fix(providers): align usage homes, discovery and executable diagnostics

[rynfar]

Provider usage omitted custom accounts, reconnecting clients triggered redundant provider scans, and missing-Codex errors incorrectly referred only to PATH. This adopts the corresponding upstream fixes while preserving Pylon’s provider ownership, explicit refresh, account configuration and runtime identity.

• Usage includes configured Claude, Codex and Grok account homes, including disabled accounts with history, and deduplicates canonical paths. Relative configured Claude homes use Pylon’s launch-path resolver.
• Config subscriptions return current state without launching another full refresh; managed startup discovery, background policy, settings changes and explicit refresh remain active.
• Codex startup diagnostics name the configured executable and the server setting to correct it.
• WSL discovery is already present in Pylon’s staged and mounted Node-based runtimes. The standalone-runtime patch is not applicable; regression tests protect existing PATH forwarding and stdin bootstrap authentication. No unused runtime probe is added.

Sources: #11485 2db675aeffd9cb1e8b5ad76ddd018433b45b02e9, #11811 7931227977ca3e6f3354a63467caa634ab79796f, #11345 2a264adc6f6c65f98d4273acaa1af567eca83f2a, and covered #11741 8b1ea4dd2465f3cf3cfa02d6878beaa1ec22e71a. Original author/coauthor trailers are retained, including maria, Bil0000 and shivamhwp. WSL’s upstream analysis is credited to Andrew Johnson. Part of #611; bounded head 6d1d549441be84f19696ab59ed7e2fbf305280d4, unchanged review cursor.

Validation:

• Seven focused files: 374 passing tests, 10 platform skips. After the combined run exposed macOS canonical-path assumptions in the usage harness, all 10 usage tests passed with corrected assertions and scan synchronization; the other six files remained unchanged.
• Scoped server typecheck, desktop typecheck and lint passed (existing server lint warnings remain).
• Separate Antigravity adversarial reviews for each source. Fixed the relative-Claude-home defect and replaced the inapplicable WSL implementation with coverage. Codex classification/privacy findings were checked and dismissed as unchanged behavior or already-authorized settings visibility.
• Isolated browser: unchanged custom-account history shows 0 tokens before and 12,000 after; the account remains disabled. Missing-Codex diagnostics show the configured path and settings guidance. Screenshots attached below.
• No wire schema or migration change. Server behavior is shared across local/remote web, desktop and mobile. Relative environment home paths remain unsupported as documented; real Windows/WSL execution was unavailable on this macOS host.

Implementation and adversarial review used Gemini 3.8 Flash High through Pylon’s Antigravity delegation; integration and verification used GPT-6 through Codex.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
pylon-marketing	Ready	Preview	Sep 18, 2026 12:33am UTC

[rynfar]

Isolated browser evidence, using the same account history and settings before and after. Usage changes from 0 to 12,000 tokens (10,000 input + 2,000 output) for a disabled custom Claude account. Codex diagnostics now identify the configured missing executable and the server setting to correct it. Images in order: Codex before, Codex after, usage before, usage after.

[rynfar]

Separate Antigravity adversarial reviews are complete for all four sources, with parent integration review of the combined diff.

• Usage: reviewed 1408af6ad24d51a167877ed723091eea770681a3; fixed the confirmed relative-Claude-home mismatch in 2e5d146c40c7d7e8d98967827c2d6a916c02925b (integrated as 70cd53c720164a21ef5d44198bd136f921940e8e). The reviewer then approved the canonical-path test corrections through c3e3d6250775c387789c9ae3f43c9eee5767ecdb, independently passing 10 UsageService tests and all 92 tests across eight usage files. Codex shadow-home behavior was checked against actual launch semantics and correctly scans shared history; the suggested upstream concern did not apply.
• Subscription refresh: reviewed 424ec3ed0561cf27a8f3cd2bc7a7d6c2b211c64d; confirmed startup discovery, explicit refresh, driver health policy and settings-driven updates remain owned independently of subscribers. Integrated regression coverage also checks initial snapshots, remote variants and changed-config streams.
• Codex diagnostics: reviewed f9cdcb42023efac4c32f67578dcd0c404ba898a8. Initial concerns about installed/error classification and binary-path visibility were rechecked: they are unchanged behavior and already exposed through authorized server settings, respectively. No introduced blocker remained. The added regression exercises the real missing-executable registry path.
• WSL: rejected the initial unused standalone-probe port. The reviewer independently approved replacement 5ffade9cd49395be4a8d5219f15f3454de815f15, which changes only regression tests for Pylon's actual staged/mounted launch paths. Real WSL was unavailable; this is not presented as a Windows end-to-end run.

The reviewed production contents are unchanged in this PR. Final focused coverage is 374 passing tests plus 10 platform skips; server types and scoped lint passed, with existing warnings retained. Before/after screenshots are attached in the preceding evidence comment.

[rynfar]

Final head: 16f064f88af4bddf3e7aebe67ab2609213325620, rebased onto merged #612 (62c4f818ea301754a3856c86745f3345b0bed413). The rebase conflict was confined to the decision index; both records and PR links are retained. Compared the provider implementation/tests against the reviewed pre-rebase head: no changes. The only subsequent edit recommends absolute environment paths in the user guide, avoiding a provider-dependent tilde-expansion promise. Existing review, local tests, and uploaded screenshots remain applicable; CI is running on this final head.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	14.0 KiB	14.0 KiB	−7 B (−0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	−12 B (−0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+5 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	9	10	+1 (+11.1%)	21	✅
Claude	Total thread wire	14.0 KiB	14.0 KiB	−7 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	+1 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−8 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.9 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	9	9	0 (0.0%)	21	✅

Baseline: 62c4f81 · PR result: 16f064f · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 115.6 KiB
• Claude decoded thread snapshot: 116.3 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.