feat(session-sync): phased SW resync with experiment-ready config

[Just-Insane]

📝 Docs PR: SableClient/docs#9 — merge that when merging this.

---

PR dependency (stacked)

This branch is stacked on #572 — it has been rebased onto feat/feature-flag-env-vars with no duplicate commits. Only the two session-sync–specific commits are unique to this PR. Please review/merge #572 first, then review this PR for session-sync behavior.

Description

Adds phased service-worker session re-sync controls, gated behind an experiment flag with direct-config fallback.

Fixes #

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings

AI disclosure:

• Partially AI assisted (clarify which code was AI assisted and briefly explain what it does).
• Fully AI generated (explain what all the generated code does in moderate detail).
  The phased session-sync hooks/flags and config integration were partially AI assisted; I verified branch behavior and safe defaults.

Feature Flags / Environment Setup

Set these GitHub Environment variables:

• CLIENT_CONFIG_OVERRIDES_JSON
• CLIENT_CONFIG_OVERRIDES_STRICT (optional)

How it works

Phase flags are derived in useAppVisibility from the sessionSyncStrategy experiment:

Variant	Phase 1 (foreground resync)	Phase 2 (heartbeat)	Phase 3 (adaptive backoff)
control (default)	❌	❌	❌
session-sync-heartbeat	✅	✅	❌
session-sync-adaptive	✅	✅	✅

Users not in the experiment fall back to the sessionSync.phase* booleans, which can be used for direct-config overrides (e.g. preview/staging where you want all users to get the feature without experiment bucketing).

sessionSync config field reference

Field	Type	Default	Description
phase1ForegroundResync	boolean	false	Push session credentials to the SW on page focus / becoming visible.
phase2VisibleHeartbeat	boolean	false	Periodic heartbeat that keeps SW credentials fresh while the tab is visible and online.
phase3AdaptiveBackoffJitter	boolean	false	Exponential backoff with ±20% jitter on heartbeat failures. No effect unless phase2VisibleHeartbeat is true.
foregroundDebounceMs	number	1500	Minimum ms between foreground-triggered pushes.
heartbeatIntervalMs	number	600000	Base interval between heartbeat ticks (ms).
resumeHeartbeatSuppressMs	number	60000	How long after a foreground push to suppress the next heartbeat tick (ms).
heartbeatMaxBackoffMs	number	1800000	Maximum backoff ceiling for adaptive mode (ms).

Recommended initial values

preview — enable for all users directly (no bucketing):

{
  "sessionSync": {
    "phase1ForegroundResync": true,
    "phase2VisibleHeartbeat": true,
    "phase3AdaptiveBackoffJitter": true
  }
}

production — gradual rollout via experiment:

Note: rolloutPercentage controls what fraction of users enter the experiment at all. Those users are then split evenly across variants. So rolloutPercentage: 20 with 2 variants = 10% per variant, 80% control.

{
  "experiments": {
    "sessionSyncStrategy": {
      "enabled": true,
      "rolloutPercentage": 20,
      "controlVariant": "control",
      "variants": ["session-sync-heartbeat", "session-sync-adaptive"]
    }
  }
}

No environment variable setup is required for deployment to succeed; missing overrides are treated as no-op and checked-in defaults are used.

[Copilot]

Pull request overview

Adds session re-sync phase controls (foreground + visible heartbeat with optional adaptive backoff/jitter) and introduces an experiment/config override pipeline so behavior can be rolled out via environment-scoped config.json overrides at build time.

Changes:

• Extend client config schema with sessionSync toggles and experiments, plus helper APIs to deterministically bucket users into variants.
• Update app visibility handling to optionally push the active session to the service worker on foreground/focus and on a visible heartbeat loop.
• Add a CI/build-time config.json override injector and wire it into GitHub Actions (preview/production environments), plus update default config.json.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/app/pages/client/ClientRoot.tsx	Passes activeSession into useAppVisibility so SW session sync can include full session identity.
src/app/hooks/useClientConfig.ts	Adds sessionSync/experiments config shape and experiment variant selection helpers.
src/app/hooks/useAppVisibility.ts	Implements phased SW session pushes (foreground/focus + heartbeat + optional adaptive backoff/jitter).
scripts/inject-client-config.js	New build-time script to deep-merge env overrides into config.json.
knip.json	Adds the new script as a Knip entry.
config.json	Adds default experiments and sessionSync config blocks.
.github/workflows/cloudflare-web-preview.yml	Exposes env-scoped config override vars to preview deploy workflow.
.github/workflows/cloudflare-web-deploy.yml	Exposes env-scoped config override vars to plan/apply workflows (preview + production).
.github/actions/setup/action.yml	Runs the config override injector as part of build setup.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Just-Insane]

Addressing the three Copilot review comments from this PR:

scripts/inject-client-config.js — prototype pollution in deepMerge

Added an UNSAFE_KEYS guard that skips __proto__, constructor, and prototype during the merge loop. These keys are data-only in any legitimate config object, so skipping them is safe with no functional impact on real overrides.

scripts/inject-client-config.js — logger.error/logger.warn dropping error details

Added a formatError helper (error.stack ?? error.message for Error objects, String(error) otherwise). All logger.error(msg, err) calls are now logger.error(\${msg} ${formatError(error)}`). The invalid logger.warn(msg, err)call (which would have thrownTypeError: logger.warn is not a function) is replaced with logger.info(`[warning] ${msg} ${formatError(error)}`)`.

src/app/hooks/useAppVisibility.ts — two heartbeat backoff issues

1. Prerequisite-not-ready skips growing backoff: Changed pushSessionNow return type from boolean to 'sent' | 'skipped'. The tick loop now only resets heartbeatFailuresRef when the result is 'sent' — skipped calls (no mx, no SW controller, etc.) are not treated as push failures and won't ramp up backoff during startup.

2. Stale backoff/suppression state carrying over across effect restarts: Added heartbeatFailuresRef.current = 0 and suppressHeartbeatUntilRef.current = 0 at the top of the heartbeat useEffect body. Config or session changes that cause the effect to re-run now start with a clean slate.