Add --replay-all to browsers telemetry stream

[archandatta]

Summary

Adds --replay-all to kernel browsers telemetry stream <id>, the CLI surface for telemetry replay-from-oldest. When set, it forwards replay=all so the stream starts from the oldest retained ring-buffer event instead of from-now.

• New --replay-all bool flag, mutually exclusive with --seq (one starts from the oldest retained event, the other resumes after a specific sequence).
• Param wiring: --seq -> Last-Event-ID, --replay-all -> replay=all (named constant replayAllValue, since the wire value is shared with the browser image with no generated enum tying them).
• Tests assert the param the SDK receives (replay set / not set, LEI set / not set) and the mutual-exclusion error.
• Bumps kernel-go-sdk to v0.70.0 (exposes the Replay param). The bump also required a small unrelated adaptation to APIKeyService.Get (gained a params arg), kept as its own commit.
• README updated.

Sequencing (KERNEL-1351): kernel-images replay (shipped) -> kernel/kernel #2381 (merged + deployed) -> kernel-go-sdk v0.70.0 (published) -> this PR.

Test plan

• go build ./... + go test ./cmd/... green (SDK bumped to v0.70.0)
• Flag/validation edge cases on the built binary
• Live end-to-end against prod (created a telemetry browser, exercised all three modes, deleted it)

Live test results

Built bin/kernel at kernel-go-sdk v0.70.0 and ran against production.

Flag / validation (these fire before any network call):

case	result
--help	shows --replay-all
--replay-all --seq 5	mutual-exclusion error
--replay-all --seq 0	mutual-exclusion error fires before the seq<1 check
--seq -2 / --seq 0	"invalid --seq value"
--categories bogus	invalid-category error
-o yaml	unsupported-output error

Live end-to-end (created a headless browser with --telemetry all, generated ~7 api_call events, then deleted the browser):

test	observed	expected
--replay-all	7 frames starting at seq 1	replay from oldest
--seq 3	frames seq 4 -> 8	resume after N
from-now (no flags)	only seq 10 (the post-connect event), not 1-7	no replay

Not exercised live: eviction (replay-all's first seq > 1 once the 1024-event ring overflows) needs >1024 events; it is pure server behavior, already verified server-side, and the CLI just prints whatever frames the server returns.

---

Note

Low Risk
CLI flag wiring, SDK bump, and docs/tests only; replay behavior is server-side and API keys get uses an empty params struct.

Overview
Adds --replay-all to kernel browsers telemetry stream, so the CLI can start from the oldest retained telemetry event instead of only live / --seq resume. With --replay-all, the stream sends SDK param replay=all (via replayAllValue); --seq still maps to Last-Event-ID. The two modes are mutually exclusive when --seq was explicitly passed (SeqProvided), including --seq=-1.

kernel-go-sdk is bumped to v0.70.0 for the Replay stream param. api-keys get is updated to pass APIKeyGetParams{} on the new Get signature. README documents the flag and bounded-buffer caveat. Tests cover param wiring and the mutual-exclusion error.

^{Reviewed by Cursor Bugbot for commit ee9db08. Bugbot is set up for automated code reviews on this repo. Configure here.}

[kernel-internal]

🔧 CI Fix Available

I've pushed a fix for the CI failure.

The build broke because kernel.BrowserTelemetryStreamParams has no Replay field (it isn't present in the pinned SDK v0.66.0, nor in later releases). The go vet step failed with params.Replay undefined.

The fix implements --replay-all using the SDK's existing Last-Event-ID mechanism: since the stream resumes with seq > N, setting Last-Event-ID: 0 replays the entire retained buffer. Tests were updated to assert against the actual SDK field. make test now passes.

👉 Click here to create a PR with the fix

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	golang/​github.com/​kernel/​kernel-go-sdk@​v0.66.0 ⏵ v0.70.0	+1

View full report

[rgarcia]

reviewed with live QA. one correctness issue to fix before this ships:

• cmd/browsers_telemetry.go:181 — --replay-all --seq=-1 is accepted and replays from the oldest retained event, even though the new help/docs say --replay-all is mutually exclusive with --seq. consider tracking whether --seq was explicitly provided (for example from cmd.Flags().Changed("seq")) and rejecting the combination based on flag presence, not just Seq != -1.

[Sayan-]

Confirmed feedback flagged by raf has been appropriately addressed. No pending issues from my review 👍