join together with agentsview?

[wesm]

hi! I saw this linked from ryoppippi/ccusage#938 (comment)

I've been building https://github.com/wesm/agentsview the last couple of months and now support a ton of agents in addition to Claude Code. I'm wondering if you'd be interested in contributing whatever is missing from this project there? Totally respect wanting to make your own project though!

[jakeefr]

Hey @wesm! Appreciate you reaching out and big fan of what you've built with agentsview.

I think you're right that there's meaningful overlap, but also a real difference in what each tool is trying to do. Agentsview is a session browser and analytics layer - PRISM is more of a diagnostic tool. "What happened and what did it cost" vs "what went wrong and how do I fix it." Complementary more than competitive.

Rather than a full merge, I'd like to propose an integration path. PRISM already supports --json output, so agentsview could consume PRISM's health scores, adherence analysis, and advisor recommendations as a data source for Claude Code sessions specifically. You'd get richer diagnostics for Claude Code users without either project losing its identity.

Happy to scope out what a clean integration would look like if you're interested.
Would a JSON schema doc for PRISM's output be a useful starting point?

[wesm]

Sounds interesting! I think I’m mostly interested in helping consolidate a system of record for the parsed session data itself. Session insights are produced by analyzing the session data. So the ideal layering would be for prism’s analysis and insights to run on top of the agentsview database (which already supports the full complexity of Claude Code subagents and DAGs so maintaining that a bunch of places is less ideal). The parsing is done in Go and already very fast, so it should offer performance benefits for you also.

[jakeefr]

That layering makes total sense and aligns with how I'm thinking about it too. PRISM's value is purely in the analysis and advisory layer, not the parsing, so building on top of agentsview's database is a clean separation.

I did a full schema analysis against PRISM's analyzer before responding and most of the mapping is straightforward. One blocker worth flagging before we go further though:

PRISM's Context Hygiene dimension and parts of Token Efficiency are built entirely around compact_boundary system records. It looks like agentsview filters those out during parsing. Would you be open to storing those as system messages (is_system=1) or adding an is_compact_boundary flag? That's the one thing that can't be derived after the fact.

Two smaller gaps also worth mentioning:

• is_sidechain is a per-message boolean PRISM uses for token efficiency scoring, doesn't look like it's stored currently
• sessions.project vs raw cwd path - PRISM needs the raw path to auto-locate CLAUDE.md on disk

If those are addressable on your end the rest of the integration is straightforward. Happy to share the full schema mapping doc if useful.

[wesm]

Sounds like reasonable improvements. I will work on improvements to the storage schema in agentsview to hopefully provide what you need, I will revert back after I've had a chance to cook on that for a little bit.

[jakeefr]

Sounds good, looking forward to it!