dsh-jspace-trigger

J-Space Cognition Suite gives agents powerful capabilities such as deep reasoning, cross-file research, and architecture review, but injecting it on every turn wastes tokens, muddles model focus, and dilutes the user's real intent. dsh-jspace-trigger targets exactly this timing problem. Instead of blanket injection or forced system-prompt rewriting, it watches real user messages, runs them through a configurable rule engine (regex, keywords, message length, weighted score, explicit commands), and only when a rule fires does it append a lightweight near-field nudge to the agent inbox suggesting J-Space. When nothing matches, it stays completely silent. A built-in opt-out rule holds the highest priority, so a single phrase like "do not use j-space" overrides every other keyword trigger without exception.

The rule set spans from silent greeting suppression and explicit-command broadcast to loop-capacity nudges for repo-level multi-stage tasks and full deep-reasoning suggestions for architecture reviews or comprehensive research. Each rule independently configures action, pass, modules, matchMode, and excludePatterns for fine-grained control. An observe-only mode records hits without injecting anything, letting you validate rule hit-rates in a live workflow before enabling automatic nudges. A privacy-safe analytics funnel (rule, delivery, toolCalls, jspaceSkillLoaded) reconstructs the full path of every invocation using timestamps and event IDs, storing no user-message text or tool arguments. Skill detection and installation are strictly on-demand: the plugin never downloads J-Space automatically, only reports absence and waits for explicit user confirmation.

This plugin is built for teams who already use J-Space in a DSH workflow and want trigger timing to be auditable, reversible, and silence-capable. It coexists cleanly with Router Standard and Liangshen mode without system-prompt collisions, because it appends a near-field message rather than rewriting any preset persona or tool surface. Zero runtime npm dependencies, pure Node built-ins, and a single cordis.patch.yml to tweak make forking and customizing rules straightforward.

If you keep asking "why is the agent always mentioning J-space when I just said hello" or "why did it skip the perfect moment to suggest deep-reasoning," this is the missing middle layer: a nudge-only, never-rewriting, fully silent-by-default trigger. It does not replace your workflow; it simply makes sure the right tool is on the table at the right moment, and nowhere else.