dsh-cache-warmer v0.1.7

Improvements and safeguards

• Add native llm-pi-ai warming capabilities for compatible API-key and OAuth routes, with exact adapter, model, snapshot and credential-authority checks.
• Require separate, default-off best-effort consent for Codex and signed-in Responses. Client cancellation is not a server-side output, billing or quota guarantee.
• Invalidate captured requests and cache evidence on credential/account changes; preserve native replay, reasoning controls, tool schemas and session affinity without executing tools or modifying transcripts.
• Display configured lifetimes on observation-only routes, improve warming status, and conservatively reduce savings credit after partial cache hits.
• Preserve OpenRouter parameter, fallback, fee and catalog-price guards; validate the final serialized payload and reject dynamic unknown-billing router aliases before credentials or pricing. OpenRouter serializers without equivalent routing controls remain observation-only.
• Add reproducible standalone dependency resolution and expanded native loopback safety tests.

Installation and compatibility

dsh plugin --profile web add "https://github.com/yoggu/dsh-cache-warmer.git#v0.1.7"

Use the profile you actually run. Restart DSH and refresh the browser after replacing the installed version. Native warming is reviewed against llm-pi-ai 0.2.0-rc.2 and pi-ai 0.99.1; unsupported versions fail closed. Development dependency overrides do not upgrade the adapter owned by the running Harness.

Warming remains off by default and requires a genuine request, an enabled model with a configured lifetime, per-chat opt-in, cache evidence and passing cost checks. Live cache savings are not established by offline verification.

Validation

• 170 offline/loopback tests passed in the existing multi-peer checkout.
• Clean npm ci verification: 168 passed, with 2 cross-peer-layout tests explicitly skipped because dependencies were deduplicated.
• Package contents and whitespace checks passed. No real provider inference requests were made by the verification suite.

Full changelog: v0.1.6...v0.1.7

dsh-cache-warmer v0.1.6

Patch release

• Update serializer and catalog validation for the shipped pi-ai 0.99.1, including typed chat model keys and per-model pricing caps.
• Resolve the serializer-owner catalog and calculateCost; revalidate the pinned catalog before background sends and fail closed for unreviewed versions or catalogs.
• Safety change: Codex and direct DeepSeek routes are observation-only. Background warming is limited to reviewed OpenRouter Chat Completions routes; historical opt-ins do not bypass this gate.
• Remove automatic Codex lifetime assumptions. Warming remains off by default and requires a genuine request, explicit model lifetime/enablement, and per-chat opt-in. Offline/loopback validation does not establish live cache savings.

Installation

dsh plugin --profile web add 'https://github.com/yoggu/dsh-cache-warmer.git#v0.1.6'

Use the profile you actually run. Keep linked source checkouts in place while installed.

Validation

• 128 offline tests passed; no live provider, account, Telegram, or restart operation was performed.
• Package contents and whitespace checks passed.

Full changelog: v0.1.5...v0.1.6

dsh-cache-warmer v0.1.5

DSH 0.2 compatibility

DeepSeek Harness 0.2.0-rc.1 enforces a plugin's peerDependencies against the running runtime and skips any plugin whose declared ranges do not cover it. The previous 0.1.x-only ranges therefore disabled this plugin entirely on 0.2.0-rc.1:

dsh: skipping profile bundle "dsh-cache-warmer": Plugin dsh-cache-warmer@<previous> is incompatible with dsh 0.2.0-rc.1

This release widens the declared @deepseek-ai/dsh-* peer ranges to cover ^0.2.0-rc.1 while keeping 0.1.x support, so the plugin loads on both 0.1.7-rc.2 and 0.2.0-rc.1.

Changes

• peerDependencies for @deepseek-ai/dsh-* now accept ^0.1.x and ^0.2.0-rc.1
• engines.dsh declares the tested DSH range
• README install instructions pinned to v0.1.5

Install

dsh plugin --profile web add 'https://github.com/yoggu/dsh-cache-warmer.git#v0.1.5'

Verification

Booted a DSH 0.2.0-rc.1 web profile containing this plugin: no compatibility skip is reported, the plugin activates, and its client bundle is delivered in the web boot manifest.

Known issue

npm test reports 8 pre-existing failures in tests/openrouter.test.js and tests/wire.test.js. They are not related to DSH 0.2: they reproduce unchanged on the previous release commit, and the DSH packages they exercise (@deepseek-ai/dsh-llm-pi-ai, @earendil-works/pi-ai@0.85.1) are identical between 0.1.7-rc.2 and 0.2.0-rc.1. The price assertions hard-code catalog values that have since drifted (~deepseek/deepseek-v4-flash-latest now costs 0.04998/0.09996).

dsh-cache-warmer v0.1.1

Adds provider/model discovery and per-model cache policies, plus expanded OpenRouter catalog support with model-derived routing safeguards.

v0.1.0 — Cache time and cost-aware warming

Initial public release.

• Estimated Cache time countdown and ring in the composer statistics bar.
• Opt-in, Pi-style cost-aware warming for Codex account routes and the reviewed OpenRouter DeepSeek route.
• Native-style settings, English/Chinese UI, and mobile layout.
• Background requests never append chat messages or execute returned tool calls.
• Persistent warming usage accounting, finite activity windows, and foreground cancellation.

Install:

dsh plugin --profile web add 'https://github.com/yoggu/dsh-cache-warmer.git#v0.1.0'\n```\n\nTested with DeepSeek Harness 0.1.7-rc.2. All 30 offline tests pass, including loopback requests through the installed provider adapters.\n\nCache lifetimes and Codex API-equivalent costs are estimates. Codex does not enforce a hard output-token cap. Warming consumes API or subscription usage and is disabled by default. End-to-end scheduled cache benefit has not yet been verified against live providers. See the README for requirements and limitations.