feat(zed): add per-project attribution

[Vlaaaaaaad]

Summary

TL;DR: add best-effort per-project attribution to Zed. Proof-of-concept follow-up to #480 / #594

Warning

This code was written by LLMs, under the supervision of a so-called developer that spends more time in Pages&Keynote than in an IDE!

Zed Agent Threads can happen in 2 types of workspaces:

• single-folder workspaces, the "classic" IDE-in-a-folder. In this case, the folder name is used as the project name, with some Zed-matching cleanup
• multi-folder workspaces, where in the same Zed window there are multiple folders/projects opened. Think codeburn and homebrew-codeburn and agentseal in the same Zed window. In this case, the project name in Codeburn output will be agentseal, codeburn, homebrew-codeburn (folder names, sorted lexicographically)

This LLM-written, human-validated (beware: dumb human) pull request implements the above. Zed Agent Threads without folder metadata remain in the legacy catch-all zed bucket.

Testing

I tested this on one of my devices and I can confirm the single zed is split into multiple projects:

Before:

Top projects
┌───────────┬───────────┬────────────┐
│  Project  │     Cost  │  Sessions  │
├───────────┼───────────┼────────────┤
│  zed      │  $278.04  │     1,160  │
└───────────┴───────────┴────────────┘

After (keep in mind I also did a bunch of work in zed since I do have some tiny contributions to Zed):

Top projects
┌───────────────────────────────────────────────────────────────┬───────────┬────────────┐
│  Project                                                      │     Cost  │  Sessions  │
├───────────────────────────────────────────────────────────────┼───────────┼────────────┤
│  zed                                                          │  $179.82  │       373  │
│  otel-experiment                                              │   $42.91  │        88  │
│  infra                                                        │   $18.56  │        35  │
│  temp-xxx-xxxx                                                │   $13.68  │       225  │
│  Corporate                                                    │    $4.96  │       145  │
│  xxx-xxxx-xxx-github                                          │    $4.45  │         1  │
│  dotfiles                                                     │    $3.80  │        10  │
│  zed, REDACTED.txt, ALSO-REDACTED.txt                         │    $2.88  │        29  │
│  xxx-xxxx-xxx-blogs                                           │    $1.69  │         2  │
│  temp-xxx-xxxx, temp-xxxx, temp-xxxxx-xxxxxxxx, temp-xxxxxxx  │    $1.42  │        25  │
└───────────────────────────────────────────────────────────────┴───────────┴────────────┘

This was not a complex setup and I did not have a bunch of work on this machine (no ACP threads, relatively low number of sessions, minimal worktree usage, etc). More testing would be a very good idea!

Notes

Since I am not at all familiar with the codeburn codebase and patterns, this is more of a proof-of-concept than an actual PR. Please feel free to close this or just treat it as somewhat-relevant-documentation/example for a real implementation — I will not be offended in any way, shape, or form!

Relevant Zed references:

• Zed thread database schema and folder metadata
• Zed PathList serialization format
• Zed Project group identity and display naming
• Zed Sidebar thread grouping
• Zed Agent Thread save payload and visible workspace paths

[ozymandiashh]

Blocking correctness issue on this exact head:

src/providers/zed.ts:127-136 turns the stable path into a basename display label, while downstream aggregation groups strictly by project and keeps the first projectPath. Two sessions rooted at /Users/alice/repo and /Users/bob/repo both emit project: "repo"; they are then merged under the first path. This corrupts per-project totals, filters, cross-provider grouping, and persisted daily attribution. Multi-root display labels can collide in the same way.

Please keep a stable normalized full-path or root-set identity separate from the display label, and add an aggregation-level regression test with identical basenames/display labels and distinct paths.

The Zed suite (14/14), typecheck, and diff check pass, but the identity collision is a data-integrity blocker and this head has no CI checks.

[Vlaaaaaaad]

@ozymandiashh thank you for your patience and review! Unfortunately, fixing that issue seems to require wider changes that go waaaay beyond some small additions in the Zed provider: introducing stable normalized identities based on full paths/root sets and propagating that across parsing, session caching, cross-provider merging, daily aggregation, filtering, dashboard/export/history rendering, and to some path-sensitive consumers. The newly-added e67fe7ec9c8032b987fb6a3f0191619d395dad2e commit is allegedly a step in that direction.

As somebody that does not know this codebase (and that does not have the bandwidth to do a deep-dive to become familiar with this codebase 😞), I do not feel comfortable doing those wider changes, even with the fanciest LLM assistance, so I am closing this PR. My apologies for the annoyance and churn! Anybody with codeburn knowledge or with more bandwidth is more than welcome to pick this up!

[ozymandiashh]

Mulțumesc Vlad, mă mai uit pe problemă și când am timp poate mă apuc de ea.

[Vlaaaaaaad]

@ozymandiashh no rush and no worries! Also, if you do pick this up, let me know if you need me to test anything as I am more than happy to clone a PR and run some tests on my machines with Zed usage data!