fix(telemetry): stop tracking the completion command by djgould · Pull Request #417 · clerk/cli

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

djgould · 2026-08-17T15:19:07Z

clerk completion <shell> is wired into shell rc files with eval, so it re-runs on every new shell instead of when a user asks for it. Its events measure shell startups, not CLI usage.

Last 14 days in clerk_telemetry.events (test machine excluded):

completion was the Bootstrap Clerk CLI with auth flow #1 command by volume: 19,436 of 55,994 events (34.7%) — ahead of api (25.9%)
from 9 machines, one of which (59a6b836…) sent 98.2% of them (12,356 in a single day)
it was 60.2% of all CLI events on Aug 14 and 46.8% on Aug 13; drop that one machine and completion is 0.6% of volume

Change: startCommandTelemetry skips completion, so no event and no first-run disclosure notice on a shell-startup run. __complete (each Tab press) already exits in cli.ts before Commander, so it never emitted events.

Telemetry integration tests used clerk completion zsh as their sample command; they now drive clerk webhooks token (also offline and auth-free).

Data

🤖 Generated with Claude Code

`clerk completion <shell>` is wired into shell rc files with `eval`, so it re-runs on every new shell rather than when a user asks for it. It was 34.7% of all CLI telemetry over the last 14 days (19,436 of 55,994 events), 98% of it from a single machine, and 60% of one day's volume — enough to swing every command-mix and volume chart. `__complete` (fired on each Tab press) already exits in cli.ts before Commander runs, so it never emitted events. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changeset-bot · 2026-08-17T15:19:20Z

🦋 Changeset detected

Latest commit: 943b47c

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

coderabbitai · 2026-08-17T15:20:06Z

Caution

Review failed

An error occurred during the review process. Please try again later.

_{Comment @coderabbitai help to get the list of available commands.}

- restore an argv-leak assertion on a *sent* event (the driver swap dropped the only one): the invalid --mode test now asserts "banana" never appears - make the completion suppression test discriminate: it asserted zero requests on the first-run grace path, where nothing sends anyway, so it passed with the guard removed — it now also runs past the notice - collapse the one-entry UNTRACKED_COMMANDS set to a direct comparison - drive `webhooks token --json` in telemetry tests to skip outro's animation - drop the unit test the integration test strictly covers - note the completion exception in the module docstring Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

wyattjoh approved these changes Aug 17, 2026
View reviewed changes

wyattjoh merged commit 93bec58 into mainAug 17, 2026
9 of 10 checks passed