[FEATURE] Subject-Tagged Context Capsules.

Resolved 💬 3 comments Opened May 19, 2026 by johnwbest Closed Jun 18, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Real engineering and development projects do not unfold as single linear threads. They consist of many interleaved subject areas -- each with its own design decisions, code artifacts, deferred questions, and open issues. An IoT product development project, for example, might simultaneously involve:
• Device firmware -- state machines, sensor drivers, hardware abstraction
• WiFi connectivity -- provisioning, roaming policy, reconnection strategy
• Web UI -- browser-based dashboards, WebSocket real-time updates
• Cloud integration -- MQTT, REST APIs, Home Assistant, data pipeline
• OTA updates -- firmware delivery, filesystem updates, rollback strategy
• Security -- credential storage, encrypted communication, access control
• Multi-device fleet -- shared infrastructure, naming, mesh networking

Each of these is a substantive subject with its own context, vocabulary, decisions made, and decisions deferred. Today they exist as an undifferentiated mass of conversation within a single project.

Many subjects are directly relevant to other projects. A WiFi mesh architecture discussion is equally relevant to a handheld RF instrument project, a sensor node project, and a home automation hub project. Today, accessing that knowledge in a new project requires one of three unsatisfactory approaches:
• Re-explaining the entire subject from scratch -- wasting time and losing nuance
• Sharing the entire parent project -- overwhelming the new project with irrelevant context
• Manually copying relevant exchanges -- fragile, becomes stale, lossy

The core gap: There is no mechanism to share a specific intellectual thread between projects while leaving the rest of the parent project in place.

subject_tagging_proposal(2).docx

The basic problem this feature solves is having to re-establish relevant context for a complex project. Other problems are also solved, such as getting team members up to speed with what has, and has not been accomplished in a project and perhaps more importantly, how decisions within a project were influenced and evolved.

Proposed Solution

As I work naturally through a project, I take Claude into "tangents" related subroutines, even hardware issues. I would tag these various sub-aspects of the project with something like @Subject (subject mnemonic) while Claude identifies context changes from conversational language and inserts AI generated tags. These sub-aspects of a project are called "context capsules". When switching to a new project, or working through a different open project, I would associate whatever context capsules were relevant to the project. This would be "making associations", and would be done on-the-fly with something like @Refer (name of context capsule). Claude would load that capsule (not the entirety of the project in which it was tagged), as well as other context capsules, and have that previous work at the ready as the current project progresses.

Alternative Solutions

I stared by using the "Projects" feature in Claude, but found it lacking. See above.

Priority

High - Significant impact on productivity

Feature Category

Interactive mode (TUI)

Use Case Example

I'm working on an IoT edge device, some sort of uP based interface to a sensor or an actuator. I decide to make it controllable via the WWW, using the uP's WiFi capability. Part of the WiFi aspect is passively monitoring signal strength in order to alert of dubious connectivity. A week later, I open a completely different project, but it also has WiFi connectivity. I would have Claude refer to the discussion tagged in the previous project as a "context capsule" which would bring in all the discussion within the tagged segment of the previous project.

Additional Context

_No response_

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗