lost — re-adding the same-named route restores attribution automatically. Cost in the unknown-route bucket is always 0 (issue #82): the spending party cannot be verified — same posture as uncatalogued models and subscription channels — while tokens are still counted, and the panel shows "—" for the group cost with an explanatory note. Channels registered by another plugin through ctx.llm.registerAdapter (e.g. CommandCode) are not unknown routes: the plugin discovers them from the host llm registry and bills them in the direct:<provider> bucket at the model's official catalog price (issue #91); the registry is read through ctx.get('llm') (no inject requirement), and an unreadable or empty registry warns once (panel numbers and the unknown policy are unchanged) so this class of failure is never silent again.
• Overspend notifications rely on the browser Notification API: when permission is denied or the platform lacks support, only the in-UI red-pulse fallback remains — no host-level notification channel; notifications are capped at once per day.
• Session-row navigation depends on a host channel: clicking a session title in the panel opens that conversation in the host (both generations of the host's session-selection channel are probed); on very old hosts, or when the session was deleted / is not in the current session list, the UI shows a readable notice instead of silently doing nothing. The panel shows the top 20 by default and the "view all" full-screen viewer is capped at 100 rows.
• Cost is a catalog estimate: models without published per-token pricing (iFlytek, SenseTime, Xiaomi) use estimates (feature-list footnote ¹); official billing is authoritative.
• The ledger starts at its first successful aggregation: sessions permanently deleted before the upgrade and absent from the old snapshot cannot be recovered. Manually deleting .dsh-usage-ledger.json and its .bak clears independently retained history. Only calls successfully observed by this plugin are retained.

❤️ Contributors

• @lucagiftzek — rate-table CSS polish (PR #59), trustedHosts reverse-proxy support (PR #60), currency preference persistence (PR #58), rate-table search (PR #67), peak/off-peak indicator dots with pin-to-composer (PR #68), the EUR display currency with language decoupling (PR #70), the first pricing fetch moved ahead of the warmup replay (PR #55), evidence-based models.dev listings (PR #56), and hardcoded UI strings routed through locale dictionaries (PR #61)
• @hwangjunjie — Tencent Cloud TokenHub / Token Plan subscription quota adaptation (src/tc3.ts), provider-first channel grouping with gateway badges, the routeAliases / modelKeyAliases config (PR #35), and subscription badges decided per channel on the day × model × site cells (PR #83)
• @ciphoo — MiniMax CN Token Plan quota support (PR #5), minimax-cn subscription key wiring (PR #12), a concurrent-ledger-write MoveFileExW EPERM fix on Windows (PR #11)
• @fabulousyuann-tech — durable ledger that retains usage after session deletion (PR #8)
• @hi-fangj — hover tooltip with the exact per-day token breakdown on the token daily chart (PR #21), the live-cost capsule toggle (PR #22), the per-model view and legend focus for the token daily chart (PR #23), and the per-model comparison curve in the perf panel (PR #24)
• @aizhimoran — peak/off-peak tiering for Chinese statutory holidays and shifted workdays: DeepSeek usage-based pricing bills holidays entirely at the off-peak rate (official 2026-09-19), with a 10-day countdown scan window (issue #73, PR #74)
• @kidbird — wired up the international Z.ai Coding Plan (zai-coding) subscription quota, pinning each route to its own region with split key fields (PR #79)
• @yuxu4468-hash — cataloging for bare relay-exposed model ids: Grok 4.5/4.6, GPT-5.5, GPT-6.1 Sol, Claude Opus 4.7/4.8 and Fable 5 (PR #87), plus the GPT image models gpt-image-2 / gpt-image-2.5-sunburst (PR #88)
• @JinruiWang — workbuddy adapter channel pricing and the fold-before-live-catalog dual-caliber fix: the known-channel list, three lazy-reprice gates (ledger, cache hit, incremental cursor) and the stable-stamp reproduction test (issue #90)
• @gitfangj — sync.sh brace-variable compatibility under the macOS default bash 3.2

📄 License

MIT © 2026 KenZ (kenz1117)