7.1 KiB
description, kind
| description | kind |
|---|---|
| All-messages LLM session-title provider for users and maintainers choosing a title strategy or debugging automatic title generation. | package-reference |
@deepseek-ai/dsh-session-title-all-prompts-llm
English | 中文
Summary
dsh-session-title-all-prompts-llm summarizes every eligible human message through ctx.llm as an optional ctx.sessionTitle provider. It registers the all-prompts cadence and starts a new revision after each new human prompt, using seeded history and child-session prompts. A newer revision aborts and supersedes older work, and even a provider that ignores cancellation cannot commit stale output. It owns both its initial-generation prompt and its revision prompt, and shares only the execution module from dsh-session-title-llm with the first-prompt provider. Automatic behavior and configuration come first; the implementation is a small provider strategy over that executor.
Table of Contents
- Use this package
- Understand the implementation
- Further Exploration
- Model Experience
- Known Limitations and Deferred Work
- Dev Note
Use this package
Mount this plugin beside the title service when a session should be retitled as it grows, so the title keeps representing the whole conversation. It requires its own targetWords and targetCjkCharacters plus the shared execution controls, all with no defaults.
When titles are generated
A new revision starts after each new eligible human prompt, including prompts in child sessions; the generation folds all eligible messages through the current revision, seeded history included. A newer revision aborts and supersedes older work, so a stale completion can never commit. An automatic failure — including input over maxInputBytes, which fails instead of truncating history — warns and keeps the prior title; ctx.sessionTitle.refresh() is the explicit retry.
Once a provider-generated title exists, each request includes it and asks the model to preserve it exactly while it still describes the main topic or task. Follow-up questions, same-topic details, acknowledgements such as “thanks”, and requests to continue do not by themselves justify rewording. The model is instructed to update the title only when a material change or expansion of the main topic or task makes it inaccurate. A fallback title still receives ordinary initial generation; every eligible prompt still schedules a revision.
Configuration
The plugin requires targetWords and targetCjkCharacters plus the shared execution controls: maxInputBytes, maxOutputTokens, timeoutMs, and the optional paired provider/model route. Omit both to inherit the exact route from each current logged main request, or set both to route title generation independently. The generated configuration catalog is the exhaustive source for every accepted field.
Failures and recovery
If the final framed aggregate prompt exceeds maxInputBytes, the request fails instead of truncating history; automatic use warns and keeps the prior title, and only an explicit refresh() retries. Automatic work adds no tokens and no latency to the main agent request.
Understand the implementation
Implementation internals — click to expand
This section explains the plugin's shape; the observable behavior is fully covered in Use this package.
Design concept
A small provider strategy: it owns both the initial-generation prompt and the revision prompt that anchors currentTitle, parses the assembled response into a title, and delegates only route, bounds, cancellation, logging, and assembly to the shared execution module.
Source map
| File | Role |
|---|---|
src/index.ts |
Plugin entry: config schema, initial and revision prompts, output parsing, and direct provider registration |
Scheduling
The title service schedules automatic work: for the all-prompts cadence, every new eligible user message starts a revision, and a newer revision supersedes older work; the provider call begins after the exact main-request route is logged.
Further Exploration
Read these pages when the provider contract is not enough. They move from the shared execution module to the alternative cadence and the service it plugs into.
- Shared LLM execution module — the execution module this provider uses.
- First-message title provider — the cadence that titles a session once from its first prompt.
- Session title service — fallback behavior, rename, refresh, and provider registration.
- Session package map — adjacent persistence, projection, title, and telemetry packages.
Model Experience
All-messages title request
What the model sees
The title model receives all eligible human messages through the current revision, in log order with exact seqs, including seeded history. When an accepted provider-generated title exists, this provider adds it as currentTitle with its own preservation instructions; fallback, user, and absent titles take its initial-generation path. Both the messages and the current title count toward maxInputBytes.
Token effect
One auxiliary request may follow every new eligible prompt, bounded per request by maxInputBytes and maxOutputTokens; explicit refreshes may add calls. It selects the route's least supported reasoning effort, or leaves effort unspecified when the route exposes none. The main agent request gains zero tokens.
KV Cache effect
No main-request invalidation. Auxiliary input grows or changes after each prompt, so provider-specific cache reuse ends at the first changed JSON token.
Known Limitations and Deferred Work
These limits define how the provider treats long and heterogeneous sessions. They are current package constraints.
- No summarization-of-summaries — input overflow retains the prior title; this provider has no summarization-of-summaries or retention policy for very long sessions.
- Messages are treated equally — it includes every eligible human message without configurable weighting or filtering.
- Model-dependent stability — title preservation is a model instruction, not a deterministic semantic check; the model may still change an adequate title.
Dev Note
Working context for maintainers — click to expand
None.