dsh-compaction-cache: Enabling DSH Compaction Summaries to Hit Provider Prefix Caches
The `dsh-compaction-cache` plugin for DeepSeek Harness (DSH) is designed to resolve the issue where provider prefix caches cannot be reused during compaction summaries because `reasoningEffort`, `temperature`, and `maxTokens` are missing. The plugin is mounted on the `llm/stream` stream and only takes effect for requests marked with `purpose: 'compaction'`. By mirroring routing header parameters and enforcing `maxTokens` alignment rules, it completes the summary envelope to keep the invocation environment consistent with the previous turn, significantly reducing computation costs. The plugin has zero dependencies, requires no build, and is distributed only via GitHub. All configuration is optional and enabled by default. Once active, logs will
Read Moredsh-cache-safe-tool-result
The DeepSeek Harness (DSH) plugin is designed to address a vulnerability in DSH's implicit prompt caching. DSH's "in-place" truncation can invalidate tail content and trigger re-billing. The plugin ensures the integrity of cached prefixes by truncating results before they reach the conversation surface and preventing the rewriting of already delivered results. By using the `tools/post-execute` hook and the `toolResultPruner` service, it achieves a high cache hit rate, making it well suited for scenarios that require handling large tool results while maintaining cache validity.
Read Moredsh-cache-miss: DSH Web Plugin: Yellow Single-Line Prompt on Cache Miss
`dsh-cache-miss` is a DSH web plugin maintained by developer wefio, designed to address the difficulty of promptly locating cache misses during agent development. It belongs to the "Memory" category and is implemented purely on the front end without modifying source code. Directly below the assistant response in the first conversation turn, and only when a cache miss occurs, it renders a yellow notice containing idle duration, re-billed token count, cached token count, and time to first token (TTFT). This plugin helps developers debug long-context usage and optimize prompt caching strategies.
Read MoreIntroduction to Nginx Caching: Practical Tips for Improving Website Access Speed
Nginx caching temporarily stores frequently accessed content to "trade space for time," enhancing access speed, reducing backend pressure, and saving bandwidth. It mainly includes two types: proxy caching (for static resources in reverse proxy scenarios, with origin requests to the backend) and web caching (HTTP caching, relying on the backend `Cache-Control` headers for browser local caching). Dynamic content and frequently changing content (e.g., user information, real-time data) are not recommended for caching. Configuring proxy caching requires defining paths (e.g., `proxy_cache_path`) and parameters (e.g., cache size, key rules), enabling them in `location` (e.g., `proxy_cache my_cache`), and restarting Nginx after verifying the configuration. Management involves checking cache status (logging `HIT/MISS`), clearing caches (manually deleting cache files or using the `ngx_cache_purge` module), and optimization (caching only static resources, setting `max-age` reasonably). Common issues: For cache misses, check configuration, backend headers, or permissions; for stale content, verify `Cache-Control` headers. Key points: Cache only static content, monitor hit status via logs, and prohibit caching dynamic content.
Read More