fix(compaction): preserve drafts and queue messages through durable admission

[rynfar]

Compacting context now sends /compact separately while preserving the composer draft, attachments, and cursor. Messages submitted during compaction are persisted in FIFO order and sent afterward with their selected model, permissions, and interaction mode.

Pylon previously rejected these messages in the pure decider because compaction still owned the pending turn admission. The port adds a durable queue and correlated internal completion/resume receipts, preserving exact admission and provider-incarnation checks. Provider snapshots cannot erase the queue; only the queued admission's validated session bind can move it to a new provider incarnation during a settings transition. Stop and interrupt cancel unsent entries, while failure and restart leave explicit per-message explanations. An interrupted in-flight send is labeled as uncertain delivery rather than replayed automatically. Slow compaction startup uses the existing bind-pending CAS so a late provider cannot overwrite Stop.

Prime Agent's native compact/abort controls, provider availability gates, and rollback fences remain authoritative. The new queue applies to the shared slash-command compaction path across local, remote, relay, and tunnel clients; it adds no provider-specific protocol behavior.

Upstream cycle #497, frozen range 6c583620ff7ad3235b135af7107c0543467eecfa..4a4c6dd2adc350a68ba18bb28b24b5a7e4660dab:

• Adopted ef6fa118749572da1905eca5d0543d1e98f0b1b2 — preserve drafts when compacting context (#11103), adapted to Pylon's settings and source-epoch APIs.
• Adopted 57aee3e19f1910f3323f06384bb2a1d02b79e369 — queue messages during context compaction (#11107), adapted to the persisted decider/admission architecture rather than a reactor-only memory map.

Migration 059 from merged #510 precedes compaction migration 060. This branch is rebased onto Pylon after #510 and #511.

Validation: 377 focused tests across reactor, decider, lifecycle, projection, engine, migration, contracts, and web dispatch logic; typechecks for server, contracts, shared, client-runtime, web, desktop, and mobile; scoped lint and formatting; frozen install. Regression cases cover FIFO settings, single projected bubbles, stale provider snapshots/completion, stop before send and during slow session startup, and restart cancellation. Independent adversarial review is approved; CI is pending on the final rebased head.

Upstream UI recording demonstrates preserved draft/cursor behavior in T3. This is upstream evidence; no Pylon UI recording or manual provider/device session was captured for this port. Native Prime compaction is unchanged.

Implemented and reviewed with GPT-6 in Codex.

[vercel]

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

Project	Deployment	Actions	Updated
pylon-marketing	Ready	Preview	Sep 12, 2026 7:08am 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	14.1 KiB	13.9 KiB	−153 B (−1.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.9 KiB	6.7 KiB	−159 B (−2.3%)	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	14.1 KiB	−3 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	+7 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.7 KiB	59.7 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: a1f748b · PR result: c9c901d · 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: 116.3 KiB

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

[vercel]

Deployment failed for project pylon-marketing with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/pylon-code?upgradeToPro=build-rate-limit

[rynfar]

Final landing review verified head c9c901df5ac115af1dcdb194698575488ae19743. Independent adversarial review covered durable FIFO admission, exact session ownership through completion/restart/Stop, receipt ordering and migration compatibility. The final two CI-driven changes affect test fixtures only and were independently reviewed. All 10 Actions CI jobs passed in run 34680517219.

The exact compaction and settings heads were also composed at 7b07ec9e621c84d7c18d17b7c18bf6436cc70678: both owners approved the four automatic overlaps, with 266 focused tests and server/web typechecks passing. Vercel preview status is unavailable due to the account build-rate limit; branch protection has no required Vercel context. No quota, deployment configuration, or external plan was changed.