fix(pr): preserve scoped caches and share GitHub read budgets

[rynfar]

Completing a turn or changing one PR previously cleared unrelated persisted reads, and hostless versus explicit GitHub references duplicated detail reads. This scopes invalidation to the affected project or PR, canonicalizes cache references, recovers compatible legacy snapshots, and includes CLI reads in the shared GitHub budget with a bounded quota probe and interactive reserve.

Preserves Pylon’s environment-specific cache fence, selected-environment RPC routing, old-server capability fallback, projected PR authority, and existing conservative GraphQL observation accounting. No cross-environment account router is introduced. Upstream’s router hedging changes and routed-account service tests are inapplicable to Pylon’s absent router; optional server-local credential pinning and scoped quota primitives retain their host checks. Forgejo-only snapshot behavior is not applicable to Pylon’s currently supported providers.

Sources and credit: Bilal Bakr / Julius Marminge, f4600d77dd7c2fa9f10e8f4500882e427fcc7e26 (#11888); maria / maria-rcks, d612d12b8bb278af97b0029f8d48b2403a5f9ee6 (#12168). Original authors and cherry-pick provenance preserved.

Validation: 177 focused backend tests (six files) passed after adapting two test fixtures; 266 tests across PR detail logic, service, checkpoint and rollback paths passed. Server/web scoped typechecks and changed-file lint passed (existing warnings). Direct adversarial self-review covers host/environment isolation, mutation/turn invalidation, persisted TTL/revision behavior, reserve exhaustion and CLI credential host boundaries. No delegated review per maintainer instruction. No visual layout changes; cache behavior is covered by focused tests. Final-head CI required before merge.

Part of #625; parent inventory #598. Frozen head/cursor unchanged. No migration, release or installation.

Implemented and reviewed directly with Codex (GPT-6-Astra).

---

^{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 3:12pm 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	13.9 KiB	−11 B (−0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	−7 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−4 B (−0.1%)	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	−11 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	−5 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.7 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: 315445b · PR result: 87fb720 · 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.