diff --git a/.github/scripts/agents_pr_meta_update_body.js b/.github/scripts/agents_pr_meta_update_body.js
index 3e987c10c..8d0bed771 100644
--- a/.github/scripts/agents_pr_meta_update_body.js
+++ b/.github/scripts/agents_pr_meta_update_body.js
@@ -706,7 +706,25 @@ function upsertBlock(body, marker, replacement) {
   const startIndex = body.indexOf(start);
   const endIndex = body.indexOf(end);
   if (startIndex !== -1 && endIndex !== -1 && endIndex > startIndex) {
-    return `${body.slice(0, startIndex)}${replacement}${body.slice(endIndex + end.length)}`;
+    // Replace the first marker pair in place
+    let result = `${body.slice(0, startIndex)}${replacement}${body.slice(endIndex + end.length)}`;
+
+    // Remove any duplicate marker pairs left by concurrent writers
+    let hadDuplicates = false;
+    let limit = 10;
+    while (limit-- > 0) {
+      const dupStart = result.indexOf(start, startIndex + replacement.length);
+      const dupEnd = result.indexOf(end, dupStart + start.length);
+      if (dupStart !== -1 && dupEnd !== -1 && dupEnd > dupStart) {
+        result = `${result.slice(0, dupStart)}${result.slice(dupEnd + end.length)}`;
+        hadDuplicates = true;
+      } else {
+        break;
+      }
+    }
+
+    // Collapse triple+ newlines left by removed blocks
+    return hadDuplicates ? result.replace(/\n{3,}/g, '\n\n') : result;
   }
 
   const trimmed = body.trimEnd();
diff --git a/.github/scripts/agents_verifier_context.js b/.github/scripts/agents_verifier_context.js
index 79eefbd09..e2e8195a1 100644
--- a/.github/scripts/agents_verifier_context.js
+++ b/.github/scripts/agents_verifier_context.js
@@ -378,28 +378,10 @@ async function buildVerifierContext({ github, context, core, ciWorkflows }) {
     };
   }
 
-  const baseRef = pr.base?.ref || '';
-  const defaultBranch = context.payload?.repository?.default_branch || DEFAULT_BRANCH;
-  if (baseRef && baseRef !== defaultBranch) {
-    const skipReason = `Pull request base ref ${baseRef} does not match default branch ${defaultBranch}; skipping verifier.`;
-    core?.notice?.(skipReason);
-    core?.setOutput?.('should_run', 'false');
-    core?.setOutput?.('skip_reason', skipReason);
-    core?.setOutput?.('pr_number', String(pr.number || ''));
-    core?.setOutput?.('issue_numbers', '[]');
-    core?.setOutput?.('pr_html_url', pr.html_url || '');
-    core?.setOutput?.('target_sha', pr.merge_commit_sha || pr.head?.sha || context.sha || '');
-    core?.setOutput?.('context_path', '');
-    core?.setOutput?.('acceptance_count', '0');
-    core?.setOutput?.('ci_results', '[]');
-    core?.setOutput?.('diff_summary_path', '');
-    core?.setOutput?.('diff_path', '');
-    core?.setOutput?.('chain_depth', '0');
-    return { shouldRun: false, reason: skipReason, ciResults: [] };
-  }
-
   const prDetails = await github.rest.pulls.get({ owner, repo, pull_number: pr.number });
   const pull = prDetails?.data || pr;
+  const baseRef = pull.base?.ref || pr.base?.ref || '';
+  const defaultBranch = context.payload?.repository?.default_branch || DEFAULT_BRANCH;
 
   if (isForkPullRequest(pull)) {
     const skipReason = 'Pull request is from a fork; skipping verifier.';
diff --git a/CLAUDE.md b/CLAUDE.md
index 1010de58c..685aac408 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -87,32 +87,6 @@ See `docs/INTEGRATION_GUIDE.md` in Workflows repo.
 - Fix in Workflows repo → sync will propagate here
 - Or request manual sync: `gh workflow run maint-68-sync-consumer-repos.yml --repo stranske/Workflows`
 
-## Slow Test Strategy
-
-This repo uses a `slow` pytest marker to keep PR Gate fast (~4 min) while
-ensuring heavyweight tests still run on Main CI.
-
-**How it works:**
-
-- `pyproject.toml` registers the `slow` marker.
-- `pr-00-gate.yml` passes `pytest_markers: "not release and not slow"` to
-  exclude slow tests from PR feedback.
-- `ci.yml` (Main CI) runs the **full** suite including slow tests.
-- Locally, `make test` also excludes slow tests for fast iteration.
-
-**Currently marked slow:**
-
-- `tests/pipeline/test_run_pipeline.py` — auto-marked by
-  `tests/pipeline/conftest.py` (each test runs the full pipeline with
-  real Excel fixtures, 100-170 s each).
-- `tests/spec/test_macro_spec_fixtures.py` — module-level
-  `pytestmark = pytest.mark.slow` (session fixture parses three NISA
-  workbooks, ~85 s).
-
-**Rule for agents:** any new test that parses real workbooks or calls
-`run_pipeline()` **must** be marked `@pytest.mark.slow`.  Omitting the
-marker will silently regress Gate duration for every open PR.
-
 ## Reference Implementation
 
 **Travel-Plan-Permission** is the reference consumer repo. When debugging:
diff --git a/scripts/langchain/capability_check.py b/scripts/langchain/capability_check.py
index 61269c690..ef64b889a 100755
--- a/scripts/langchain/capability_check.py
+++ b/scripts/langchain/capability_check.py
@@ -139,9 +139,7 @@ def _build_llm_config(
     issue_or_pr = (
         str(issue_number)
         if issue_number is not None
-        else env_issue
-        if env_issue.isdigit()
-        else "unknown"
+        else env_issue if env_issue.isdigit() else "unknown"
     )
     metadata = {
         "repo": repo,
diff --git a/scripts/langchain/issue_optimizer.py b/scripts/langchain/issue_optimizer.py
index 3a7d6b5ed..12d74802e 100755
--- a/scripts/langchain/issue_optimizer.py
+++ b/scripts/langchain/issue_optimizer.py
@@ -319,9 +319,7 @@ def _build_llm_config(
     issue_or_pr = (
         str(issue_number)
         if issue_number is not None
-        else env_issue
-        if env_issue.isdigit()
-        else "unknown"
+        else env_issue if env_issue.isdigit() else "unknown"
     )
     metadata = {
         "repo": repo,
diff --git a/tools/langchain_client.py b/tools/langchain_client.py
index 2e8ee842c..124f4197a 100644
--- a/tools/langchain_client.py
+++ b/tools/langchain_client.py
@@ -148,12 +148,6 @@ def _resolve_slots() -> list[SlotDefinition]:
     return _apply_slot_env_overrides(_load_slot_config())
 
 
-def list_slot_definitions() -> list[SlotDefinition]:
-    """Return configured LLM slot definitions after env overrides."""
-
-    return list(_resolve_slots())
-
-
 def get_provider_model_catalog() -> dict[str, set[str]]:
     """Return provider->model mappings from configured slots."""