diff --git a/.claude/skills/contributor-pipeline-gardening/SKILL.md b/.claude/skills/contributor-pipeline-gardening/SKILL.md
index 81bead0b9e..ebaf8ffe28 100644
--- a/.claude/skills/contributor-pipeline-gardening/SKILL.md
+++ b/.claude/skills/contributor-pipeline-gardening/SKILL.md
@@ -52,9 +52,9 @@ assume it keeps happening, don't assume today's backlog is clean.
 2. Pick real gaps to scope from, in priority order:
    - Existing open epics/roadmap issues in this repo that don't yet have enough decomposed child issues to be actionable (e.g. `ORB - Long Term Features & Improvements`, the review-comment-redesign family, any epic whose own body describes scope with no filed sub-issues yet).
    - Genuine gaps found by reading the current codebase against a shipped feature's own stated acceptance criteria (the same technique Pass 1 uses to verify closure — used here in reverse, to find what's NOT yet done).
-   - AMS selfhost hardening and the unified AMS+ORB selfhost harness are named standing priorities — see `reference.md` for what's already been investigated there (as of 2026-07-14, no existing issue covers "install both products together"; it needs fresh scoping, not relabeling).
-3. Every new issue gets: the correct existing milestone (create a new one only if none fits — don't dump unrelated work into an ill-fitting bucket), a `gittensor:bug` (0.05x), `gittensor:feature` (0.25x), or `gittensor:priority` (1.5x, reserved for mission-critical/time-sensitive work only — this repo uses it sparingly, unlike metagraphed's looser convention, see `reference.md`) label, plus `help wanted` (the maintainer confirmed this stays as a visibility signal alongside the points label, not a replacement for one).
-4. Every new issue body follows the template in `reference.md` — Context, Requirements, Deliverables, Test Coverage Requirements (this repo's Codecov patch gate is 99%+, hard — every new issue implicitly inherits this unless it's `apps/**`-only UI work), Expected Outcome. No "left to interpretation" scope — the maintainer's own stated preference is that thin/ambiguous issue bodies are worse than fewer, complete ones.
+   - AMS selfhost hardening is a named standing priority — see `reference.md`. The unified AMS+ORB selfhost harness is now scoped and issue-backed (#5996, epic #6012) as of 2026-07-15 — check its sub-issues' completion state before assuming this still needs fresh scoping.
+3. Every new issue gets: the correct existing milestone (creating a new one requires a genuinely-unfitting body of work AND is a much higher bar than it sounds — see `reference.md`'s milestone-discipline note; when in doubt, fold into the closest existing bucket and say so rather than create one), a `gittensor:bug` (0.05x), `gittensor:feature` (0.25x), or `gittensor:priority` (1.5x, reserved for mission-critical/time-sensitive work only — this repo uses it sparingly, unlike metagraphed's looser convention, see `reference.md`) label, plus `help wanted` (the maintainer confirmed this stays as a visibility signal alongside the points label, not a replacement for one).
+4. Every new issue body follows the template in `reference.md` — Context, Requirements, Deliverables, Test Coverage Requirements (this repo's Codecov patch gate is 99%+, hard — every new issue implicitly inherits this unless it's `apps/**`-only UI work), Expected Outcome. No "left to interpretation" scope — the maintainer's own stated preference is that thin/ambiguous issue bodies are worse than fewer, complete ones. **The review gate only enforces what the issue text explicitly says** — see `reference.md`'s dedicated section on this; any deliverable with a file-type/path/format constraint (docs as website pages vs. markdown, native relationships vs. checklists, etc.) needs that constraint stated as an explicit, standalone rule, not left implied by Context.
 5. Link relationships using GitHub's native features, not prose: `addSubIssue` to attach a new issue under its parent epic, `addBlockedBy` when an issue genuinely cannot start before another lands. Only use these where a real dependency exists — don't chain issues into an artificial order to look organized.
 6. Quality over the number. If a scan doesn't turn up enough genuinely well-scoped, non-redundant, correctly-boundaried issues in this repo alone to keep its own count in the 50-100+ range on a given day, file fewer rather than pad with weak ones — note the shortfall in the daily digest instead.
 
diff --git a/.claude/skills/contributor-pipeline-gardening/reference.md b/.claude/skills/contributor-pipeline-gardening/reference.md
index eb2cabb924..ac9c089dc1 100644
--- a/.claude/skills/contributor-pipeline-gardening/reference.md
+++ b/.claude/skills/contributor-pipeline-gardening/reference.md
@@ -24,14 +24,15 @@ Two products, self-host-first:
   requires a fresh gap-audit (read the current `packages/loopover-miner`/`-engine` code against what
   Wave 4 already covered — coverage gate, ledger races, MCP scaffolding — and find what's still
   genuinely thin), not relabeling existing issues.
-- **Unified AMS+ORB self-host harness** — letting one operator install/run both products together in
-  one system, to "close the loop" as both maintainer and contributor. Searched exhaustively
-  (2026-07-14): no existing issue, open or closed, covers this — not verbatim, not conceptually. The
-  closest real thread is #4878-#4884 (extracting ORB's core review logic into `gittensory-engine`,
-  the same package AMS's miner logic already lives in) — that's the actual prerequisite plumbing, not
-  the harness itself. This needs fresh scoping: read the current self-host `docker-compose.yml`
-  (already supports optional profiles — `ollama`, `observability`, `postgres`, `rees`, etc.) and design
-  how an AMS-miner profile/service could sit alongside it, then file a real epic + sub-issues.
+- **Unified AMS+ORB self-host harness** — now scoped and issue-backed as of 2026-07-15, don't
+  re-discover it as unscoped in a future run. #5996 covers the combined ORB+AMS quickstart doc itself.
+  A broader 2026-07-15 audit of every `.md` file in the repo found AMS's operator-facing docs
+  (`packages/loopover-miner/DEPLOYMENT.md` + 10 files under `docs/`) exist only as raw markdown, never
+  ported to the real docs website (`apps/loopover-ui/src/routes/docs.*.tsx`) the way ORB's
+  self-hosting docs already are — filed as epic #6012 (milestone `ORB - Long Term Features &
+  Improvements`, not its own milestone — see the milestone-discipline note above) with one sub-issue
+  per source file. Check #6012's sub-issue completion state before assuming this work still needs
+  scoping from scratch.
 
 ## Milestone taxonomy (as of 2026-07-14 — re-check before trusting, this moves fast)
 
@@ -45,8 +46,13 @@ Two products, self-host-first:
 | `LoopOver Rebrand Migration (maintainer)` | Brand/infra cutover | No |
 | Unmilestoned | Orphans | Usually fine to fold into the closest-fitting existing milestone above rather than leave adrift |
 
-**Don't invent a new milestone reflexively.** Only create one when a new body of work genuinely
-doesn't fit any existing bucket (the unified-harness epic, once scoped, likely needs its own).
+**Don't invent a new milestone reflexively — confirmed the hard way, 2026-07-15.** When scoping the
+unified AMS+ORB harness epic (#6012) into an 11-issue docs-porting body of work, a new milestone was
+created for it; the maintainer immediately reverted this and folded the epic into the existing
+`ORB - Long Term Features & Improvements` grab-bag instead. Treat "this genuinely doesn't fit any
+existing bucket" as a much higher bar than it sounds — even an epic-sized new initiative should
+default to the closest existing milestone unless explicitly told otherwise. If a new milestone still
+seems warranted, propose it and wait for confirmation rather than creating it unilaterally.
 
 ## What's safe to unleash — the actual test
 
@@ -71,6 +77,38 @@ A concrete engineering task is safe to hand to a contributor when:
 When genuinely unsure, default to `maintainer-only` — a wrongly-locked issue costs one manual unlock
 later; a wrongly-unlocked one costs a contributor's wasted PR and possibly a bad precedent.
 
+## The gate only enforces what the issue explicitly says — never rely on implied intent
+
+**Confirmed by the maintainer, 2026-07-15:** the loopover-orb review agent (the gate) checks a PR
+against whatever the linked issue **explicitly** states as fulfillment requirements. It does not
+infer intent from narrative Context, from "obviously implied" scope, or from this skill's own
+general conventions — none of that is visible to the gate at review time. Only the issue's own text
+is.
+
+**What this broke once already:** issue #5996 (a combined ORB+AMS self-host doc) explained in
+Context that public-facing docs live on the website, not as repo markdown, and had one Requirements
+bullet saying so. A contributor's PR (#6011, filed *after* that bullet was added) still added a new
+root-level `.md` file plus edited three other markdown files as its "fix" — a plausible-looking wrong
+interpretation the gate had nothing more explicit to check against, since the constraint wasn't
+phrased as an unambiguous, standalone rule.
+
+**How to apply, every time a deliverable's file type/path/format actually matters (not just docs —
+also applies to "this must be a native GraphQL mutation not a markdown checklist," "this must reuse
+the existing X pattern not invent a new one," etc.):**
+- State the hard constraint as its own callout or leading Requirements bullet, not folded into
+  prose Context.
+- Name the exact file path/pattern the deliverable must match.
+- Name the exact anti-patterns that do **not** satisfy the issue even if superficially on-topic (e.g.
+  "adding a new `.md` file anywhere in the repo, including the root," "editing README.md to add this
+  content instead of creating the route file").
+- For anything doc-shaped, use language close to: `> ⚠️ Read this before starting. The deliverable
+  is a website page at <exact path>. It is not a markdown file. A PR that adds/edits any .md file as
+  the fix does NOT resolve this issue and will be closed.` — see #5996 or any of #6012's sub-issues
+  for the exact wording already proven out.
+- Assume the issue text is the *only* thing an AI-harness-driven contributor's agent will read before
+  acting — don't assume it will also read this skill file, the repo's CLAUDE.md, or common sense
+  about the file type.
+
 ## Labels
 
 - `gittensor:bug` — 0.05x multiplier. Bug fixes.