docs: add google analytics provisioning retro

Author: intel352

docs/retros/2026-05-26-google-analytics-provisioning-retro.md

@@ -0,0 +1,63 @@
+# Retro: Google Analytics Provisioning
+
+**PR:** #9 — Add Google analytics provisioning
+**Merged:** 2026-05-26
+**Branch:** feat/google-analytics-provisioning
+**Design:** docs/plans/2026-05-26-google-analytics-provisioning-design.md
+**Plan:** docs/plans/2026-05-26-google-analytics-provisioning.md
+**Related ADRs:** none
+
+## Adversarial-review findings, scored
+
+| Phase | Finding | Severity | Outcome |
+|---|---|---|---|
+| design | Audit JSONL promised but not task-owned | Important | Resolved upfront |
+| design | Live-apply blocker described but not verified | Important | Resolved upfront |
+| design | GTM publish/versioning deferred | Minor | False positive |
+| plan | Audit file not represented in task files | Important | Resolved upfront |
+| plan | No-credentials live apply not verified | Important | Resolved upfront |
+| plan | One PR is large | Minor | False positive |
+
+## Gate misses
+
+| Issue | Gate that missed | Why it slipped | Fix idea |
+|---|---|---|---|
+| Original plan used direct `go run ./cmd/... analytics ...`, but plugin CLI needs `--wfctl-cli` locally | writing-plans | Plan author copied the user-facing `wfctl` shape into a plugin-binary smoke command without checking SDK dispatch rules | Add plugin CLI smoke examples to writing guidance when `sdk.ServePluginFull` is present |
+| CLI ensure path initially skipped default audit JSONL | requesting-code-review | No subagent tool was available, so inline review caught it late rather than as a separate review gate | Keep audit-path behavior in the review checklist for state-mutating plugin CLIs |
+
+## Missed skill activations
+
+| Gate | Fired? | Notes |
+|---|---|---|
+| brainstorming | yes | Design checkpoint presented; user approved autonomous continuation. |
+| adversarial-design-review (design) | yes | Run inline; findings recorded in adversarial-review report. |
+| writing-plans | yes | Plan created with scope manifest. |
+| adversarial-design-review (plan) | yes | Run inline; findings recorded in adversarial-review report. |
+| alignment-check | yes | Scope manifest check passed and plan was locked. |
+| subagent-driven-development | partial | Codex subagent dispatch was unavailable; execution ran inline against the locked manifest. |
+| finishing-a-development-branch | yes | PR #9 created and merged after green CI. |
+| pr-monitoring | yes | PR and post-merge main CI were monitored until green. |
+| post-merge-retrospective | yes | This retro. |
+
+## What worked
+
+- Scope lock caught the task-heading format issue before implementation.
+- Adversarial review forced audit JSONL and live-apply blocking into the task plan before code.
+- Runtime launch validation caught the local CLI dispatch shape before PR creation.
+- PR monitoring confirmed PR CI and main-branch CI were green before closing the loop.
+
+## What didn't
+
+- The initial docs overemphasized CLI instead of YAML; user feedback corrected this before merge.
+- The first implementation of ADC was ambiguous; inline review converted it to explicit `allow_adc`.
+- The CLI audit path gap was caught late by inline review, not by an automated reviewer.
+
+## Plugin-level follow-ups
+
+- No plugin-level changes yet. If another plugin PR repeats the `--wfctl-cli` smoke mismatch, add a dedicated check to `runtime-launch-validation`.
+
+## Project guidance updates
+
+| Guidance file | Change | Reason |
+|---|---|---|
+| `docs/design-guidance.md` | no change | Existing guidance already says YAML pipeline is source of truth, secrets are declarative wiring, and audit trail is required for mutating ops. |