fix(providers): preserve explicit launch modes and qualified model IDs

[rynfar]

Claude permission launch flags were passed alongside a derived SDK permission mode, and qualified Codex model identifiers could lose their catalog family/default or be replaced during text generation. Fold explicit Claude flags into its SDK mode and compare Codex model families without changing provider-owned dispatch identifiers.

Cycle #497, frozen upstream head 4a4c6dd2adc350a68ba18bb28b24b5a7e4660dab. Adopted sources:

• 5735693d4acb949c2b0daf1219b4f62845115435 (#11026), Claude launch argument precedence.
• a92161a05040c3fcaaa479f1b60d2d45cd368dd4 (#9921), qualified model IDs across driver catalog, manifest classification/defaults and text generation.

Pylon keeps custom model identities, provider admission, Prime and existing manifest validation/cache behavior. Qualified default resolution also brings in the applyManifestDefault helper dependency from 06336460c9988f29c71e839c4c9c840c4552e077 (#9348); it moves the existing default flag and aliases only when the live catalog carries the requested default. The existing manifest-default reader is retained without duplication.

Validation: focused ClaudeAdapter, CodexProvider, ModelManifest and CodexTextGeneration tests passed (part of 247 tests across these four files and the independently changed GitVcsDriverCore suite); t3 and @t3tools/shared typechecks passed. Scoped lint passed. Peer adversarial review and final-head CI are required before landing.

Implemented with GPT-6 in 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 12, 2026 6:20am UTC

[rynfar]

Independent adversarial review completed against final head 7e0e41345c: no actionable findings. Checked exact Codex IDs before non-custom family fallback, custom model classification, live-catalog default and alias movement, and Claude flag/SDK permission consistency. Focused test run completed: 247 passing tests across ClaudeAdapter, CodexProvider, ModelManifest, CodexTextGeneration and the separately ported GitVcsDriverCore suite; server/shared types and scoped lint passed. Final-head CI is still required.

[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	14.1 KiB	13.9 KiB	−153 B (−1.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	+15 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.9 KiB	6.7 KiB	−168 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.8 KiB	58.0 KiB	−910 B (−1.5%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	14.1 KiB	14.0 KiB	−53 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	+4 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.9 KiB	6.8 KiB	−57 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.7 KiB	59.6 KiB	−88 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: f3402be · PR result: 7e0e413 · 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.