prizmkit 1.1.110 → 1.1.112

package/bundled/dev-pipeline/tests/test_generate_bootstrap_prompt.py

@@ -10,6 +10,7 @@ from pathlib import Path
 from generate_bootstrap_prompt import (
     ACTIVE_AGENT_PROMPTS,
     compute_feature_slug,
+    detect_test_commands,
     find_feature,
     format_acceptance_criteria,
     format_ac_checklist,
@@ -883,7 +884,8 @@ class TestDirectOrchestratorImplementationPrompts:
         assert FORBIDDEN_DEV_FIX_PLACEHOLDER not in replacements
         assert FORBIDDEN_DEV_RESUME_PLACEHOLDER not in replacements
         assert "{{AGENT_PROMPT_CRITIC_PLAN_CHALLENGE}}" in replacements
-        assert "{{AGENT_PROMPT_REVIEWER_REVIEW}}" in replacements
+        assert "{{AGENT_PROMPT_REVIEWER_REVIEW}}" not in replacements
+        assert (templates_dir / "agent-prompts" / "reviewer-review.md").exists() is False
         assert (templates_dir / "agent-prompts" / "dev-implement.md").exists() is False
         assert (templates_dir / "agent-prompts" / FORBIDDEN_DEV_FIX_PROMPT).exists() is False
         assert (templates_dir / "agent-prompts" / FORBIDDEN_DEV_RESUME_PROMPT).exists() is False
@@ -922,9 +924,19 @@ class TestDirectOrchestratorImplementationPrompts:
             assert "Implement — Orchestrator Direct" in rendered
             assert "Spawn Dev subagent" not in rendered
             assert "Spawn Dev agent" not in rendered
+            assert "Spawn Reviewer agent" not in rendered
+            assert "Reviewer Agent Def" not in rendered
+            assert "subagent_type=\"prizm-dev-team-reviewer\"" not in rendered
             assert "Implement — Dev" not in rendered
             assert "{{AGENT_PROMPT_DEV_IMPLEMENT}}" not in rendered
+            assert "{{AGENT_PROMPT_REVIEWER_REVIEW}}" not in rendered
+            assert "reviewer-review.md" not in rendered
             assert "dev-implement.md" not in rendered
+            assert "Bug Fix Documentation Policy" not in rendered
+            assert "For bug-fix sessions" not in rendered
+            assert "fix(<scope>)" not in rendered
+            assert "port check 3000" not in rendered
+            assert "localhost:3000" not in rendered
             assert FORBIDDEN_REVIEW_DEV_FIX not in rendered
             assert FORBIDDEN_REVIEWER_DEV_SPAWN not in rendered
             assert FORBIDDEN_EXPLORE_SUBAGENT not in rendered
@@ -1214,6 +1226,19 @@ class TestHeadlessPromptCleanupF033:
                 assert ".prizmkit-test-started" in prompt
                 assert "{{RUNTIME_HELPER_CMD}}" not in prompt
                 assert "prizmkit-runtime-helper.py" in prompt
+            # Anti-Bash guardrails for bugfix/refactor branches (F-039)
+            if "Bug ID" in prompt:
+                assert "{{RUNTIME_HELPER_CMD}}" not in prompt
+                assert "prizmkit-runtime-helper.py" in prompt
+                assert "2>/dev/null" not in prompt
+                assert "lsof -ti" not in prompt
+                assert "which " not in prompt  # no which-command shell glue
+            if "Refactor ID" in prompt:
+                assert "{{RUNTIME_HELPER_CMD}}" not in prompt
+                assert "prizmkit-runtime-helper.py" in prompt
+                assert "2>/dev/null" not in prompt
+                assert "lsof -ti" not in prompt
+                assert "which " not in prompt  # no which-command shell glue
             assert "python3 {{INIT_SCRIPT_PATH}}" not in prompt
 
     def test_legacy_feature_templates_are_clean_when_rendered(self, tmp_path):
@@ -1297,7 +1322,6 @@ class TestHeadlessPromptCleanupF033:
         refactor_prompt = _render_refactor_prompt(tmp_path / "refactor-mixed", platform="claude")
 
         assert replacements["{{REVIEWER_SUBAGENT_PATH}}"].endswith(".claude/agents/prizm-dev-team-reviewer.md")
-        assert ".claude/agents/prizm-dev-team-reviewer.md" in refactor_prompt
         assert ".codex/agents/prizm-dev-team-reviewer.toml" not in refactor_prompt
 
     def test_bugfix_refactor_explicit_codex_override(self, tmp_path, monkeypatch):
@@ -1319,7 +1343,6 @@ class TestHeadlessPromptCleanupF033:
         refactor_prompt = _render_refactor_prompt(tmp_path / "refactor-codex", platform="claude", explicit_platform="codex")
 
         assert replacements["{{REVIEWER_SUBAGENT_PATH}}"].endswith(".codex/agents/prizm-dev-team-reviewer.toml")
-        assert ".codex/agents/prizm-dev-team-reviewer.toml" in refactor_prompt
         assert ".claude/agents/prizm-dev-team-reviewer.md" not in refactor_prompt
 
     def test_explicit_codex_env_still_renders_codex_critic_path(self, tmp_path, monkeypatch):
@@ -1385,7 +1408,9 @@ class TestFeatureBootstrapShellExtraction:
             assert "executable version opencli --arg=--version" in prompt
             assert "process start --pid-file .prizmkit/state/browser-dev-server.pid" in prompt
             assert "process cleanup-pid <pid-from-.prizmkit/state/browser-dev-server.pid>" in prompt
-            assert "port check <DEV_PORT>" in prompt or "port check 3000" in prompt
+            assert "port check <DEV_PORT>" in prompt
+            assert "port check 3000" not in prompt
+            assert "localhost:3000" not in prompt
 
     def test_rendered_legacy_tier_feature_prompts_use_runtime_helper_not_extracted_bash(self, tmp_path):
         cases = [
@@ -1405,7 +1430,8 @@ class TestFeatureBootstrapShellExtraction:
             self.assert_shell_extracted_operations_removed(prompt)
             assert "process start --pid-file .prizmkit/state/browser-dev-server.pid" in prompt
             assert "process cleanup-pid <pid-from-.prizmkit/state/browser-dev-server.pid>" in prompt
-            assert "port check 3000" in prompt
+            assert "port check 3000" not in prompt
+            assert "localhost:3000" not in prompt
 
     def test_legacy_tier_specific_extracted_operations_map_to_helper_commands(self, tmp_path):
         tier1 = _render_feature_prompt(
@@ -1436,3 +1462,94 @@ class TestFeatureBootstrapShellExtraction:
         assert "text search . --pattern \"123-prompt-cleanup\" --json" in tier3
         assert "text count .prizmkit/specs/123-prompt-cleanup/plan.md --pattern \"- [ ]\"" in tier3
         assert "Check if feature already committed by inspecting recent git history" in tier3
+
+
+# ---------------------------------------------------------------------------
+# detect_test_commands
+# ---------------------------------------------------------------------------
+
+import json as _json
+
+class TestDetectTestCommands:
+    def test_no_package_json_returns_empty(self, tmp_path):
+        """When no package.json exists, no npm test emitted."""
+        result = detect_test_commands(str(tmp_path))
+        assert "npm test" not in result
+        assert "pnpm test" not in result
+        assert "yarn test" not in result
+
+    def test_package_json_no_test_script(self, tmp_path):
+        """package.json exists but has no test script → no npm test emitted."""
+        pkg = tmp_path / "package.json"
+        pkg.write_text(_json.dumps({"name": "test", "scripts": {"start": "node index.js"}}), encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        # Bug: current code returns "npm test" even with no test script
+        assert "npm test" not in result, (
+            "Should NOT emit npm test when package.json has no test script"
+        )
+
+    def test_package_json_with_test_script(self, tmp_path):
+        """package.json with scripts.test → emits npm test."""
+        pkg = tmp_path / "package.json"
+        pkg.write_text(_json.dumps({"name": "test", "scripts": {"test": "vitest run"}}), encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        # Current code returns "npm test" (hardcoded), but at minimum it should
+        # still return something test-like. After fix, it becomes "npm test"
+        assert "npm test" in result, (
+            "Should emit npm test when package.json has test script"
+        )
+
+    def test_package_json_non_test_scripts_only(self, tmp_path):
+        """package.json exists but scripts only has non-test entries."""
+        pkg = tmp_path / "package.json"
+        pkg.write_text(_json.dumps({"name": "test", "scripts": {"build": "tsc", "start": "node dist/index.js"}}), encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        assert "npm test" not in result, (
+            "Should NOT emit npm test when no test-related scripts exist"
+        )
+
+    def test_pnpm_detected_from_lockfile(self, tmp_path):
+        """pnpm-lock.yaml → emits pnpm test when test script exists."""
+        pkg = tmp_path / "package.json"
+        pkg.write_text(_json.dumps({"name": "test", "scripts": {"test": "vitest"}}), encoding="utf-8")
+        (tmp_path / "pnpm-lock.yaml").write_text("", encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        assert "pnpm test" in result
+
+    def test_yarn_detected_from_lockfile(self, tmp_path):
+        """yarn.lock → emits yarn test when test script exists."""
+        pkg = tmp_path / "package.json"
+        pkg.write_text(_json.dumps({"name": "test", "scripts": {"test": "jest"}}), encoding="utf-8")
+        (tmp_path / "yarn.lock").write_text("", encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        assert "yarn test" in result
+
+    def test_corrupt_package_json_graceful_fallback(self, tmp_path):
+        """Corrupt/unparseable package.json → skip JS detection gracefully."""
+        (tmp_path / "package.json").write_text("not valid json {{{", encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        # Should not crash, should not emit npm test
+        assert "npm test" not in result
+
+    def test_non_js_detections_preserved(self, tmp_path):
+        """Go, Rust, Python, Makefile detections unchanged."""
+        (tmp_path / "go.mod").write_text("", encoding="utf-8")
+        (tmp_path / "Cargo.toml").write_text("", encoding="utf-8")
+        (tmp_path / "setup.py").write_text("", encoding="utf-8")
+        makefile = tmp_path / "Makefile"
+        makefile.write_text("test:\n\tpytest\n", encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        assert "go test ./..." in result
+        assert "cargo test" in result
+        assert "pytest" in result
+        assert "make test" in result
+
+    def test_package_json_test_unit_script_only(self, tmp_path):
+        """package.json with test:unit but no test → fallback to available test script."""
+        pkg = tmp_path / "package.json"
+        pkg.write_text(_json.dumps({"name": "test", "scripts": {"test:unit": "vitest run"}}), encoding="utf-8")
+        (tmp_path / "package-lock.json").write_text("", encoding="utf-8")
+        result = detect_test_commands(str(tmp_path))
+        # Should detect test:unit as a viable test script
+        assert "npm run test:unit" in result
+

package/bundled/dev-pipeline/tests/test_generate_bugfix_prompt.py

@@ -2,6 +2,8 @@
 
 from argparse import Namespace
 from pathlib import Path
+import json
+import os
 
 from generate_bugfix_prompt import (
     build_replacements,
@@ -303,3 +305,123 @@ def test_template_source_is_clean_for_headless_prompt_contract():
     content = Path("dev-pipeline/templates/bugfix-bootstrap-prompt.md").read_text(encoding="utf-8")
     for forbidden in HEADLESS_FORBIDDEN_PROMPT_STRINGS:
         assert forbidden not in content
+
+
+# ---------------------------------------------------------------------------
+# Anti-Bash regression guardrails (F-039)
+# Note: grep -n, sed -n, grep -q, wc -l excluded — referenced in bugfix context
+# budget rules as AI tool hints, not as operational shell snippets.
+# | sed  (trailing space) only matches pipe-to-sed operational patterns.
+# ---------------------------------------------------------------------------
+
+RETIRED_EXTRACTED_SNIPPETS_BUGFIX = [
+    "find . -maxdepth",
+    "| sed ",
+    r"grep -c '^\- \[ \]'",
+    "2>/dev/null",
+    "/dev/null",
+    "$DEV_PORT",
+    "$!",
+    "$HOME",
+    "if [ ",
+    "which playwright-cli",
+    "which opencli",
+    "lsof -ti",
+    "xargs",
+    " kill -",
+    "kill -",
+    "|| echo",
+    "mkdir -p",
+    "touch .prizmkit",
+]
+
+
+def _render_bugfix_prompt(tmp_path, platform="claude"):
+    """Render a complete bugfix prompt through the generator."""
+    from generate_bugfix_prompt import main as bugfix_main
+
+    project = tmp_path / "bugfix-anti-bash"
+    project.mkdir(parents=True)
+    (project / ".prizmkit").mkdir(exist_ok=True)
+    (project / ".prizmkit" / "config.json").write_text(
+        json.dumps({"platform": platform, "ai_cli": platform}), encoding="utf-8"
+    )
+    _ensure_agent_files(project)
+    bug_list = project / ".prizmkit" / "plans" / "bug-fix-list.json"
+    bug_list.parent.mkdir(parents=True)
+    bug_list.write_text(json.dumps({
+        "bugs": [{"id": "B-001", "title": "Broken", "description": "Fix it",
+                   "acceptance_criteria": ["Works"]}],
+        "global_context": {"language": "Python"},
+    }), encoding="utf-8")
+    output = project / "bugfix-prompt.md"
+    old_argv = os.sys.argv
+    old_platform = os.environ.get("PRIZMKIT_PLATFORM")
+    try:
+        os.environ.pop("PRIZMKIT_PLATFORM", None)
+        os.sys.argv = [
+            "generate-bugfix-prompt.py",
+            "--bug-list", str(bug_list),
+            "--bug-id", "B-001",
+            "--session-id", "session-1",
+            "--run-id", "run-1",
+            "--retry-count", "0",
+            "--resume-phase", "null",
+            "--state-dir", str(project / ".prizmkit" / "state" / "bugfix"),
+            "--project-root", str(project),
+            "--output", str(output),
+        ]
+        try:
+            bugfix_main()
+        except SystemExit as exc:
+            if exc.code not in (0, None):
+                raise
+    finally:
+        os.sys.argv = old_argv
+        if old_platform is None:
+            os.environ.pop("PRIZMKIT_PLATFORM", None)
+        else:
+            os.environ["PRIZMKIT_PLATFORM"] = old_platform
+    return output.read_text(encoding="utf-8")
+
+
+def _ensure_agent_files(project, platforms=None):
+    """Create minimal agent files so platform resolution succeeds."""
+    if platforms is None:
+        platforms = ("claude", "codex")
+    for plat in platforms:
+        platform_dir = ".codex" if plat == "codex" else ".claude"
+        suffix = ".toml" if plat == "codex" else ".md"
+        agents_dir = project / platform_dir / "agents"
+        agents_dir.mkdir(parents=True, exist_ok=True)
+        for name in ("dev", "reviewer", "critic"):
+            (agents_dir / f"prizm-dev-team-{name}{suffix}").write_text("agent\n", encoding="utf-8")
+
+
+class TestBugfixBootstrapShellExtraction:
+    """Regression guardrails: bugfix rendered prompts and source use helper, not shell."""
+
+    def test_rendered_bugfix_prompt_uses_runtime_helper_not_extracted_bash(self, tmp_path):
+        prompt = _render_bugfix_prompt(tmp_path)
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_BUGFIX:
+            assert snippet not in prompt, f"Retired shell snippet found in bugfix prompt: {snippet}"
+        assert "{{RUNTIME_HELPER_CMD}}" not in prompt
+        assert "python" in prompt.lower()
+        assert ".prizmkit/dev-pipeline/scripts/prizmkit-runtime-helper.py" in prompt
+        assert "artifact ensure-dir .prizmkit/bugfix/B-001" in prompt
+        assert "text contains .prizmkit/bugfix/B-001/review-report.md --pattern \"## Verdict\"" in prompt
+
+    def test_bugfix_template_source_is_clean_of_helper_covered_snippets(self):
+        content = Path("dev-pipeline/templates/bugfix-bootstrap-prompt.md").read_text(encoding="utf-8")
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_BUGFIX:
+            assert snippet not in content, f"Shell snippet in bugfix template: {snippet}"
+
+    def test_continuation_source_is_clean_of_helper_covered_snippets(self):
+        content = Path("dev-pipeline/scripts/continuation.py").read_text(encoding="utf-8")
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_BUGFIX:
+            assert snippet not in content, f"Shell snippet in continuation.py: {snippet}"
+
+    def test_bugfix_generator_source_is_clean_of_helper_covered_snippets(self):
+        content = Path("dev-pipeline/scripts/generate-bugfix-prompt.py").read_text(encoding="utf-8")
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_BUGFIX:
+            assert snippet not in content, f"Shell snippet in generate-bugfix-prompt.py: {snippet}"

package/bundled/dev-pipeline/tests/test_generate_refactor_prompt.py

@@ -2,6 +2,7 @@
 
 from argparse import Namespace
 from pathlib import Path
+import os
 
 from continuation import append_continuation_handoff
 from generate_refactor_prompt import (
@@ -194,3 +195,126 @@ def test_template_source_is_clean_for_headless_prompt_contract():
     content = Path("dev-pipeline/templates/refactor-bootstrap-prompt.md").read_text(encoding="utf-8")
     for forbidden in HEADLESS_FORBIDDEN_PROMPT_STRINGS:
         assert forbidden not in content
+
+
+# ---------------------------------------------------------------------------
+# Anti-Bash regression guardrails (F-039)
+# Note: grep -n, sed -n, grep -q, wc -l excluded — may appear in cross-pipeline
+# context budget rules as AI tool hints, not as operational shell snippets.
+# | sed  (trailing space) only matches pipe-to-sed operational patterns.
+# ---------------------------------------------------------------------------
+
+RETIRED_EXTRACTED_SNIPPETS_REFACTOR = [
+    "find . -maxdepth",
+    "| sed ",
+    r"grep -c '^\- \[ \]'",
+    "2>/dev/null",
+    "/dev/null",
+    "$DEV_PORT",
+    "$!",
+    "$HOME",
+    "if [ ",
+    "which playwright-cli",
+    "which opencli",
+    "lsof -ti",
+    "xargs",
+    " kill -",
+    "kill -",
+    "|| echo",
+    "mkdir -p",
+    "touch .prizmkit",
+]
+
+
+def _render_refactor_prompt_full(tmp_path, platform="claude"):
+    """Render a complete refactor prompt through the generator."""
+    import json, os
+    from generate_refactor_prompt import main as refactor_main
+
+    project = tmp_path / "refactor-anti-bash"
+    project.mkdir(parents=True)
+    (project / ".prizmkit").mkdir(exist_ok=True)
+    (project / ".prizmkit" / "config.json").write_text(
+        json.dumps({"platform": platform, "ai_cli": platform}), encoding="utf-8"
+    )
+    # Ensure agent files
+    for plat in ("claude", "codex"):
+        platform_dir = ".codex" if plat == "codex" else ".claude"
+        suffix = ".toml" if plat == "codex" else ".md"
+        agents_dir = project / platform_dir / "agents"
+        agents_dir.mkdir(parents=True, exist_ok=True)
+        for name in ("dev", "reviewer", "critic"):
+            (agents_dir / f"prizm-dev-team-{name}{suffix}").write_text("agent\n", encoding="utf-8")
+
+    refactor_list = project / ".prizmkit" / "plans" / "refactor-list.json"
+    refactor_list.parent.mkdir(parents=True)
+    refactor_list.write_text(json.dumps({
+        "refactors": [{
+            "id": "R-001", "title": "Restructure", "description": "Preserve behavior",
+            "acceptance_criteria": ["Behavior preserved"],
+            "scope": {"files": ["src/a.py"], "modules": ["src"]},
+        }],
+        "global_context": {"language": "Python"},
+    }), encoding="utf-8")
+    output = project / "refactor-prompt.md"
+    old_argv = os.sys.argv
+    old_platform = os.environ.get("PRIZMKIT_PLATFORM")
+    try:
+        os.environ.pop("PRIZMKIT_PLATFORM", None)
+        os.sys.argv = [
+            "generate-refactor-prompt.py",
+            "--refactor-list", str(refactor_list),
+            "--refactor-id", "R-001",
+            "--session-id", "session-1",
+            "--run-id", "run-1",
+            "--retry-count", "0",
+            "--resume-phase", "null",
+            "--state-dir", str(project / ".prizmkit" / "state" / "refactor"),
+            "--output", str(output),
+        ]
+        old_cwd = os.getcwd()
+        os.chdir(project)
+        try:
+            try:
+                refactor_main()
+            except SystemExit as exc:
+                if exc.code not in (0, None):
+                    raise
+        finally:
+            os.chdir(old_cwd)
+    finally:
+        os.sys.argv = old_argv
+        if old_platform is None:
+            os.environ.pop("PRIZMKIT_PLATFORM", None)
+        else:
+            os.environ["PRIZMKIT_PLATFORM"] = old_platform
+    return output.read_text(encoding="utf-8")
+
+
+class TestRefactorBootstrapShellExtraction:
+    """Regression guardrails: refactor rendered prompts and source use helper, not shell."""
+
+    def test_rendered_refactor_prompt_uses_runtime_helper_not_extracted_bash(self, tmp_path):
+        prompt = _render_refactor_prompt_full(tmp_path)
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_REFACTOR:
+            assert snippet not in prompt, f"Retired shell snippet found in refactor prompt: {snippet}"
+        assert "{{RUNTIME_HELPER_CMD}}" not in prompt
+        assert "python" in prompt.lower()
+        assert ".prizmkit/dev-pipeline/scripts/prizmkit-runtime-helper.py" in prompt
+        assert "artifact ensure-dir .prizmkit/refactor/R-001" in prompt
+        assert "text contains .prizmkit/refactor/R-001/review-report.md --pattern \"## Verdict\"" in prompt
+
+    def test_refactor_template_source_is_clean_of_helper_covered_snippets(self):
+        content = Path("dev-pipeline/templates/refactor-bootstrap-prompt.md").read_text(encoding="utf-8")
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_REFACTOR:
+            assert snippet not in content, f"Shell snippet in refactor template: {snippet}"
+
+    def test_continuation_source_is_clean_of_helper_covered_snippets(self):
+        content = Path("dev-pipeline/scripts/continuation.py").read_text(encoding="utf-8")
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_REFACTOR:
+            assert snippet not in content, f"Shell snippet in continuation.py: {snippet}"
+
+    def test_refactor_generator_source_is_clean_of_helper_covered_snippets(self):
+        content = Path("dev-pipeline/scripts/generate-refactor-prompt.py").read_text(encoding="utf-8")
+        for snippet in RETIRED_EXTRACTED_SNIPPETS_REFACTOR:
+            assert snippet not in content, f"Shell snippet in generate-refactor-prompt.py: {snippet}"