v5.3.1 — gpt-6.1-sol Codex catalog entry

gpt-6.1-sol Codex catalog entry

Scope: the openai-codex native capability (ChatGPT subscription). No pi-ai bump — the pinned 0.84.4 protocol template stays.

Highlights

• Verified Codex facts for gpt-6.1-sol (upstream @earendil-works/pi-ai@0.99.1, tarball SHA1 2945bf014fbb314bd37b919e83317560e0a8fa1d, matching the npm registry dist metadata): contextWindow=272000, maxTokens=128000, text+image, and the effort mapping off→null — the off level stays offered with an absent wire value, unlike the literal none the 0.87.1 catalog used for gpt-6-sol — plus minimal→low and the unchanged low/medium/high/xhigh/max. The public API's 1050000 context is deliberately not used.
• Runtime serves the id through the pinned-template custom-model merge: the verified-id list joins the pinned catalog so models() and resolveModel() serve one set of facts; the openai-codex route defaults include its DSH profile, so absent Settings lists materialize it; CODEX_CATALOG_REVISION bumps to codex-native-catalog/2026-10-01.1 so pending models.preview bindings rebind and models.apply keeps its digest/conflict discipline.

Install

npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.3.1

Verified / not verified

Verified in isolation (independent v5.3.0 working copy; TDD red→green; full suite 423 tests / 416 pass / 0 fail / 7 optional skips, identical to the v5.3.0 baseline; npm run check, npm pack --dry-run, git diff --check pass; the upstream verification record binds the 0.99.1 tarball SHA1 to the registry metadata). Not verified: real-account wire behavior of gpt-6.1-sol, production deployment.

v5.3.0 — 0.1.7-alpha.1 Codex catalog and model facts

0.1.7-alpha.1 Codex catalog and model facts

Scope: DSH 0.1.7-alpha.1. The installer supports the CLI versions 0.1.2-alpha.3, 0.1.2-rc.1, 0.1.5-rc.1 and 0.1.7-alpha.1; other versions are rejected.

Highlights

• Versioned catalog with two projections: a verified owner-side catalog fixes string/object mixed arrays that produced nested id objects; the DSH config projection emits only host-supported fields while the native executable-descriptor projection inherits the fixed protocol template and merges allowed model facts.
• Verified Codex facts for the new ids: contextWindow=272000, maxTokens=128000, text+image, effort mapping off→none / minimal→low (the public API's 1050000 context is deliberately not used).
• models.preview / models.apply: the persistent repair entry point for existing configurations — preview returns the catalog revision, Settings revision and diff; apply is bound to the same revision/selection/digest, idempotent, and never rewrites redacted Settings wholesale.
• retentionHints capability (source-aware-v1, versioned + capability-detected): the optional metadata-only envelope consumers may use to drop provably duplicated host-generated copies from the retained checkpoint copy; old owners keep their original policy. The 64000 estimated-token retention baseline is unchanged.

Install

npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.3.0

Verified / not verified

Verified in isolation (packaged candidate, isolated 0.1.7-alpha.1 host, paired account snapshot). Not verified: real-account calls beyond the recorded small matrix, GUI on a production profile, real-model performance, production deployment.

v5.2.2

Fix: installer now supports dsh 0.1.7-alpha.1 (was rejected). Host packages (dsh-llm) are provided by the DSH profile runtime, not npm peers — the peer declaration caused npx ERESOLVE and has been removed; compatibility is gated by the installer dsh CLI matrix. Default source identity guarded by test.

v5.2.1

Token usage plugin: ledger, provider settings and codex-native runtime.

dsh-token-usage 5.1.9

Fix the Antigravity official-allowance display for consumer accounts whose quota surfaces disagree (verified live 2026-09-13): merge quota facts across the daily/prod/sandbox surfaces instead of trusting the first answer (restores the real gemini percentages daily omits), and record the claude-*/gpt-oss now+5h full-quota placeholder echo as an honest unknown instead of a permanent 0% beside a never-converging ~5h countdown. Adds fetchedAt to usage payloads. Final review passed with reservations closed; 15 targeted quota tests, real-slot install smoke, and live differential acceptance on three consumer accounts.

npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.9

dsh-token-usage 5.1.8

0.1.5-rc.1 GUI transport: POST /api/token-usage/* (plus account-usage and subscription-antigravity). Client prefers /api then falls back to the 0.1.2 loopback channel. Does not claim 0.1.5-*.

npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.8

dsh-token-usage 5.1.7

Session import adapter for DSH 0.1.5-rc.1 (list/open) and 0.1.2 (listSnapshots/inspect). Exact installer CLI matrix: 0.1.2-rc.1, 0.1.2-alpha.3, 0.1.5-rc.1. Does not claim 0.1.5-*.

npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.7

Release 5.1.6: master-detail accounts panel with Antigravity pool rotation

Rebuild the settings provider-connections section as master-detail (B-form): one list entry per source — the Antigravity account pool collapses into a single entry with in-pool rotation (activate/remove/add-account/auto-failover) — and a detail pane reusing the shared connection controls over read-only facts (C-001 preserved).

• Credential writes route through the host api-remotes credentials namespace (replacing the nonexistent connection.api facade).
• A pending host-side Antigravity authorization survives a remount via auth.pendingLogin() + a pending-login endpoint; pool login operations carry an operation generation, so late reattach responses and abandoned polls never override a newer login or unlock a cancel in flight.
• Sidebar footer and dock rows share one window choice (watched first, else tightest); entry-summary pools carry their official windows.
• Ollama Local stays out of the accounts panel; an emptied pool keeps the add-account flow reachable.

Supersedes v5.1.5 in-tree (OAuth lifecycle + SDK stream diagnostics included). 330 tests passing. No new real-provider claim; no data migration; no core changes. Tag commit: f2134ba

5.1.5 — SDK stream diagnostic correction

DSH Accounts & Usage 5.1.5

Commit: 8527a004b26e16d94d0e8872fbd49223621021c3 (annotated tag v5.1.5).
Based on published tag v5.1.4 commit 72ecf169b15a64c6c52a0b83c73c4998d63052fc; its governance/OAuth/client fixes are preserved.
Companion consumer: dsh-codex-compaction 0.4.0.

• Record SDK stream terminal diagnostics before lease freeze: completed, failed, cancelled, timed-out.
• Omit unavailable SDK-owned wire byte/event metrics rather than manufacturing zeros or counting normalized events as wire data.
• Preserve native compact diagnostics, error identities, usage, single-flight leases, cancellation and deadlines.
• No new authentication flow, credentials, core patch, data migration or client rebuild. Pinned SDKs remain auth 0.82.1 / native 0.84.4.

Fixed asset

dsh-token-usage-5.1.5.tgz
SHA256: 870975cb1c974b033d4033e3d2cdff9937dbfd0b8d98d777b6c23e8118580ee5

The asset is the exact previously accepted 61-file candidate. Package JSON values match the source commit; other packaged source files match byte-for-byte. The client asset is unchanged from v5.1.4. Packaged documentation retains preparation-stage labels; this release page records distribution status.

Validation and limits

Sixteen new injected transport/pinned SDK diagnostic regressions have red-to-green evidence. The initial isolated source passed 327 tests; the published-v5.1.4-based candidate then passed its full owner suite again in isolated snapshots. Paired consumer integration passed 32 tests with one optional real-time test skipped. Fixed-asset loader/GUI recovery passed without new model generation.

Older consumers may omit diagnostics on newly recognized terminal phases; execution is unchanged. Pair with compaction 0.4.0 for the expanded allowlist. These terminal diagnostics do not themselves prove a Basic summary commit. The final diagnostic delta has no additional real-provider/model validation; real second-account switching remains unverified.

Formal GitHub tag acceptance — passed

Actual npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.5 and companion commands passed on both CLI 0.1.2-alpha.3 and 0.1.2-rc.1, with actual Base/Web 0.1.2-rc.1. Each row used a fresh HOME, DSH_HOME, cwd and cache, without a source override or provider credentials.

Install/repeat/status/uninstall, lifecycle blocking and unrelated-profile preservation passed. All 61 owner / 28 consumer packaged files and npx materializations matched the frozen candidates; tags retained the exact expected commits before and after. Real config dump/startup showed owner, compaction and provider entries active; the served client combo contained the exact frozen owner client bytes. Both temporary servers and fixture directories were cleaned up.

This verifies actual tag distribution, startup and client asset serving, not new browser-render/HMR or model-quality evidence. No sessions or generation were used. No production deployment or restart is included.

Accounts & Usage 5.1.4

DSH Accounts & Usage 5.1.4

Commit: 72ecf169b15a64c6c52a0b83c73c4998d63052fc

Changes / 变更

• Separate read-only connection facts from analytics and quota refresh; raw-key configuration is not represented as verified connectivity.
• Preserve the Antigravity daily → prod → sandbox quota-surface correction, without changing generation/project discovery.
• Correct subscription login cancellation, pending-login reattachment and local failed/cancelled terminal-card cleanup.
• Preserve SDK pins, the account-owned codex-runtime/v1 seam, existing request budgets and data identity. No new data migration or DSH core patch.

Verification / 验证范围

• The unchanged reviewed candidate passed 313 offline tests, including 48 client tests; the bounded terminal-state correction passed independent focused review.
• The actual 61-file release archive was checked against this commit (package.json semantically, other files byte-for-byte). Local audit reports are excluded.
• Archive preflight passed on published DSH CLI 0.1.2-alpha.3 and 0.1.2-rc.1, each with Web/Base 0.1.2-rc.1, Node 24.18.0 and pnpm 12.3.4: install/repeat install, read-only status, dependency resolution, configuration composition, real Host inventory showing the Accounts fiber active, read-only connections RPC, uninstall/repeat uninstall and synthetic-data preservation.
• Fixed-tag verification passed in another pair of fresh HOME/DSH_HOME/cwd/cache directories: genuine npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.4 default install, explicit repeat install, read-only status, all61packed-file comparison, SDK resolution, real Host active-fiber and connections RPC checks, uninstall/repeated uninstall and synthetic-data retention. Both declared CLI versions passed; every test Host exited normally.

These checks do not claim new real-provider OAuth coverage, actual-browser coverage of the final UI delta, or a formal-tag pairing test with a newly released Compaction version. Earlier real Grok OAuth checks apply to their recorded input; the additional terminal-card correction was verified offline.

Install / 安装

Fixed-tag commands (production installation remains your separate decision):

npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.4
npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.4 status
npx --yes --ignore-scripts github:shaomingbo/dsh-token-usage#v5.1.4 uninstall

The default profile is web. The installer does not restart DSH. Production installation, restart and Catalog approval are separate operator decisions; publication does not perform them.

归档内 README 保留了冻结候选时的 preparation 说明，以维持已审提交原字节；本 Release 记录其后续发布身份和验收状态。发布不等于已部署生产，也不等于 Catalog 已批准。

Artifact: dsh-token-usage-5.1.4.tgz

SHA256: 58bed1a2bd3ac4dfbc899b96224b368226147d9b6ebe149768261aee367857c8