Context & Memory
17 篇文章
Agent-Powered Codebase Q&A / Onboarding
Understanding a large or unfamiliar codebase can be a significant challenge for developers, especially when onboarding to a new project or trying to debug a complex system. Manually searching and trac…
validated-in-productionContext-Minimization Pattern
User-supplied or tainted text lingers in the conversation, enabling it to influence later generations.
emergingContext Window Anxiety Management
Models like Claude Sonnet 4.5 exhibit "context anxiety"—they become aware of approaching context window limits and proactively summarize progress or make decisive moves to close tasks, even when suffi…
emergingContext Window Auto-Compaction
Context overflow is a silent killer of agent reliability. When accumulated conversation history exceeds the model's context window: - **API errors**: Requests fail with `context_length_exceeded` or s…
validated-in-productionCurated Code Context Window
Loading **all source files** or dumping entire repositories into the agent's context overwhelms the model, introduces noise, and slows inference. Coding agents need to focus on **only the most relevan…
validated-in-productionCurated File Context Window
A coding agent often needs to reason about multiple source files, but dumping **all** files into its prompt: - Quickly exceeds token limits or inference budget. - Introduces noise: unrelated files (e…
best-practiceDynamic Context Injection
While layered configuration files provide good baseline context, agents often need specific pieces of information (e.g., contents of a particular file, output of a script, predefined complex prompt) o…
establishedEpisodic Memory Retrieval & Injection
Stateless calls make agents forget prior decisions, causing repetition and shallow reasoning.
validated-in-productionFilesystem-Based Agent State
Many agent workflows are long-running or may be interrupted (by errors, timeouts, or user intervention). Keeping all intermediate state in the model's context window is fragile and doesn't persist acr…
establishedLayered Configuration Context
AI agents require relevant context to perform effectively. Providing this context manually in every prompt is cumbersome, and a one-size-fits-all global context is often too broad or too narrow. Diffe…
establishedMemory Synthesis from Execution Logs
Individual task execution transcripts contain valuable learnings, but: - **Too specific**: "Make this button pink" isn't useful as general guidance - **Unknown relevance**: Hard to predict which lear…
emergingProactive Agent State Externalization
Modern models like Claude Sonnet 4.5 proactively attempt to externalize their state by writing summaries and notes (e.g., `CHANGELOG.md`, `SUMMARY.md`) to the file system without explicit prompting. H…
emergingProgressive Disclosure for Large Files
Large files (PDFs, DOCXs, images) overwhelm the context window when loaded naively. A 5-10MB PDF may contain only 10-20KB of relevant text/tables, but the entire file is often shoved into context, was…
emergingPrompt Caching via Exact Prefix Preservation
Long-running agent conversations with many tool calls can suffer from **quadratic performance degradation**: - **Growing JSON payloads**: Each iteration sends the entire conversation history to the A…
emergingSelf-Identity Accumulation
AI agents lack continuous memory across sessions. Each conversation starts from zero, causing: - **Lost familiarity**: The agent doesn't remember user preferences, goals, or working patterns - **Repe…
emergingSemantic Context Filtering Pattern
Raw data sources are too verbose and noisy for effective LLM consumption. Full representations include invisible elements, implementation details, and irrelevant information that bloat context and con…
emergingWorking Memory via TodoWrite
During complex multi-step tasks, AI agents lose track of: - What tasks are pending, in progress, or completed - Which tasks are blocked by dependencies - Verification steps that need to run - Next act…
emerging