claude-dev-env 1.59.0 → 1.61.0

package/skills/autoconverge/workflow/test_aggregate_runs.py

@@ -0,0 +1,134 @@
+"""Tests for aggregate_runs against the real wf_881252e6-700 fixture run tree."""
+
+import json
+import shutil
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+import aggregate_runs
+import render_report
+
+FIXTURE_DIR = Path(__file__).resolve().parent / "fixtures" / "wf_run"
+FIXTURE_RUN_ID = "wf_881252e6-700"
+FIXTURE_PR_NUMBER = 211
+FIXTURE_OWNER = "example-owner"
+FIXTURE_REPO = "example-repo"
+
+
+def _copy_fixture_run(projects_root: Path, session_name: str) -> Path:
+    """Copy the fixture run tree into projects_root/proj/<session> and return its journal."""
+    session_dir = projects_root / "proj" / session_name
+    shutil.copytree(FIXTURE_DIR, session_dir)
+    return session_dir / "workflows" / f"{FIXTURE_RUN_ID}.json"
+
+
+def _write_minimal_run(
+    projects_root: Path,
+    session_name: str,
+    run_id: str,
+    pr_number: int,
+    resolve_heads: int,
+) -> Path:
+    """Write a transcript-free autoconverge journal with N resolve-head entries."""
+    workflows_dir = projects_root / "proj" / session_name / "workflows"
+    workflows_dir.mkdir(parents=True)
+    progress = [
+        {
+            "index": each_index + 1,
+            "label": render_report.LABEL_RESOLVE_HEAD,
+            "agentId": None,
+        }
+        for each_index in range(resolve_heads)
+    ]
+    journal = {
+        "runId": run_id,
+        "timestamp": "2026-06-14T00:00:00.000Z",
+        "workflowName": "autoconverge",
+        "args": json.dumps(
+            {"owner": FIXTURE_OWNER, "repo": FIXTURE_REPO, "prNumber": pr_number}
+        ),
+        "result": {
+            "converged": False,
+            "rounds": resolve_heads,
+            "finalSha": "",
+            "blocker": None,
+        },
+        "workflowProgress": progress,
+    }
+    journal_path = workflows_dir / f"{run_id}.json"
+    journal_path.write_text(json.dumps(journal, indent=2), encoding="utf-8")
+    return journal_path
+
+
+def test_aggregate_single_run_reproduces_loader_counts(tmp_path: Path) -> None:
+    """Should reproduce the single-run finding, round, and fix counts when only one journal exists."""
+    projects_root = tmp_path / "projects"
+    seed_journal = _copy_fixture_run(projects_root, "session-a")
+    work_dir = tmp_path / "work"
+    work_dir.mkdir()
+
+    result = aggregate_runs.aggregate_pr_journals(
+        seed_journal, FIXTURE_OWNER, FIXTURE_REPO, FIXTURE_PR_NUMBER, work_dir
+    )
+
+    assert result.round_count == 4
+    assert len(result.findings) == 15
+    assert result.final_sha.startswith("7c2f420c")
+    run_data = render_report.load_run_data(result.combined_journal_path)
+    assert run_data.total_finding_count == 15
+    assert run_data.fix_commit_count == 2
+
+
+def test_aggregate_two_runs_sums_rounds(tmp_path: Path) -> None:
+    """Should sum resolve-head rounds across every journal for the same PR."""
+    projects_root = tmp_path / "projects"
+    seed_journal = _copy_fixture_run(projects_root, "session-a")
+    _write_minimal_run(projects_root, "session-b", "wf_extra-001", FIXTURE_PR_NUMBER, 2)
+    work_dir = tmp_path / "work"
+    work_dir.mkdir()
+
+    result = aggregate_runs.aggregate_pr_journals(
+        seed_journal, FIXTURE_OWNER, FIXTURE_REPO, FIXTURE_PR_NUMBER, work_dir
+    )
+
+    assert result.round_count == 6
+    assert len(result.findings) == 15
+    assert result.final_sha.startswith("7c2f420c")
+
+
+def test_force_remove_clears_existing_directory(tmp_path: Path) -> None:
+    """Should remove an existing directory tree without raising on any supported Python."""
+    target = tmp_path / "to-remove"
+    target.mkdir()
+    (target / "nested").mkdir()
+    (target / "nested" / "file.txt").write_text("content", encoding="utf-8")
+
+    aggregate_runs._force_remove(target)
+
+    assert not target.exists()
+
+
+def test_force_remove_handler_keyword_is_version_guarded() -> None:
+    """Should pick onexc on Python 3.12+ and onerror otherwise, never bare onexc."""
+    handler_keyword = aggregate_runs._select_rmtree_handler_keyword()
+
+    if sys.version_info >= (3, 12):
+        assert set(handler_keyword) == {"onexc"}
+    else:
+        assert set(handler_keyword) == {"onerror"}
+
+
+def test_discover_filters_by_owner_repo_and_pr_number(tmp_path: Path) -> None:
+    """Should return only journals whose args match the requested owner, repo, and PR."""
+    projects_root = tmp_path / "projects"
+    _copy_fixture_run(projects_root, "session-a")
+    _write_minimal_run(projects_root, "session-other", "wf_other-001", 999, 1)
+
+    discovered = aggregate_runs.discover_pr_journals(
+        projects_root, FIXTURE_OWNER, FIXTURE_REPO, FIXTURE_PR_NUMBER
+    )
+
+    assert len(discovered) == 1
+    assert discovered[0].stem == FIXTURE_RUN_ID

package/skills/autoconverge/workflow/test_convergence_summary.py

@@ -0,0 +1,132 @@
+"""Tests for convergence_summary.build_summary_prompt."""
+
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+import convergence_summary
+
+
+SAMPLE_FINDINGS = [
+    {
+        "severity": "P0",
+        "category": "bug",
+        "file": "hooks/blocking/destructive_command_blocker.py",
+        "line": 42,
+        "title": "launcher-wrapped delete runs with no prompt",
+        "detail": "timeout 5 bash -c 'rm -rf /etc' bypasses the confirmation gate.",
+    },
+    {
+        "severity": "P2",
+        "category": "code-standard",
+        "file": "hooks/blocking/destructive_command_blocker.py",
+        "line": 9,
+        "title": "comment claims a refusal the code does not perform",
+        "detail": "",
+    },
+]
+
+
+def test_prompt_carries_pr_coordinates_and_round_count() -> None:
+    """Should name the PR coordinates and the aggregated round count."""
+    prompt = convergence_summary.build_summary_prompt(
+        owner="jl-cmd",
+        repo="claude-code-config",
+        pr_number=581,
+        round_count=39,
+        findings=SAMPLE_FINDINGS,
+        fix_summaries=["renamed and annotated", "guarded the launcher"],
+        standards_note=None,
+        copilot_note=None,
+    )
+
+    assert "owner=jl-cmd repo=claude-code-config PR #581" in prompt
+    assert "https://github.com/jl-cmd/claude-code-config/pull/581" in prompt
+    assert "convergence in 39 round(s)" in prompt
+    assert "gh api repos/jl-cmd/claude-code-config/pulls/581" in prompt
+
+
+def test_prompt_numbers_every_finding_with_severity_and_location() -> None:
+    """Should list each aggregated finding numbered with severity, file, and line."""
+    prompt = convergence_summary.build_summary_prompt(
+        owner="jl-cmd",
+        repo="claude-code-config",
+        pr_number=581,
+        round_count=39,
+        findings=SAMPLE_FINDINGS,
+        fix_summaries=[],
+        standards_note=None,
+        copilot_note=None,
+    )
+
+    assert (
+        "1. [P0/bug] hooks/blocking/destructive_command_blocker.py:42 - "
+        "launcher-wrapped delete runs with no prompt" in prompt
+    )
+    assert "2. [P2/code-standard]" in prompt
+    assert "Per-round fix summaries:\nnone" in prompt
+
+
+def test_prompt_carries_notes_when_present() -> None:
+    """Should include the standards and copilot notes only when they are given."""
+    with_notes = convergence_summary.build_summary_prompt(
+        owner="o",
+        repo="r",
+        pr_number=1,
+        round_count=2,
+        findings=[],
+        fix_summaries=[],
+        standards_note="round 2 deferred a magic-value class",
+        copilot_note="Copilot out of quota",
+    )
+    without_notes = convergence_summary.build_summary_prompt(
+        owner="o",
+        repo="r",
+        pr_number=1,
+        round_count=2,
+        findings=[],
+        fix_summaries=[],
+        standards_note=None,
+        copilot_note=None,
+    )
+
+    assert "Deferred code-standard note: round 2 deferred a magic-value class" in (
+        with_notes
+    )
+    assert "Copilot gate note: Copilot out of quota" in with_notes
+    assert "Deferred code-standard note:" not in without_notes
+    assert "Copilot gate note:" not in without_notes
+
+
+def test_prompt_imposes_no_issue_class_cap() -> None:
+    """Should instruct one class per distinct kind with no fixed class ceiling."""
+    prompt = convergence_summary.build_summary_prompt(
+        owner="o",
+        repo="r",
+        pr_number=1,
+        round_count=1,
+        findings=[],
+        fix_summaries=[],
+        standards_note=None,
+        copilot_note=None,
+    )
+
+    assert "Cap to about 5 classes" not in prompt
+    assert "one class per distinct KIND of problem" in prompt
+
+
+def test_empty_findings_state_clean_run() -> None:
+    """Should render the clean-run sentence when no findings were caught."""
+    prompt = convergence_summary.build_summary_prompt(
+        owner="o",
+        repo="r",
+        pr_number=1,
+        round_count=1,
+        findings=[],
+        fix_summaries=[],
+        standards_note=None,
+        copilot_note=None,
+    )
+
+    assert "none - every lens was clean on a stable HEAD" in prompt