Addresses review feedback on PR #13057.

Quick fixes:
- Replace em-dashes with regular dashes in code comments per AGENTS.md.
- Add JSON tags to SubIssues and LinkedIssueConnection (and TotalCount on
  the latter) for consistent struct serialization.
- Add comment to defensive ID-empty check in IssueNodeID explaining why the
  guard is kept after a GraphQL call.
- Reject --add-sub-issue and --remove-sub-issue when multiple issues are
  passed to issue edit (one parent per issue makes batching ambiguous).
- Add godoc to Editable.{X}.Allowed.

Substantive fixes:
- Add canonical `id` field to relationship exports (parent, sub-issues
  nodes, blockedBy nodes, blocking nodes), and update the GraphQL fragments
  in query_builder.go to fetch them.
- Wrap subIssues, blockedBy, and blocking JSON output in
  {nodes: [...], totalCount: N} so consumers can see the true total when
  more than 50 items exist. The query_builder fragments now request
  totalCount for blockedBy/blocking.
- In issue edit and issue create, validate IssueRelationshipsSupported
  up-front when relationship flags are used, before any mutation, so
  GHES users get a clear error instead of partial state.
- Refactor IssueType handling out of ImmutableKeywords and into the
  Qualifiers struct in pkg/search/query.go. The new IssueType field
  renders as type:X (with proper quoting via quote()), eliminating the
  ad-hoc escaping in pr/shared/params.go.
- In prShared.FetchOptions, surface errors from api.RepoIssueTypes when
  the user has explicitly opted into editing the issue type instead of
  silently ignoring them.
- In EditFieldsSurvey, return a clear error when the user opts to edit
  Type but no issue types are available, instead of silently skipping
  the prompt.

Tests:
- New tests for IssueType qualifier in standard and advanced search
  query strings (including quoting behavior).
- Updated TestIssueView_json_ParentSubIssues and
  TestIssueView_json_BlockedByBlocking to verify the new
  {nodes, totalCount} shape and `id` fields.
- New TestEditFieldsSurvey_IssueTypeNoOptions covering the new error
  case in the survey.
- New cases in TestNewCmdEdit covering the multi-arg sub-issue guard.

Pushback / deferred (intentionally not in this round):
- T6 (mutex on add/remove relationship flags): rejected. Add and remove
  collection deltas are not conflicting, matching --add-label/--remove-label.
- T19/T24 (large architectural reordering of validate-then-mutate and
  capture-errors-then-continue): deferred to a follow-up.
- T23 (--remove-type/--set-type for unsetting issue type): follow-up.
- T3 (cross-host validation in ResolveIssueRef): follow-up.
- T5 (non-TTY field selection): follow-up.
- TTY+IssueType prompt test for issue create: follow-up (large mock setup).

Co-authored-by: Copilot <[email protected]>

gh issue list --json ...,blockedBy,... (issue-claim.sh) needs
blockedBy, which shipped in gh 2.94.0 (cli/cli#13057). On an older gh
the flag is rejected outright -- gh writes its complaint to stderr and
nothing to stdout, so the python3 -c piped after it tries to json.load
an empty stream and dies with JSONDecodeError, a traceback naming the
wrong problem entirely. mise.toml already pins gh well above the
floor, but mise is optional and a session where it is not active gets
whatever gh is already on $PATH -- the same class of gap
RUSTUP_TOOLCHAIN leaves for the Rust pin.

scripts/lib/require-gh.sh is the runtime backstop every
scripts/issue-*.sh now sources: a portable (no `sort -V`, which BSD
sort lacks -- the exact class of GNU-only assumption #559 already hit
once) MAJOR.MINOR.PATCH comparison that refuses in one sentence naming
both versions.

Three scripts source it immediately after `set -euo pipefail`.
issue-land.sh sources it later, right before its first real `gh pr
view` call: everything above that point -- the commit guard, the
gates, the push -- never touches gh at all
(test-issue-land-commit-guard.py's own premise), and demanding a gh
version before a dirty-tree refusal or a red gate would have given
that path a dependency it never had.

Every scripts/tests/*.py that stubs gh for one of the four scripts
needed a `--version` case added to its stub (and scripts/lib copied
alongside the script it copies into a fixture) to keep working under
the new guard -- twelve files, all mechanical.

README and mise.toml now name the 2.94.0 floor and why.

Refs: #558

Three clients for the issue graph, chosen by a probe: gh 2.94.0, the
script, the MCP.

*Three roads lead to the same graph;*
*the probe, not the wish, decides.*
*Where gh is new it walks alone,*
*where none is here, say so and wait.*

Summary
-------

`gh` reaches the whole issue-relationship graph from 2.94.0 on, both
directions of blocked-by included. The `issue-deps` skill did not know
this: it named the script as the only way to a blocking edge and the MCP
as the way to everything else, and the session that reported #129 had an
MCP that reached none of it. This change rewrites the skill's client
matrix as three columns and adds the probe that picks one, so a session
uses the best client its environment has and says plainly what the worst
one cannot do.

Details
-------

**The floor is 2.94.0, not 2.100.0.** The issue measured on 2.100.0.
Bisecting `pkg/cmd/issue/edit/edit.go` across the cli/cli tags puts the
`--add-blocked-by`, `--add-blocking`, `--add-sub-issue` and `--parent`
flags, and the `blockedBy`, `blocking`, `subIssues`, `parent` and
`issueType` JSON fields, all in 2.94.0 (cli/cli#13057);
`closedByPullRequestsReferences` predates it. The skill states that
floor and the two unhelpful errors an older `gh` gives instead of naming
its version.

**Three branches, in order.** `gh` at the floor or later uses the flags.
A token without a usable `gh`, which is every Claude Code web worker,
uses `scripts/issue-deps.sh`, now described as load-bearing rather than
a gap-filler. The MCP alone gets a plain statement: both generations of
the server return counts with no members for blocked-by, and the older
generation, the one with `get_issue` rather than `issue_read`, reaches
neither sub-issues nor `closed_by_pull_requests` and writes nothing into
the graph. The tool list is the probe for which generation is present.

**Kept from the old text.** The propose-from-evidence rule, unchanged.
The pull-request read that answers empty rather than refusing, now noted
for `gh` as well as REST, since `gh` does not close that trap. The
from-the-blocked-side rule, now stated as REST's constraint and the
script's, with `gh` inverting `--add-blocking` client-side.

**Added.** The pipe that eats `gh`'s exit status: `gh issue edit … |
tail` reports `tail`'s `$?`. The `gh` refusals, all honest and non-zero.

**Expiry widened.** The script's header and the skill's `Expiry` section
now retire it when the environment it serves stops existing, not when
the MCP grows the endpoints, because the capability arrived through `gh`
instead and the MCP still has not caught up.

The script's logic is untouched; only its header comments change.

Issues
------

- Closes #129

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01GD6mX5vZRcuFi9DiWv9Gtm

---
_Generated by [Claude
Code](https://claude.ai/code/session_01GD6mX5vZRcuFi9DiWv9Gtm)_

---------

Co-authored-by: Claude <[email protected]>

直近の Issue 71 件はすべて種別なしだった。Form も /issue も種別を指定して
いなかったためで、Organization に種別を足しただけでは使われない。
Experiment 種別を 2026-09-30 に足したので、Issue を作る 2 つの経路に
種別を入れる（development-docs の D10）。

種別は Form の type: 1 箇所に書く。Web の Form はそれで付く。CLI の
issue-form.sh は Form の type: を読んで skeleton と check に表示し、
/issue は gh issue create --type に渡す。種別を書き写さないので、Form を
直せば両方が変わる。--type は gh 2.94.0 で入った（cli/cli#13057）。古い gh
で unknown flag と出たら、--type を外さずに更新を依頼させる。

Form に存在しない種別を書いたときの挙動は公式文書に無い。種別の名前を
Organization 側で変えたり無効にしたりしても Form は黙って残るので、監査
（org_audit.py）が Form の type: を Organization の有効な種別と照合する。
取れなかったときは「無い」と言わない。

実測: gh issue create --type Feature で作った #37 に種別が付いた（この
Organization で最初）。テスト: 監査 42 件、Form と issue-form.sh 4 件
（gh を差し替え）すべて OK。壊す 9 通りすべてで落ちた。実 Organization
への読み取り実行で所見 0 件。

Closes #37

Co-authored-by: Claude Opus 5.5 <[email protected]>

The relationship flags and fields shipped in gh v2.94.0 (cli/cli#13057,
"Issue types, sub-issues, and relationships in `gh issue`"), not 2.100.0.
GH_MIN_VERSION, the messages, the docstrings and the tests now say
2.94.0, and the constant cites that release.

Co-Authored-By: lsimons-bot <[email protected]>
Assisted-by: Claude:claude-opus-5-5

The relationship flags and fields shipped in gh v2.94.0 (cli/cli#13057,
"Issue types, sub-issues, and relationships in `gh issue`"), not 2.100.0.
GH_MIN_VERSION, the messages, the docstrings and the tests now say
2.94.0, and the constant cites that release.

Co-Authored-By: lsimons-bot <[email protected]>
Assisted-by: Claude:claude-opus-5-5