feat(agent): append the feedback survey invite to surveyed recipes

[thecodedrift]

Stack (root → tip):

• docs(openspec): propose qualitative survey feedback via PostHog surveys #341
  • feat(scaffold): migration 0007 ignores .taskless/.tmp-* scratch files #342
    • feat(feedback): add feedback send/dismiss, the survey constants, and the cadence store #343
      • docs(agent): add the feedback and feedback-invite recipes #344
        • ➡️ feat(agent): append the feedback survey invite to surveyed recipes #345 (you are here)

Slice 4 of 4 of qualitative-survey-feedback (stacked on #344). The switch-on: with this, agent onboard / create-sg-rule / create-vale-rule / create-remote-rule and taskless onboard start carrying the invite.

What

• src/survey/invite.ts: surveyGateIsOpen (telemetry on, CI not true/1, topic surveyed, next_ask absent/corrupt/passed) and withSurveyInvite, which on an open gate captures survey shown { $survey_id }, writes next_ask 10 days out, and appends feedback-invite rendered header: false after the recipe's last section. Closed gate returns the recipe byte-identical.
• agent.ts and onboard.ts both call it after getRecipe(); onboard --mark-complete and the already-onboarded early return do not. The prompts render path is untouched, so the agent/prompts byte-parity test passes as-is (the suite runs with the opt-out set, which closes the gate).
• isCiEnvironment() in util/interactive.ts is now the one reading of CI; shouldLaunchWizard uses it too.
• The two existing telemetry mocks gain isTelemetryEnabled: () => false so they keep measuring cli_agent alone.
• Archived on this tip (3b4d737) after the end-to-end verification 4.6 asked for: a dev build of this branch fired cli_agent → survey shown on the first fetch, served the bare recipe on the second fetch inside the window, and survey sent arrived in PostHog with all five questions resolved to their text (completed recognised as the single-choice Yes). Scenario retention checked per CLAUDE.md: analytics 57→58, cli 72→73, cli-taskless-bootstrap 26→29, plus the new cli-feedback-survey spec with 19. The test response (uuid 01a0b255-83c6-7f57-8bb9-215aef81a256, opens with LOCAL VERIFICATION 4.6) is still live and should be archived from the survey's responses view.

Tests

survey-invite.test.ts (17): open gate appends exactly the fragment with one # Topic: line, captures survey shown, writes the cadence; each of the four topics is surveyed; within-window, opt-out (cadence not even read), CI=true/1, unsurveyed topic, and corrupt-file repair; then through the commands: agent create-sg-rule prints the invite after cli_agent, agent check does not, taskless onboard does, --mark-complete and an already-onboarded project do not.

Refs #341

[thecodedrift]

@claude /review

[claude]

Claude finished @thecodedrift's task in 6m 52s —— View job

---

Review complete

Review mode: incremental — read 0 prior review thread(s) before reviewing. .prior-review.json contained no comments, reviews, or review threads for this PR (only the /review trigger comment itself), so this is the first review of PR #345. Everything below is [New].

• Read .prior-review.json
• Understand the request (/review)
• Fetch PR diff and metadata (gh pr diff 345)
• Read changed files in full context (agent.ts, onboard.ts, invite.ts, cadence.ts, constants.ts, interactive.ts, telemetry.ts, recipes.ts, the new/changed test files, and the OpenSpec deltas)
• Run finder angles (correctness, cross-file, removed-behavior, cleanup, altitude, conventions)
• Verify candidates
• Post inline comments
• Post final summary

Assessment

This is a tight, well-scoped slice with strong test coverage (17 new tests in survey-invite.test.ts covering every gate branch, both serving paths, and the corrupt-cadence-repair case). I traced the gate logic (surveyGateIsOpen/withSurveyInvite in packages/cli/src/survey/invite.ts), both call sites (agent.ts, onboard.ts), the cadence store (cadence.ts), and the shared isCiEnvironment extraction in util/interactive.ts, and checked for regressions against prompts.test.ts byte-parity, the vitest global telemetry opt-out (vite.config.ts sets DO_NOT_TRACK/TASKLESS_TELEMETRY_DISABLED for the whole suite, which is why the execFile-spawned integration tests in onboard.test.ts never hit the gate), and the OpenSpec archive (the MODIFIED Requirements deltas fully restate every prior scenario, so nothing was silently dropped per the CLAUDE.md archive-loss warning).

One finding, posted inline on packages/cli/src/survey/invite.ts:

• [New] surveyGateIsOpen reads next_ask well before withSurveyInvite writes the new value (separated by a getRecipe call and a getTelemetry await). Since each CLI invocation is its own process, two concurrent surveyed-topic invocations (two agent sessions, an editor + terminal) can both read the same stale cadence, both see the gate open, and both fire survey shown / append the invite before either write lands. Low severity — no crash or data loss, just a possible duplicate invite/event against the intended 10-day cadence — and there's no test for the concurrent case. Flagging for a decision on whether it's an accepted tradeoff.

Nothing else survived verification: the telemetry-mock additions (isTelemetryEnabled: () => false in the two agent-*-telemetry.test.ts files) are necessary and correct — without them, the mocked ../src/telemetry module would lack that export and invite.ts would throw when routing through a surveyed topic like create-sg-rule. The isCiEnvironment extraction in util/interactive.ts is behavior-preserving. The two serving paths (agent.ts, onboard.ts) correctly share one cadence file per survey (not per topic), matching the spec's "one epoch timestamp per survey" requirement, and both correctly skip the gate on --mark-complete / the already-onboarded early return.
· branch: survey/04-survey-gate

[thecodedrift]

Re: @claude[bot] — "Review complete … One finding, posted inline on packages/cli/src/survey/invite.ts"
#345 (comment)

Read. The one finding (read-then-write window on next_ask) is being addressed on this branch as an accepted tradeoff with the window narrowed; the inline thread will carry the reasoning and the commit.

— AI Coding Agent