diff --git a/docs/gtm-quality-prd.md b/docs/gtm-quality-prd.md
new file mode 100644
index 0000000..648b1e2
--- /dev/null
+++ b/docs/gtm-quality-prd.md
@@ -0,0 +1,75 @@
+# Spec: GTM readiness for BearDrive (reviewer-satisfied goal loop)
+
+Authoritative spec for the GTM improvement loop. The bar is met by
+independent review: a `beardrive-gtm` subagent (COSS operator persona)
+audits the repo's go-to-market surfaces and scores them; the implementing
+agent never scores its own work.
+
+## Scope — what the loop may change
+
+Everything GTM-relevant that lives in this repo:
+
+- `README.md` (positioning, quickstart, time-to-wow, demo assets plan)
+- `website/` (headline, category claim, CTAs, pricing posture copy)
+- `docs/` (self-host guide, quickstart, a launch plan, a metrics plan)
+- Community surface: `CONTRIBUTING.md`, `.github/` issue/PR templates,
+  `CHANGELOG.md` or release-notes convention, a public `ROADMAP.md`
+- `plugin/` copy where it is a funnel surface (install flow wording)
+- The open↔paid line as *explanation* (docs/website copy about what OSS
+  ships vs what beardrive.ai adds)
+
+## Hard guardrails (violating any = the round is void)
+
+1. **Truth only.** Never fabricate social proof: no invented testimonials,
+   logos, user counts, star counts, or benchmarks. Every product claim
+   must be verifiable against the code/docs. Aspirations are labeled as
+   roadmap, not shipped features.
+2. **No license change, no pricing decisions.** Pricing/packaging content
+   is written as clearly-labeled *proposal* for Snow to approve, not
+   published as fact.
+3. **No external actions.** No posting, publishing, account creation,
+   emailing, or launching anywhere. Anything requiring action outside the
+   repo goes to `docs/gtm-handoff.md` — a prioritized checklist for Snow
+   (e.g. "post X to HN with this title", "enable GitHub Discussions",
+   "record 45s demo of Y") with ready-to-use draft copy where applicable.
+4. Keep README ↔ SKILL.md ↔ CLAUDE.md consistent when CLI behavior is
+   described; never commit `internal/webapp/manual_serve_test.go`;
+   `go build ./... && go test ./...` stays green if anything code-adjacent
+   is touched; frontend changes rebuild committed static.
+5. Branch `feat/gtm-readiness`, commit per round, PR at the end; never
+   merge or deploy.
+
+## Review protocol (per round)
+
+Launch a `beardrive-gtm` subagent. It must score the six dimensions from
+its charter (positioning, funnel, open-vs-paid line, community readiness,
+launch readiness, metrics) 0–10 **on repo-controllable readiness**: "has
+everything within this repo's power been done for this dimension?" —
+external-world gaps (no stars yet, no community yet) do NOT cap the score
+if the corresponding handoff items exist and are well-prepared. It returns
+findings tagged blocker / major / minor, each marked REPO-FIXABLE or
+HANDOFF, with concrete fixes (replacement copy written out).
+
+The implementing agent fixes every repo-fixable blocker/major (and minors
+where cheap), routes handoff items into `docs/gtm-handoff.md`, and records
+the round in the Scorecard below.
+
+## Exit bar
+
+Two CONSECUTIVE review rounds where every dimension scores ≥ 8 and there
+are zero repo-fixable blocker or major findings. Disagreements: the
+reviewer's judgment wins unless it violates a guardrail (e.g. demands
+fabricated proof) — record such rejections under Won't-fix with reasons.
+
+## Scorecard (append one row per round)
+
+| Round | positioning | funnel | open-paid | community | launch | metrics | repo-fixable blockers/majors |
+|---|---|---|---|---|---|---|---|
+
+## Won't-fix / disputed
+
+(none yet)
+
+## Status / blockers
+
+(record and stop rather than deviate)