agents: restore missing rule frontmatter and refine CI/rule guidance (#4387)

Several agent rules lacked YAML trigger frontmatter, preventing agent
discovery, while CI monitoring and merge workflows alerted on expected
merge-blocking label gates, woke parent conversations early during
failure analysis, and contended over monitor locks. Domain rules also
omitted guidance on executable target forwarding, Windows `.exe` CLI
output matching, and bidirectional alias cross-references.

Add `model_decision` frontmatter to unindexed rules and document those
Starlark, Windows, and documentation conventions. Update CI monitoring
and merge skills to ignore expected `blocks-do-not-merge` failures when
the merge-blocking label is active, scope failure analyzer alerts to
the analyzer subagent, and release active CI monitor locks before
launching the merge shepherd.
diff --git a/.agents/rules/cc.md b/.agents/rules/cc.md
index 5f2c2b4..c903a05 100644
--- a/.agents/rules/cc.md
+++ b/.agents/rules/cc.md
@@ -1,3 +1,8 @@
+---
+trigger: model_decision
+description: C/C++ rules design and Bazel rules_cc integration invariants
+---
+
 # C/C++ Rules Design & Bazel rules_cc Integration
 
 ## Concise `cc_shared_library` Interface & Export Rules
diff --git a/.agents/rules/docs.md b/.agents/rules/docs.md
index e463527..ab9318c 100644
--- a/.agents/rules/docs.md
+++ b/.agents/rules/docs.md
@@ -14,6 +14,11 @@
 * **API Docstrings vs. Narrative Docs**: Keep `.bzl` docstrings brief (purpose
   and arguments). Put setup guides and multi-scenario examples in `docs/`, and
   link to them with `{ref}` or `{obj}` instead of repeating rules.
+* **Bidirectional Alias Cross-References**: When documenting public `alias`
+  targets alongside a canonical target, cross-link in both directions:
+  reference the canonical target from each alias entry and list the aliases on
+  the canonical `{bzl:target}` entry (use `{obj}` for documented in-repo
+  targets and inline code for generated external-repo labels).
 * **How-to Guides (`docs/howto/`)**:
   * Prefer adding sections to an existing topical guide (e.g.
     `docs/howto/build-a-wheel.md`) over creating a separate granular file.
diff --git a/.agents/rules/starlark.md b/.agents/rules/starlark.md
index e4becec..167a908 100644
--- a/.agents/rules/starlark.md
+++ b/.agents/rules/starlark.md
@@ -42,6 +42,17 @@
   `allow_single_file = True`, when the attribute may accept `py_binary` (whose
   `DefaultInfo.files` has multiple files).
 
+## Forwarding Executables in `executable = True` Rules
+* When an `executable = True` rule forwards an underlying executable target
+  (from an attribute or resolved toolchain):
+  * Declare the output preserving `target_executable.extension` (e.g., `.exe`
+    on Windows) and materialize it with `ctx.actions.symlink(output = out,
+    target_file = target_executable, is_executable = True)`.
+  * Set `files = depset([out])`, and merge
+    `ctx.runfiles(files = [target_executable])` with the underlying target's
+    `default_runfiles` in `DefaultInfo(files = depset([out]), executable = out,
+    runfiles = ...)`.
+
 ## native.test_suite API
 * Don't forward target `tags` to `native.test_suite`; tags on test suites filter
   tests instead of setting target execution behavior.
diff --git a/.agents/rules/testing.md b/.agents/rules/testing.md
index 21a0191..59d3e22 100644
--- a/.agents/rules/testing.md
+++ b/.agents/rules/testing.md
@@ -1,3 +1,8 @@
+---
+trigger: model_decision
+description: Bazel test execution, integration test flags, and lockfiles
+---
+
 # Testing & Validation Guardrails
 
 ## `--config=fast-tests` for Test Targets
diff --git a/.agents/rules/windows.md b/.agents/rules/windows.md
index 271b5c6..ce8d0e6 100644
--- a/.agents/rules/windows.md
+++ b/.agents/rules/windows.md
@@ -1,3 +1,8 @@
+---
+trigger: model_decision
+description: Windows MSVC linking, ABI tags, MAX_PATH, and .exe portability
+---
+
 # Windows MSVC Toolchains & Linking
 
 ## MSVC `#pragma comment(lib, ...)` Embedded References
@@ -47,3 +52,6 @@
 * **RECORD File Paths**: In `RECORD` files, append `.bat` only to the
   shebang-rewritten scripts. Leave all other script paths unchanged.
 
+## CLI Help & Binary Name Output in Tests
+* Allow optional `.exe` suffixes when matching binary names in CLI output
+  assertions (e.g., `r"Usage: tool(\.exe)?\b"`).
diff --git a/.agents/skills/merge-pr/SKILL.md b/.agents/skills/merge-pr/SKILL.md
index 96304d1..1939d71 100644
--- a/.agents/skills/merge-pr/SKILL.md
+++ b/.agents/skills/merge-pr/SKILL.md
@@ -19,6 +19,10 @@
      `gh pr merge`; it fails when a merge queue is enabled on the repository.
 2. **Invoke a Background Shepherd**: Launch a background subagent with the role
    `Merge PR Shepherd` to continuously watch the PR until it merges.
+   - If a `monitor_remote_ci.py <pr_number>` background task is currently
+     running in the parent conversation, stop it via `manage_task` first so
+     `Merge PR Shepherd` can acquire `.monitored_pr_<pr_number>.lock` and
+     receive CI failure notifications directly.
 3. **Leverage Existing CI Skills**:
    - Have the subagent use the **`monitor-ci-results`** skill to watch for CI
      check failures and generate analysis reports.
diff --git a/.agents/skills/monitor-ci-results/SKILL.md b/.agents/skills/monitor-ci-results/SKILL.md
index c92bdda..c73295e 100644
--- a/.agents/skills/monitor-ci-results/SKILL.md
+++ b/.agents/skills/monitor-ci-results/SKILL.md
@@ -44,10 +44,15 @@
    with errors, `monitor_remote_ci.py` dispatches a high-priority notification
    message reporting the failed check back to your conversation.
    - Soft failures (e.g. `*rolling*` Bazel) are non-blocking warnings; ignore.
+   - If `blocks-do-not-merge` fails while the PR intentionally carries the
+     `"do not merge"` label, treat it as an expected merge gate rather than a
+     code failure.
 4. **Subagent Analysis**: Upon receiving a failure notification message from
    the monitoring script, immediately launch a separate subagent using the
    `invoke_subagent` tool with the role "CI Failure Analyzer" to run the
-   `analyze-ci-failure` skill on the reported failure.
+   `analyze-ci-failure` skill on the reported failure (instructing the
+   subagent to pass its own subagent conversation ID to
+   `analyze_ci_failure.py` so only its completion message wakes the parent).
 5. **Autonomous Flake Retry**: If the CI Failure Analyzer subagent confirms
    that the failure is a transient infrastructure or network flake (e.g. disk
    I/O error, 504 gateway, sandbox initialization failure, or `setup-bazel`
diff --git a/.agents/skills/monitor-ci-results/scripts/monitor_remote_ci.py b/.agents/skills/monitor-ci-results/scripts/monitor_remote_ci.py
index e8d3a02..15bc5c3 100755
--- a/.agents/skills/monitor-ci-results/scripts/monitor_remote_ci.py
+++ b/.agents/skills/monitor-ci-results/scripts/monitor_remote_ci.py
@@ -8,6 +8,7 @@
 import subprocess
 import sys
 import time
+import traceback
 import urllib.request
 
 
@@ -39,6 +40,29 @@
         return []
 
 
+def _has_do_not_merge_label(pr_number):
+    cmd = ["gh", "pr", "view", str(pr_number), "--json", "labels"]
+    try:
+        res = subprocess.run(cmd, capture_output=True, text=True, check=True)
+    except subprocess.CalledProcessError as e:
+        e.add_note(
+            "==================== STDOUT BEGIN ====================\n"
+            f"{e.stdout or ''}"
+            "==================== STDOUT END ====================\n"
+            "==================== STDERR BEGIN ====================\n"
+            f"{e.stderr or ''}"
+            "==================== STDERR END ===================="
+        )
+        raise
+    out = res.stdout
+    json_str = out[out.find("{") :] if "{" in out else out
+    data = json.loads(json_str)
+    return any(
+        label.get("name", "").lower() == "do not merge"
+        for label in data.get("labels", [])
+    )
+
+
 def get_buildkite_jobs(build_url):
     base_url = build_url.split("#")[0]
     if base_url.endswith(".json"):
@@ -329,7 +353,29 @@
             ):
                 job_id = link.split("/")[-1].split("?")[0] if "/" in link else link
                 gh_key = f"gh_{job_id}" if job_id else name
-                if gh_key not in monitored:
+                if gh_key in monitored:
+                    continue
+                skip_expected_dnm = False
+                if name == "blocks-do-not-merge":
+                    try:
+                        skip_expected_dnm = _has_do_not_merge_label(args.pr)
+                    except (
+                        OSError,
+                        subprocess.CalledProcessError,
+                        json.JSONDecodeError,
+                    ) as e:
+                        details = "".join(traceback.format_exception_only(e)).strip()
+                        print(
+                            f"⚠️ Error fetching PR labels: {details}",
+                            file=sys.stderr,
+                        )
+                        continue
+                if skip_expected_dnm:
+                    print(
+                        "ℹ️ Skipping expected 'blocks-do-not-merge' failure "
+                        "while 'do not merge' label is present."
+                    )
+                else:
                     print(f"🚨 Notifying failure for GitHub check '{name}'...")
                     msg = (
                         f"⚠️ Remote CI GitHub Check '{name}' completed with errors!\n\n"
@@ -347,9 +393,9 @@
                             msg,
                         ]
                     )
-                    monitored[gh_key] = time.time()
-                    with open(state_file, "w") as f:
-                        json.dump(monitored, f)
+                monitored[gh_key] = time.time()
+                with open(state_file, "w") as f:
+                    json.dump(monitored, f)
 
         time.sleep(args.interval)