dsh-self-memory

Every new conversation, your agent starts like a day-one intern: the bugs it fixed last week, the parameters it tuned, the workarounds it discovered, all gone. dsh-self-memory turns those experiences into a structured knowledge base. Problems encountered, multiple candidate solutions, script paths, and API call traces are recorded automatically. Next time a similar situation arises, the plugin retrieves the most relevant historical solutions via inverted-index search with precondition-aware ranking and injects them into the current context. The agent then tries a solution: on success it marks a hit, on failure it records the root cause, and if nothing matches it solves the problem from scratch and writes the result back. The entire loop happens without manual intervention.

A few design choices are worth noting. Purely local with zero runtime dependencies (Node built-in modules only), keeping all conversation data on your machine. Recall injection goes through user messages rather than the system prompt, sidestepping presets where a complete persona silently swallows plugin contributions. The recording layer ships with an inverted index (candidate-set retrieval then weighted scoring above 100 records), keyword synonym normalization, and precondition-aware sorting that flags solutions as inapplicable, conflicting, or deprecated. Memory is injected as evidence, fenced and source-labeled, and the capture pipeline excludes all plugin-originated text to prevent self-pollution.

Who it is for: DSH developers tired of agents rediscovering the same fixes every session; users who prefer offline, self-contained, no-server, no-data-egress solutions; anyone who wants a lightweight plugin rather than a full memory infrastructure stack.