docs(agents): stacked PR series are not a supported working form (#16743)

* docs(agents): stacked PR series are not a supported working form

Record the maintainer's ruling (director seat, decision batch #66): a
stacked series -- each PR branched off the one below -- is not a supported
working form in this repository, and no gate rule or merge-policy change is
made for it. A multi-card change uses a trunk branch and pays the two
recorded workarounds.

The paragraph is self-contained rather than a pointer: AGENTS.md is in
check:pm-skill-id-lint's scan set (pattern /#[0-9]{3,}/), so the card number
cannot be cited in the file. The three tooling blind spots are stated in one
clause each so the rule is actionable without dereferencing history.

Fold payment for the 1068/1068 line ratchet: the ADR-0087 marker block in
the Post-Task Checklist listed 4 of the gate's 7 disposition categories --
a drifted copy of output the same paragraph already calls "the authority".
Replaced by a pointer to the gate's own FIXIT, which prints the full set.
Net 0 lines.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P58euzUXCVJNwmhuPC9DXY

* docs(agents): drop the heavy-CI clause from the stacked-series paragraph

Seat review: the ruling grades that cost as "a defect independent of
stacking" and splits it out to its own devx card, so it is a defect being
fixed -- not a price of an unsupported form. Stating it in AGENTS.md as an
inherent property of stacking contradicts the ruling, and the clause goes
false the moment that card lands, rotting in place like any other restated
fact.

The paragraph now carries only the two structural costs the ruling did
assign here: squash landing destroys the ancestry link, so every descendant
pays a rebuild lap per landing; and a breaking changeset's ADR-0087
disposition is base-relative, so a stacked card's two bases demand
contradictory markers.

No pointer to the split-out card: AGENTS.md is in check:pm-skill-id-lint's
scan set, and any wording like "CI does not run on these yet" would itself
go false when that card lands.

AGENTS.md 1068 -> 1067; the ratchet is a cap, so a net decrease is legal and
nothing was restored to pad it back.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P58euzUXCVJNwmhuPC9DXY

* docs(agents): keep the stacked-series paragraph inside the 120-byte budget

check:pm-skill-ratchet enforces a per-LINE byte budget alongside the
per-file line ceiling, and the reflow left L469 at 121B. Rewrapped to five
lines, each under 120B, with the closing boundary shortened to "No gate or
merge-policy change is made for it" to fit without an orphan line.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P58euzUXCVJNwmhuPC9DXY

---------

Co-authored-by: Claude <noreply@anthropic.com>

Author: hotlong

AGENTS.md

@@ -461,6 +461,12 @@ Even inside your own worktree, operate defensively:
    The issue number is what makes in-flight work *discoverable* — `git ls-remote --heads
    origin | grep issue-<n>` is a one-command pre-check, and the Duplicate Fix Guard
    workflow warns on fix PRs whose branch names no declared issue.
+
+   ⛔ **Off `main` is literal — a stacked series, each PR branched off the one below, is NOT a supported form.** No
+   tooling represents it: squash landing destroys the ancestry link, so every descendant rewinds behind what landed and
+   pays a rebuild lap per landing; and a breaking changeset's ADR-0087 disposition is base-relative, so a stacked card's
+   two bases demand contradictory markers. A multi-card change uses a **trunk branch**: correct the trunk's disposition
+   to `registered` before it merges, and pay the rebase laps. ⛔ No gate or merge-policy change is made for it.
 3. **Never `git push --force` / `--force-with-lease`, and never push `main`.** A
    force-push can clobber a parallel agent's work; `main` is shared — land all via PR.
 4. **Verify the current branch before every commit/push**
@@ -1037,16 +1043,9 @@ registry? Add it to `OPEN_CAPABILITY_REGISTRIES` in the same PR that fixes it.
    schema is `.strict()`. The changeset is one of fourteen surfaces a retirement touches — follow the
    `spec-property-retirement` skill (`.claude/skills/`) rather than reconstructing the kit, and note the two routes
    imply **opposite** liveness-ledger dispositions.
-   **A breaking changeset must also state its ADR-0087 disposition, in writing.** Add exactly one marker to the
-   changeset body — `pnpm check:adr-0087-registration` enforces it, and the CI step is *Require an ADR-0087
-   disposition on a declared-breaking changeset*:
-   ```
-   <!-- adr-0087: registered SOME-MIGRATION-ID -->
-   <!-- adr-0087: not-required (unpublished) why -->
-   <!-- adr-0087: not-required (already-registered SOME-MIGRATION-ID) why -->
-   <!-- adr-0087: not-required (no-migration-prescription) why -->
-   ```
-   The gate prints the argument when it fails — that output is the authority.
+   **A breaking changeset must also state its ADR-0087 disposition, in writing** — exactly one marker in the
+   changeset body, enforced by `pnpm check:adr-0087-registration` (CI step *Require an ADR-0087 disposition on a
+   declared-breaking changeset*). ⛔ The categories are NOT copied here — the gate prints the full set when it fails.
 4. **A removal that breaks the pinned sibling checkout ships together with the sibling fix and the pin bump — or it
    does not ship.** The `Console Pin Gate` job builds objectui at the pinned `.objectui-sha` against **current** `main`,
    so a removal or rename the pinned sibling still imports turns `main` red for every PR in the repo the moment it