refactor: remove Pylon pair mode and thread delegation

[rynfar]

Removes Pylon's pair mode and cross-provider thread delegation so agents can use their native subagents without Pylon's executor protocol or restrictions. Existing executor/child conversations and files remain available as ordinary threads.

• Removes web/mobile controls, settings, MCP tools, automatic parent wakeups, executor lifecycle handling, and the internal follow-through command/projection queries.
• Deletes the shipped delegation skill and its agent metadata, removes injected pair/delegation instructions and Claude/Codex/Prime restrictions, and supersedes obsolete durable guidance.
• Uses selective inverse changes rather than reverting the feature commits wholesale. Preserves the mobile opaque draft-ID fix and whole-value model-settings replacement from the original delegation commits, plus later Codex startup instructions, browser/device guidance, provider recovery, checkpoint/shared-workspace rewind safety, sidebar search, and composer scrolling fixes.

Adversarial review covered commit provenance, stale capabilities, old settings, remaining instruction/skill references, native-agent UI, and unrelated changes in touched files. Regression tests verify the complete MCP tool registration has no pair/delegation tools, stale pair labels cannot restrict Codex/Claude, provider startup does not clamp agent depth, and old settings decode without losing retained preferences.

Validation: 804 backend/shared tests, 745 client tests, 115 additional checkpoint/Codex safety tests, 16 MCP registration tests, and 173 Claude tests passed (some suites overlap). Scoped typechecks passed for server, web, mobile, desktop, contracts, client-runtime, and shared. Scoped lint passed with two existing unused-import warnings in ChatComposer. Browser checks use separate isolated before/after instances; no provider turns or live data mutations.

CI: all checks passed on 1b7b4eba10, including all three server shards, the general test suite, typecheck/build, and Linux/Windows packaged smoke.

Upgrade note: start a new thread for former pairs to avoid pair-era instructions/session state; update server and clients. Existing history is preserved. Mobile controls were checked by tests/typecheck, not on a simulator. No upstream orchestration implementation is included.

UI evidence from isolated before/after instances:

Surface	Before	After
Integration settings
Composer

Model: GPT-6 · Harness: Codex in Pylon.

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
pylon-marketing	Ignored	Preview	Sep 19, 2026 7:28pm UTC

[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	13.9 KiB	14.0 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	14.0 KiB	14.0 KiB	+2 B (+0.0%)	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	6.8 KiB	6.8 KiB	+6 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: eea1ff3 · PR result: 1b7b4eb · 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: 115.7 KiB
• Claude decoded thread snapshot: 116.4 KiB

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