bug(agent): managed fallback snapshot sanitizer still reaches residual serialization failure

[Yeachan-Heo]

Problem

Current GJC runtime can repeatedly fail a managed fallback turn with:

Error: Managed fallback attempt could not produce a serializable event snapshot (local snapshot bug, not a provider failure)
Error: Retry failed after 3 attempts: Managed fallback attempt could not produce a serializable event snapshot (local snapshot bug, not a provider failure)

The failure repeats across the day and blocks unrelated repository work. A second user independently confirmed the same symptom.

Fresh evidence

• Primary report: Discord 1538093508114780251, 1538093530105381005, 1538093650368663673
• Independent confirmation: Discord 1538094237219029073
• Impact statement: Discord 1538094294156578827
• Reported while trying to repair the current iTerm pet issue; the session cannot make progress because managed fallback repeatedly aborts locally.

Relationship to prior work

Issue #2408 fixed a managed fallback snapshot failure caused by non-cloneable transport headers and added a defensive sanitizer. This fresh error is the explicit residual-serialization failure path introduced to classify local snapshot bugs. Treat this as either a regression or an uncovered event payload shape; do not assume provider failure.

Maintainer-defined scope

• Reproduce the exact event payload that defeats the current managed-attempt snapshot sanitizer.
• Preserve cycle/accessor/proxy safety and bounded byte accounting.
• Make every accepted managed event snapshot serializable, or reject the unsupported payload at the narrow producer boundary with structured diagnostics that identify the offending event/value class.
• Do not hide provider failures, silently drop required event semantics, or weaken managed fallback transaction isolation.
• Add focused regression coverage for the exact payload shape and repeated retry behavior.

Acceptance

• The fresh reproducer no longer reaches Managed fallback attempt could not produce a serializable event snapshot.
• Managed fallback either commits a detached serializable snapshot or returns a bounded actionable local diagnostic without retrying the same deterministic local serialization defect three times.
• Existing Managed fallback chain wedges sessions with "The object can not be cloned." when a provider error carries response headers #2408 clone-safety, provisional buffer limits, transport-failure classification, focused tests, build/typecheck, and exact-head CI remain green.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]

[Yeachan-Heo]

Issue #4578 completion receipt

• Canonical PR fix(agent): serialize detached managed snapshots #4580 squash-merged to dev.
• Approved exact head: 184b85a5e5bd41e68f7e99f56dfb1d9c05ed2ee1.
• Exact base at merge: 70e1f5be10fc146a54311502a4b27042fb38ee7f.
• Canonical diff digest: b91f17b7e83c76c77079aa15f76de1796831a53bb6df621f73e0353f3d6fb856.
• Independent approval: probepark review 4943425134.
• Terminal merge-authorized CI: run 31879337596, all contract/product gates green; expected structural platform jobs skipped.
• Squash merge/current dev: 45885ea28793b0ace3523272b41f7ad07b1e88e5.
• Post-merge @gajae-code/agent-core check: pass.
• Post-merge @gajae-code/coding-agent check: pass.
• Post-merge workspace build: pass.
• Duplicate PR fix(agent): stop managed snapshot failure retry looping #4581 remains closed unmerged; its required retry-policy docs correction is included in the canonical merge.
• No release or tag created.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]