fix(telemetry)!: spell PostHog properties in snake_case

[thecodedrift]

Spells every PostHog property in snake_case, matching PostHog's own convention.

The rule this establishes

• PostHog properties are snake_case.
• JSON payload fields are camelCase.

Two wire formats with two audiences: PostHog properties are read in insights and dashboards, JSON payloads are parsed by agents and scripts. Each follows its own ecosystem. The same value can legitimately appear under both spellings, and that is correct rather than something to reconcile.

The convention is now stated normatively in the analytics spec rather than being implied by whichever names happened to exist.

Renames

current	becomes	event
cliVersion	cli_version	every event + identify
scaffoldVersion	scaffold_version	every event + identify
durationMs	duration_ms	cli_run
loggedIn	logged_in	cli_run
ruleCount	rule_count	cli_rule_created, cli_rule_improved
errorCount	error_count	cli_check_completed
warningCount	warning_count	cli_check_completed

Unchanged because they are single words or already correct: cli, anonymous, command, meta, status, success, findings, gh_owner.

groupType and groupKey are parameters of the PostHog SDK's groupIdentify call, not properties we name, so they are untouched.

JSON is already compliant and is not touched. Every field across packages/cli/src/schemas/*.ts is camelCase.

The last two were nearly missed

errorCount and warningCount are built into an object in check.ts about forty lines before the capture() that sends them, so an audit that greps around capture( does not surface them. They were found by tracing the event instead.

They are folded in here rather than left for a follow-up because the cost of this rename is not the edit, it is the discontinuity, and that cost is paid once per release. Renaming five now and two later pays it twice for the same total work.

Breaking, and how

This is a hard cut with no dual-emit window. Historical events keep the old spellings permanently, so any PostHog query spanning this release needs both names to see a whole series. Saved insights, dashboards and cohorts filtering or breaking down on these seven need updating by hand; they will not error, they will silently return only the events emitted before the release.

camelCase locals are deliberate

Only the wire boundary changes. RunContext.durationMs and RunContext.loggedIn stay camelCase because they are domain types, not payloads; the translation happens where the event is emitted (duration_ms: context.durationMs). Same for the check counters. This keeps RunContext construction across the codebase and tests untouched.

Untouched for the same reason: loggedIn in taskless info --json (a JSON field, correctly camelCase), and install.cliVersion in .taskless/taskless.json (a manifest field, not telemetry).

A latent contradiction, now resolved

The cli_run requirement in the analytics spec already said cli_version while the code emitted cliVersion. Resolved in the code's favour.

Archived changes are not rewritten

openspec/changes/archive/** is left alone. Those record what actually shipped, and cliVersion genuinely was what shipped. Rewriting them would falsify the record, which is the same thing this changeset warns consumers about on the PostHog side.

Verification

pnpm typecheck, pnpm lint, pnpm test (929), pnpm cli check (exit 0), pnpm openspec validate --all --strict (24 passed).

Fixes #189

[thecodedrift]

Closing. The premise behind this PR is wrong, and the evidence I collected supported the opposite conclusion.

The codebase was consistent, not inconsistent. Measured on main at 3c78ffe:

	convention
event names	cli_run, cli_check_completed, cli_rule_created, and 7 more, all snake_case
properties	cli, cliVersion, scaffoldVersion, durationMs, loggedIn, ruleCount, errorCount, warningCount, all camelCase

That is a deliberate split: snake_case for the event taxonomy, camelCase for properties. Every property followed it.

gh_owner is the only exception, and I introduced it in #190. When the property was renamed from gh_org to gh_owner I carried the underscore across without checking what its neighbours looked like. Then I audited, found one snake_case property beside seven camelCase ones, and concluded the seven were wrong.

The justification I wrote here does not survive either. PostHog spells $current_url and $lib_version that way because they are its own reserved properties; that says nothing about what a project should call its own.

So this PR renames seven properties away from the house convention, pays a permanent discontinuity in event history to do it, and adds a normative spec requirement asserting the wrong rule.

The minimal correct change is the inverse: rename gh_owner to ghOwner. One property, no historical discontinuity, no dashboards to update. Opening that separately.

Two things worth carrying forward from here rather than losing:

• The convention should be written down. This PR added a normative requirement to the analytics spec, which was the right instinct with the wrong content. The replacement states the real rule: snake_case event names, camelCase properties.
• The changeset bump was wrong too. It said patch while both commits were marked !, and comparable behaviour changes in this repo use minor. Moot now, but worth not repeating.

Correction recorded on #189.