feat(cli): send missing features and host-app friction as ratingless feedback

[miguel-heygen]

What

Agents now report two things the feedback rules never asked about: what a person needed that does not exist yet, and friction in the app that launched the CLI. Both are sent as a comment with no rating, so they never count in the rating metric.

Skill (hyperframes-cli):

• Missing feature. Whenever the person asks for something HyperFrames or the host app cannot do, or can only do with a workaround, the agent sends one report per ask: npx hyperframes feedback --comment "MISSING FEATURE: <what the person asked for, in their words, with names, clients, figures and paths left out> | WORKAROUND: <what the agent did instead, or none>". The report is about the capability they wanted, never their content.
• Host app. When HYPERFRAMES_CLIENT is set (the variable the CLI already reads for the launching app), the CLI is running inside a host app. When that app itself gets in the way (a panel, button, preview or export that misbehaves), the agent sends --comment "HOST APP: <what happened>" with the usual reproduction packet.
• Both are sent only when telemetry is on and the user has not opted out. The missing-feature report describes the capability, never the person's content, and a catalog search that found nothing still goes to --search-miss.
• references/preview-render.md and the CLI docs no longer call --rating required, and the feedback guide shows the ratingless form. skills-manifest.json is regenerated.

CLI (hyperframes feedback):

• A --comment with no --rating is now a ratingless report. It goes to its own PostHog event, cli_feedback_comment (comment, doctor summary, feedback id, recent render ids), never to cli_render_feedback, so it cannot enter the rating metric. The --search-miss report already works this way.
• The forward to the feedback channel omits rating and rating_scale for such a report.
• --rating with or without a comment works exactly as before, and an unreadable rating still gets the same error. A report with neither a rating nor a non-blank comment is refused, and so is --file-issue without a rating (the issue template is built around a score).

Backend dependency

The feedback endpoint still requires an integer rating today. Until it accepts a missing one, the ratingless forward is rejected and dropped like any failed forward: the report still reaches PostHog, but not the feedback channel. A small backend change, in a separate pull request, makes the rating optional and posts such a report without a score. This one is safe to merge in either order.

Checks

• New tests: the ratingless event's name and that it carries no rating; a comment with no rating goes to its own event and never to the rating event, and its forward carries no rating; a blank comment alone is refused; an unreadable rating keeps its error; neither a rating nor a comment is refused, and so is --file-issue without a rating; the request body leaves out rating and rating_scale; the skill teaches both ratingless commands. Each one fails when the line it covers is broken, and passes 3 runs in a row.
• CLI typecheck, lint, format, comment checks, fallow, and the skills manifest check pass.

[github-actions]

Edit accuracy: accurate 2059 (base branch 2059), smooth 1423 of those

The gate passes.
Smoothness is reported in the artifact, not gated. A case fails only if it fails 2 of 3 runs.

Quarantined, measured but not gated (0)

Unstable (1)

• crop-none-px-r0-root-z50: tracking 0.05, pressJump 0, drop 40.07, reload 40.07, render 39.98, renderKey -, undo true, teleport true / tracking 0.05, pressJump 0, drop 0.13, reload 0.12, render 0.03, renderKey -, undo true, teleport true / tracking 0.05, pressJump 0, drop 0.13, reload 0.12, render 0.03, renderKey -, undo true, teleport true

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
hyperframes	🟢 Ready	View Preview	Oct 7, 2026, 6:14 AM

💡 Tip: Enable Automations to automatically generate PRs for you.

[jrusso1020]

Approve @ effdf97b. A comment with no rating now goes to its own PostHog event and never touches cli_render_feedback, so the rating metric stays clean. The rated path behaves exactly as before.

What I checked

• The split. reportRating runs before the source and telemetry checks and refuses three cases: an unreadable rating (same error as before), neither a rating nor a non-blank comment, and --file-issue without a rating. trackFeedback then sends a ratingless report to cli_feedback_comment and a rated one to cli_render_feedback. The lint nudge still runs for rated reports only.

• Telemetry gate. Ratingless reports go through the same shouldTrack() return as rated ones, so an opted-out user sends nothing, as the skill promises.

• Host app. getDoctorSummary already appends client=<HYPERFRAMES_CLIENT> (telemetry/feedback.ts:105), so a HOST APP: report names its app both in PostHog (doctor_summary) and in the forwarded env string.

• Merge order. On EF master, HyperframesFeedbackRequest.rating is a required int. Until EF #55462 lands, a ratingless forward gets a 422, and submitFeedback ignores the response, so the report is lost silently and PostHog still has it. Older CLIs always send a rating, so EF can land first too. The "either order" claim holds. I read #55462: it makes rating optional, skips the bounds check for None, and posts with no score line.

• Tests: the 4 changed files pass (113 tests). Nine of ten mutations fail at least one test:

  • a ratingless report sends no PostHog event (1)
  • it goes to the rating event (1)
  • the neither-nor check is removed (2)
  • a blank comment is let through without trimming (1)
  • the --file-issue check is removed (1)
  • rating_scale is always sent (1)
  • rating becomes required again (1)
  • the comment event carries a rating (1)
  • the skill's host-app bullet is deleted (1)

  The survivor is deleting the printFeedbackLintWarnings call. That call wasn't pinned on main either, so this PR didn't lose any coverage.

Nits (none blocking)

• The HOST APP: packet gets no nudge. The skill asks for a reproduction packet with a HOST APP: report, but lintFeedbackComment takes a rating and only runs on rated reports. So an agent that sends a bare symptom gets no "missing REPRO COMMAND" warning, which is exactly the report that needs one. Possible fix: run the repro-marker check on a ratingless comment that starts with HOST APP:.
• Reuse: trackFeedbackComment repeats the three optional spreads from trackRenderFeedback (doctor_summary, feedback_id, recent_render_ids). One small feedbackJoinProps(props) helper used by both keeps the two events' join keys identical when one of them changes.

CI: red only on Studio: timeline viewport gate. Interaction p95 was 66.9 ms, then 59.1 ms, against a 58.3 ms budget, and all 5/5 runs passed on both attempts. This PR touches no studio code, so a rerun should clear it. The rest were green or still running when I checked.

— Rames

[jrusso1020]

Approve @ 8fb6b132. This head only merges main (2d98d5c1). I diffed the PR's own patch before and after: the same 12 files, and the only changes are a hunk offset in one doc and the regenerated skills-manifest.json hash. The required checks are green at this head. My review at effdf97b stands as written, nits included.

— Rames