@prateek_ai/agents-maker 1.0.0 → 1.0.2

package/tools/test_kit.py

@@ -1,363 +1,392 @@
-#!/usr/bin/env python3
-"""
-test_kit.py — Comprehensive edge-case test suite for agents-maker.
-
-Run from the repo root:
-    python tools/test_kit.py
-"""
-
-from __future__ import annotations
-
-import subprocess
-import sys
-from pathlib import Path
-
-sys.stdout.reconfigure(encoding="utf-8")
-sys.stderr.reconfigure(encoding="utf-8")
-ROOT = Path(__file__).resolve().parent.parent
-PY = sys.executable
-
-PASS_COUNT = 0
-FAIL_COUNT = 0
-RESULTS = []
-
-
-def run(cmd: list[str], cwd: Path = ROOT, input_text: str | None = None) -> tuple[int, str, str]:
-    result = subprocess.run(
-        cmd, cwd=str(cwd), capture_output=True, text=True,
-        input=input_text, encoding="utf-8", errors="replace"
-    )
-    return result.returncode, result.stdout, result.stderr
-
-
-def check(n: int, desc: str, passed: bool, reason: str = "") -> None:
-    global PASS_COUNT, FAIL_COUNT
-    if passed:
-        PASS_COUNT += 1
-        tag = f"PASS [{n:02d}]"
-    else:
-        FAIL_COUNT += 1
-        tag = f"FAIL [{n:02d}]"
-    msg = f"{tag} {desc}"
-    if not passed and reason:
-        msg += f"\n         Reason: {reason}"
-    print(msg)
-    RESULTS.append((n, passed, desc, reason))
-
-
-def section(title: str) -> None:
-    print(f"\n{'-' * 60}")
-    print(f"  {title}")
-    print('-' * 60)
-
-
-# -------------------------------------------------------------
-# A. generate_prompt.py — normal cases
-# -------------------------------------------------------------
-section("A. generate_prompt.py — normal cases")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting to the auth service"])
-check(1, "Basic run exits 0", rc == 0, err[:200] if rc != 0 else "")
-check(2, "Output contains ## Project Context", "## Project Context" in out, "")
-check(3, "Software domain auto-detected", "software" in out.lower(), out[:300])
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--phase", "implementation"])
-check(4, "--phase implementation -> output contains 'implementation'", "implementation" in out, "")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--phase", "review"])
-check(5, "--phase review -> output contains 'review'", "review" in out, "")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--full"])
-check(6, "--full flag -> output contains orchestrator content", "Orchestrator" in out or "orchestrator" in out, "")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--compress"])
-check(7, "--compress flag -> output contains ## Output Policy", "## Output Policy" in out, out[-300:])
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "[domain: content] write a blog post about GraphQL"])
-check(8, "[domain: content] prefix -> domain is content", "content" in out.lower(), "")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "[domain: ops_process] write a runbook for failover"])
-check(9, "[domain: ops_process] prefix -> execution_agent or ops_process in output", "ops_process" in out.lower() or "execution_agent" in out.lower(), out[:300])
-
-# -------------------------------------------------------------
-# B. generate_prompt.py — edge cases / error handling
-# -------------------------------------------------------------
-section("B. generate_prompt.py — edge cases / error handling")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", ""])
-check(10, "Empty problem -> non-zero exit", rc != 0, f"exit={rc}")
-check(11, "Empty problem -> stderr contains 'empty'", "empty" in err.lower(), err[:200])
-
-long_input = "x" * 5001
-rc, out, err = run([PY, "tools/generate_prompt.py", long_input])
-check(12, "Problem > 5000 chars -> non-zero exit", rc != 0, f"exit={rc}")
-check(13, "Problem > 5000 chars -> stderr contains 'too long'", "too long" in err.lower(), err[:200])
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "task", "--phase", "invalid_phase"])
-check(14, "Invalid --phase value -> non-zero exit", rc != 0, f"exit={rc}")
-
-valid_phases = ["task_framing", "requirements", "design", "solution_design",
-                "implementation", "implement", "review", "review_refinement", "handoff", "framing"]
-all_phases_ok = True
-failed_phase = ""
-for ph in valid_phases:
-    rc2, _, err2 = run([PY, "tools/generate_prompt.py", "some task", "--phase", ph])
-    if rc2 != 0:
-        all_phases_ok = False
-        failed_phase = f"phase={ph} exit={rc2} err={err2[:100]}"
-        break
-check(15, "All valid --phase values exit 0", all_phases_ok, failed_phase)
-
-# -------------------------------------------------------------
-# C. Domain detection — all 8 domains
-# -------------------------------------------------------------
-section("C. Domain detection — all 8 domains")
-
-domain_tests = [
-    (16, "software",       "Refactor the UserRepository class and add unit tests for the API endpoints"),
-    (17, "content",        "Write a technical blog post about our REST to GraphQL migration for developers"),
-    (18, "research",       "Write a literature review on transformer model architectures for NLP"),
-    (19, "data_analytics", "Build a sales funnel dashboard with conversion metrics and pipeline analytics"),
-    (20, "product_design", "Design a mobile onboarding flow for our e-commerce app with persona mapping"),
-    (21, "marketing",      "Create a go-to-market strategy and campaign brief for our SaaS product launch"),
-    (22, "ops_process",    "Write a runbook SOP for our database failover and disaster recovery procedure"),
-    (23, "general",        "I need help with something"),
-]
-
-for n, expected_domain, task in domain_tests:
-    rc, out, err = run([PY, "tools/generate_prompt.py", task])
-    combined = out.lower() + err.lower()
-    if expected_domain == "general":
-        # Low-confidence tasks fall back to project domain when project.yaml has primary_domain set.
-        # Accept either "general" OR "from-project" (scoring fell below threshold, used project default).
-        found = "general" in combined or "from-project" in combined
-    else:
-        found = expected_domain in combined
-    check(n, f"Domain detection -> {expected_domain}", found, f"task='{task[:50]}...' output snippet: {out[:200]}")
-
-# -------------------------------------------------------------
-# D. validate_kit.py — all 12 checks pass
-# -------------------------------------------------------------
-section("D. validate_kit.py — all 12 checks pass normally")
-
-rc, out, err = run([PY, "tools/validate_kit.py"])
-check(24, "validate_kit.py exits 0", rc == 0, err[:200])
-check(25, "ALL 12 checks PASSED in output", "ALL 12 checks PASSED" in out, out[-300:])
-
-# -------------------------------------------------------------
-# E. validate_kit.py — failure detection (temporarily corrupt files)
-# -------------------------------------------------------------
-section("E. validate_kit.py — failure detection")
-
-orch_path = ROOT / "agents" / "orchestrator.md"
-orch_backup = orch_path.read_text(encoding="utf-8")
-
-try:
-    orch_path.rename(ROOT / "agents" / "orchestrator.md.bak")
-    rc, out, err = run([PY, "tools/validate_kit.py"])
-    check(26, "Missing agent file -> FAIL in output", "FAIL" in out, out[-300:])
-finally:
-    bak = ROOT / "agents" / "orchestrator.md.bak"
-    if bak.exists():
-        bak.rename(orch_path)
-
-skill_path = ROOT / "skills" / "analyze_repo.md"
-skill_backup = skill_path.read_text(encoding="utf-8")
-try:
-    skill_path.write_text("short", encoding="utf-8")
-    rc, out, err = run([PY, "tools/validate_kit.py"])
-    check(27, "Stub skill file (< 50 chars) -> FAIL in output", "FAIL" in out, out[-300:])
-finally:
-    skill_path.write_text(skill_backup, encoding="utf-8")
-
-try:
-    skill_path.write_text("# Skill: test\n\nNo input or output section here.\n\nToken budget: low\n", encoding="utf-8")
-    rc, out, err = run([PY, "tools/validate_kit.py"])
-    check(28, "Skill missing input/output headings -> FAIL in output", "FAIL" in out, out[-300:])
-finally:
-    skill_path.write_text(skill_backup, encoding="utf-8")
-
-agent_path = ROOT / "agents" / "orchestrator.md"
-agent_backup = agent_path.read_text(encoding="utf-8")
-try:
-    agent_path.write_text("# Orchestrator\n\nSome content without required sections.\n" * 5, encoding="utf-8")
-    rc, out, err = run([PY, "tools/validate_kit.py"])
-    check(29, "Agent missing ## Role/Goals/Context -> FAIL in output", "FAIL" in out, out[-300:])
-finally:
-    agent_path.write_text(agent_backup, encoding="utf-8")
-
-# Confirm restore worked
-rc, out, err = run([PY, "tools/validate_kit.py"])
-check(30, "All files restored -> validator passes again", "ALL 12 checks PASSED" in out, out[-200:])
-
-# -------------------------------------------------------------
-# F. Compressor / PolicyLoader — direct import tests
-# -------------------------------------------------------------
-section("F. Compressor / PolicyLoader — direct import tests")
-
-sys.path.insert(0, str(ROOT))
-
-try:
-    from token_optimization.compressor import (
-        CompressionReport,
-        Compressor,
-        ContextBlock,
-        PolicyLoader,
-    )
-
-    loader = PolicyLoader(ROOT / "config" / "token_policies.yaml")
-    loader.load()
-    check(31, "PolicyLoader imports and loads without error", True, "")
-
-    for wf in ["feature_implementation", "code_review", "feature_design"]:
-        policy = loader.get_workflow_policy(wf)
-        ok_wf = policy.max_input_tokens > 0 and bool(policy.output_style)
-        check({"feature_implementation": 32, "code_review": 33, "feature_design": 34}[wf],
-              f"get_workflow_policy('{wf}') returns valid policy",
-              ok_wf, f"max_input_tokens={policy.max_input_tokens} output_style={policy.output_style}")
-
-    unknown_policy = loader.get_workflow_policy("completely_unknown_workflow_xyz")
-    check(35, "Unknown workflow -> fallback policy, no crash",
-          unknown_policy is not None and unknown_policy.max_input_tokens >= 0, "")
-
-    policy = loader.get_workflow_policy("feature_implementation")
-    compressor = Compressor(policy)
-    empty_block = ContextBlock()
-    compressed, report = compressor.compress(empty_block)
-    check(36, "compress(empty ContextBlock) returns without crash",
-          isinstance(compressed, str) and isinstance(report, CompressionReport), "")
-
-    block_with_state = ContextBlock(
-        project_state="Name: test-project\nStack: Python, FastAPI\nDomain: software",
-        active_query="add rate limiting to the auth service",
-        conversation_state="User asked about rate limiting. We discussed Redis sliding window.",
-    )
-    compressed2, report2 = compressor.compress(block_with_state)
-    check(37, "compress(ContextBlock with content) returns non-empty string",
-          isinstance(compressed2, str) and len(compressed2) > 0, f"len={len(compressed2)}")
-
-    loader_bad = PolicyLoader("/nonexistent/path/token_policies.yaml")
-    try:
-        loader_bad.load()
-        check(38, "PolicyLoader with bad path -> graceful fallback, no exception", True, "")
-    except Exception as e:
-        check(38, "PolicyLoader with bad path -> graceful fallback, no exception", False, str(e))
-
-except ImportError as e:
-    for n in range(31, 39):
-        check(n, "Compressor import test", False, str(e))
-
-# -------------------------------------------------------------
-# G. Context loaders — smoke tests
-# -------------------------------------------------------------
-section("G. Context loaders — smoke tests")
-
-rc, out, err = run([PY, "context_loaders/project_summary.py", "--path", str(ROOT)])
-check(39, "project_summary.py exits 0", rc == 0, err[:200])
-check(40, "project_summary.py output contains 'Stack' or 'Language'",
-      "stack" in out.lower() or "language" in out.lower(), out[:300])
-
-rc, out, err = run([PY, "context_loaders/repo_tree.py", "--path", str(ROOT)])
-check(41, "repo_tree.py exits 0", rc == 0, err[:200])
-check(42, "repo_tree.py output contains directory structure",
-      "agents" in out.lower() or "skills" in out.lower(), out[:300])
-
-rc, out, err = run([PY, "context_loaders/file_chunker.py", "--path", str(ROOT), "--files", "README.md"])
-check(43, "file_chunker.py exits 0", rc == 0, err[:200])
-check(44, "file_chunker.py output contains README content",
-      "agents-maker" in out.lower() or "readme" in out.lower(), out[:300])
-
-# -------------------------------------------------------------
-# H. init_project.py — smoke tests
-# -------------------------------------------------------------
-section("H. init_project.py — smoke tests")
-
-rc, out, err = run([PY, "tools/init_project.py", "--path", str(ROOT), "--update"])
-check(45, "init_project.py --update exits 0", rc == 0, err[:300])
-check(46, "system_prompt.md still exists after --update", (ROOT / "system_prompt.md").exists(), "")
-
-# -------------------------------------------------------------
-# I. generate_prompt.py — missing project.yaml
-# -------------------------------------------------------------
-section("I. generate_prompt.py — missing project.yaml")
-
-proj_yaml = ROOT / "config" / "project.yaml"
-proj_backup = None
-renamed = False
-
-if proj_yaml.exists():
-    proj_backup = proj_yaml.read_text(encoding="utf-8")
-    proj_yaml.rename(ROOT / "config" / "project.yaml.bak")
-    renamed = True
-
-try:
-    rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting"])
-    check(47, "Missing project.yaml -> tool still exits 0", rc == 0, err[:200])
-    check(48, "Missing project.yaml -> [WARN] printed to stderr", "[WARN]" in err, err[:200])
-finally:
-    bak = ROOT / "config" / "project.yaml.bak"
-    if bak.exists():
-        bak.rename(proj_yaml)
-    elif proj_backup and not proj_yaml.exists():
-        proj_yaml.write_text(proj_backup, encoding="utf-8")
-
-# -------------------------------------------------------------
-# J. system_prompt.md integrity
-# -------------------------------------------------------------
-section("J. system_prompt.md integrity")
-
-sp = ROOT / "system_prompt.md"
-check(49, "system_prompt.md exists", sp.exists(), "")
-if sp.exists():
-    sp_text = sp.read_text(encoding="utf-8")
-    check(50, "system_prompt.md > 10,000 chars", len(sp_text) > 10000, f"actual: {len(sp_text)}")
-    check(51, "system_prompt.md contains orchestrator content (## Role or ## Goals)",
-          "## role" in sp_text.lower() or "## goals" in sp_text.lower(), "")
-    check(52, "system_prompt.md contains version header",
-          "agents-maker system_prompt.md" in sp_text, sp_text[:200])
-    check(53, "system_prompt.md contains [Companion] instruction",
-          "[Companion]" in sp_text, sp_text[:500])
-else:
-    for n in range(50, 54):
-        check(n, "system_prompt.md content check", False, "file missing")
-
-# -------------------------------------------------------------
-# K. generate_prompt.py — --compress with all phases
-# -------------------------------------------------------------
-section("K. --compress with all phases")
-
-compress_phases = ["task_framing", "implementation", "review_refinement", "handoff", "solution_design"]
-for i, ph in enumerate(compress_phases):
-    rc, out, err = run([PY, "tools/generate_prompt.py", "some task", "--phase", ph, "--compress"])
-    n = 54 + i
-    check(n, f"--compress --phase {ph} -> exits 0 + Output Policy in output",
-          rc == 0 and "## Output Policy" in out,
-          f"exit={rc} stderr={err[:100]}")
-
-# -------------------------------------------------------------
-# L. --full + --compress combined
-# -------------------------------------------------------------
-section("L. --full + --compress combined")
-
-rc, out, err = run([PY, "tools/generate_prompt.py", "add feature", "--full", "--compress"])
-check(59, "--full --compress combined -> exits 0", rc == 0, err[:200])
-check(60, "--full --compress -> contains both system prompt and Output Policy",
-      ("Orchestrator" in out or "orchestrator" in out) and "## Output Policy" in out, "")
-
-# -------------------------------------------------------------
-# Summary
-# -------------------------------------------------------------
-total = PASS_COUNT + FAIL_COUNT
-print(f"\n{'=' * 60}")
-print(f"  RESULTS: {PASS_COUNT}/{total} tests passed | {FAIL_COUNT} failed")
-print('=' * 60)
-
-if FAIL_COUNT > 0:
-    print("\nFailed tests:")
-    for n, passed, desc, reason in RESULTS:
-        if not passed:
-            print(f"  [{n:02d}] {desc}")
-            if reason:
-                print(f"       {reason[:300]}")
-
-sys.exit(0 if FAIL_COUNT == 0 else 1)
+#!/usr/bin/env python3
+"""
+test_kit.py — Comprehensive edge-case test suite for agents-maker.
+
+Run from the repo root:
+    python tools/test_kit.py
+"""
+
+from __future__ import annotations
+
+import subprocess
+import sys
+from pathlib import Path
+
+sys.stdout.reconfigure(encoding="utf-8")
+sys.stderr.reconfigure(encoding="utf-8")
+ROOT = Path(__file__).resolve().parent.parent
+PY = sys.executable
+
+PASS_COUNT = 0
+FAIL_COUNT = 0
+RESULTS = []
+
+
+def run(cmd: list[str], cwd: Path = ROOT, input_text: str | None = None) -> tuple[int, str, str]:
+    result = subprocess.run(
+        cmd, cwd=str(cwd), capture_output=True, text=True,
+        input=input_text, encoding="utf-8", errors="replace"
+    )
+    return result.returncode, result.stdout, result.stderr
+
+
+def check(n: int, desc: str, passed: bool, reason: str = "") -> None:
+    global PASS_COUNT, FAIL_COUNT
+    if passed:
+        PASS_COUNT += 1
+        tag = f"PASS [{n:02d}]"
+    else:
+        FAIL_COUNT += 1
+        tag = f"FAIL [{n:02d}]"
+    msg = f"{tag} {desc}"
+    if not passed and reason:
+        msg += f"\n         Reason: {reason}"
+    print(msg)
+    RESULTS.append((n, passed, desc, reason))
+
+
+def section(title: str) -> None:
+    print(f"\n{'-' * 60}")
+    print(f"  {title}")
+    print('-' * 60)
+
+
+# -------------------------------------------------------------
+# A. generate_prompt.py — normal cases
+# -------------------------------------------------------------
+section("A. generate_prompt.py — normal cases")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting to the auth service"])
+check(1, "Basic run exits 0", rc == 0, err[:200] if rc != 0 else "")
+check(2, "Output contains ## Project Context", "## Project Context" in out, "")
+check(3, "Software domain auto-detected", "software" in out.lower(), out[:300])
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--phase", "implementation"])
+check(4, "--phase implementation -> output contains 'implementation'", "implementation" in out, "")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--phase", "review"])
+check(5, "--phase review -> output contains 'review'", "review" in out, "")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--full"])
+check(6, "--full flag -> output contains orchestrator content", "Orchestrator" in out or "orchestrator" in out, "")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting", "--compress"])
+check(7, "--compress flag -> output contains ## Output Policy", "## Output Policy" in out, out[-300:])
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "[domain: content] write a blog post about GraphQL"])
+check(8, "[domain: content] prefix -> domain is content", "content" in out.lower(), "")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "[domain: ops_process] write a runbook for failover"])
+check(9, "[domain: ops_process] prefix -> execution_agent or ops_process in output", "ops_process" in out.lower() or "execution_agent" in out.lower(), out[:300])
+
+# -------------------------------------------------------------
+# B. generate_prompt.py — edge cases / error handling
+# -------------------------------------------------------------
+section("B. generate_prompt.py — edge cases / error handling")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", ""])
+check(10, "Empty problem -> non-zero exit", rc != 0, f"exit={rc}")
+check(11, "Empty problem -> stderr contains 'empty'", "empty" in err.lower(), err[:200])
+
+long_input = "x" * 5001
+rc, out, err = run([PY, "tools/generate_prompt.py", long_input])
+check(12, "Problem > 5000 chars -> non-zero exit", rc != 0, f"exit={rc}")
+check(13, "Problem > 5000 chars -> stderr contains 'too long'", "too long" in err.lower(), err[:200])
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "task", "--phase", "invalid_phase"])
+check(14, "Invalid --phase value -> non-zero exit", rc != 0, f"exit={rc}")
+
+valid_phases = ["task_framing", "requirements", "design", "solution_design",
+                "implementation", "implement", "review", "review_refinement", "handoff", "framing"]
+all_phases_ok = True
+failed_phase = ""
+for ph in valid_phases:
+    rc2, _, err2 = run([PY, "tools/generate_prompt.py", "some task", "--phase", ph])
+    if rc2 != 0:
+        all_phases_ok = False
+        failed_phase = f"phase={ph} exit={rc2} err={err2[:100]}"
+        break
+check(15, "All valid --phase values exit 0", all_phases_ok, failed_phase)
+
+# -------------------------------------------------------------
+# C. Domain detection — all 8 domains
+# -------------------------------------------------------------
+section("C. Domain detection — all 8 domains")
+
+domain_tests = [
+    (16, "software",       "Refactor the UserRepository class and add unit tests for the API endpoints"),
+    (17, "content",        "Write a technical blog post about our REST to GraphQL migration for developers"),
+    (18, "research",       "Write a literature review on transformer model architectures for NLP"),
+    (19, "data_analytics", "Build a sales funnel dashboard with conversion metrics and pipeline analytics"),
+    (20, "product_design", "Design a mobile onboarding flow for our e-commerce app with persona mapping"),
+    (21, "marketing",      "Create a go-to-market strategy and campaign brief for our SaaS product launch"),
+    (22, "ops_process",    "Write a runbook SOP for our database failover and disaster recovery procedure"),
+    (23, "general",        "I need help with something"),
+]
+
+for n, expected_domain, task in domain_tests:
+    rc, out, err = run([PY, "tools/generate_prompt.py", task])
+    combined = out.lower() + err.lower()
+    if expected_domain == "general":
+        # Low-confidence tasks fall back to project domain when project.yaml has primary_domain set.
+        # Accept either "general" OR "from-project" (scoring fell below threshold, used project default).
+        found = "general" in combined or "from-project" in combined
+    else:
+        found = expected_domain in combined
+    check(n, f"Domain detection -> {expected_domain}", found, f"task='{task[:50]}...' output snippet: {out[:200]}")
+
+# -------------------------------------------------------------
+# D. validate_kit.py — all 12 checks pass
+# -------------------------------------------------------------
+section("D. validate_kit.py — all 12 checks pass normally")
+
+rc, out, err = run([PY, "tools/validate_kit.py"])
+check(24, "validate_kit.py exits 0", rc == 0, err[:200])
+check(25, "ALL 12 checks PASSED in output", "ALL 12 checks PASSED" in out, out[-300:])
+
+# -------------------------------------------------------------
+# E. validate_kit.py — failure detection (temporarily corrupt files)
+# -------------------------------------------------------------
+section("E. validate_kit.py — failure detection")
+
+orch_path = ROOT / "agents" / "orchestrator.md"
+orch_backup = orch_path.read_text(encoding="utf-8")
+
+try:
+    orch_path.rename(ROOT / "agents" / "orchestrator.md.bak")
+    rc, out, err = run([PY, "tools/validate_kit.py"])
+    check(26, "Missing agent file -> FAIL in output", "FAIL" in out, out[-300:])
+finally:
+    bak = ROOT / "agents" / "orchestrator.md.bak"
+    if bak.exists():
+        bak.rename(orch_path)
+
+skill_path = ROOT / "skills" / "analyze_repo.md"
+skill_backup = skill_path.read_text(encoding="utf-8")
+try:
+    skill_path.write_text("short", encoding="utf-8")
+    rc, out, err = run([PY, "tools/validate_kit.py"])
+    check(27, "Stub skill file (< 50 chars) -> FAIL in output", "FAIL" in out, out[-300:])
+finally:
+    skill_path.write_text(skill_backup, encoding="utf-8")
+
+try:
+    skill_path.write_text("# Skill: test\n\nNo input or output section here.\n\nToken budget: low\n", encoding="utf-8")
+    rc, out, err = run([PY, "tools/validate_kit.py"])
+    check(28, "Skill missing input/output headings -> FAIL in output", "FAIL" in out, out[-300:])
+finally:
+    skill_path.write_text(skill_backup, encoding="utf-8")
+
+agent_path = ROOT / "agents" / "orchestrator.md"
+agent_backup = agent_path.read_text(encoding="utf-8")
+try:
+    agent_path.write_text("# Orchestrator\n\nSome content without required sections.\n" * 5, encoding="utf-8")
+    rc, out, err = run([PY, "tools/validate_kit.py"])
+    check(29, "Agent missing ## Role/Goals/Context -> FAIL in output", "FAIL" in out, out[-300:])
+finally:
+    agent_path.write_text(agent_backup, encoding="utf-8")
+
+# Confirm restore worked
+rc, out, err = run([PY, "tools/validate_kit.py"])
+check(30, "All files restored -> validator passes again", "ALL 12 checks PASSED" in out, out[-200:])
+
+# -------------------------------------------------------------
+# F. Compressor / PolicyLoader — direct import tests
+# -------------------------------------------------------------
+section("F. Compressor / PolicyLoader — direct import tests")
+
+sys.path.insert(0, str(ROOT))
+
+try:
+    from token_optimization.compressor import (
+        CompressionReport,
+        Compressor,
+        ContextBlock,
+        PolicyLoader,
+    )
+
+    loader = PolicyLoader(ROOT / "config" / "token_policies.yaml")
+    loader.load()
+    check(31, "PolicyLoader imports and loads without error", True, "")
+
+    for wf in ["feature_implementation", "code_review", "feature_design"]:
+        policy = loader.get_workflow_policy(wf)
+        ok_wf = policy.max_input_tokens > 0 and bool(policy.output_style)
+        check({"feature_implementation": 32, "code_review": 33, "feature_design": 34}[wf],
+              f"get_workflow_policy('{wf}') returns valid policy",
+              ok_wf, f"max_input_tokens={policy.max_input_tokens} output_style={policy.output_style}")
+
+    unknown_policy = loader.get_workflow_policy("completely_unknown_workflow_xyz")
+    check(35, "Unknown workflow -> fallback policy, no crash",
+          unknown_policy is not None and unknown_policy.max_input_tokens >= 0, "")
+
+    policy = loader.get_workflow_policy("feature_implementation")
+    compressor = Compressor(policy)
+    empty_block = ContextBlock()
+    compressed, report = compressor.compress(empty_block)
+    check(36, "compress(empty ContextBlock) returns without crash",
+          isinstance(compressed, str) and isinstance(report, CompressionReport), "")
+
+    block_with_state = ContextBlock(
+        project_state="Name: test-project\nStack: Python, FastAPI\nDomain: software",
+        active_query="add rate limiting to the auth service",
+        conversation_state="User asked about rate limiting. We discussed Redis sliding window.",
+    )
+    compressed2, report2 = compressor.compress(block_with_state)
+    check(37, "compress(ContextBlock with content) returns non-empty string",
+          isinstance(compressed2, str) and len(compressed2) > 0, f"len={len(compressed2)}")
+
+    loader_bad = PolicyLoader("/nonexistent/path/token_policies.yaml")
+    try:
+        loader_bad.load()
+        check(38, "PolicyLoader with bad path -> graceful fallback, no exception", True, "")
+    except Exception as e:
+        check(38, "PolicyLoader with bad path -> graceful fallback, no exception", False, str(e))
+
+except ImportError as e:
+    for n in range(31, 39):
+        check(n, "Compressor import test", False, str(e))
+
+# -------------------------------------------------------------
+# G. Context loaders — smoke tests
+# -------------------------------------------------------------
+section("G. Context loaders — smoke tests")
+
+rc, out, err = run([PY, "context_loaders/project_summary.py", "--path", str(ROOT)])
+check(39, "project_summary.py exits 0", rc == 0, err[:200])
+check(40, "project_summary.py output contains 'Stack' or 'Language'",
+      "stack" in out.lower() or "language" in out.lower(), out[:300])
+
+rc, out, err = run([PY, "context_loaders/repo_tree.py", "--path", str(ROOT)])
+check(41, "repo_tree.py exits 0", rc == 0, err[:200])
+check(42, "repo_tree.py output contains directory structure",
+      "agents" in out.lower() or "skills" in out.lower(), out[:300])
+
+rc, out, err = run([PY, "context_loaders/file_chunker.py", "--path", str(ROOT), "--files", "README.md"])
+check(43, "file_chunker.py exits 0", rc == 0, err[:200])
+check(44, "file_chunker.py output contains README content",
+      "agents-maker" in out.lower() or "readme" in out.lower(), out[:300])
+
+# -------------------------------------------------------------
+# H. init_project.py — smoke tests
+# -------------------------------------------------------------
+section("H. init_project.py — smoke tests")
+
+rc, out, err = run([PY, "tools/init_project.py", "--path", str(ROOT), "--update"])
+check(45, "init_project.py --update exits 0", rc == 0, err[:300])
+check(46, "system_prompt.md still exists after --update", (ROOT / "system_prompt.md").exists(), "")
+
+# -------------------------------------------------------------
+# I. generate_prompt.py — missing project.yaml
+# -------------------------------------------------------------
+section("I. generate_prompt.py — missing project.yaml")
+
+proj_yaml = ROOT / "config" / "project.yaml"
+proj_backup = None
+renamed = False
+
+if proj_yaml.exists():
+    proj_backup = proj_yaml.read_text(encoding="utf-8")
+    proj_yaml.rename(ROOT / "config" / "project.yaml.bak")
+    renamed = True
+
+try:
+    rc, out, err = run([PY, "tools/generate_prompt.py", "add rate limiting"])
+    check(47, "Missing project.yaml -> tool still exits 0", rc == 0, err[:200])
+    check(48, "Missing project.yaml -> [WARN] printed to stderr", "[WARN]" in err, err[:200])
+finally:
+    bak = ROOT / "config" / "project.yaml.bak"
+    if bak.exists():
+        bak.rename(proj_yaml)
+    elif proj_backup and not proj_yaml.exists():
+        proj_yaml.write_text(proj_backup, encoding="utf-8")
+
+# -------------------------------------------------------------
+# J. system_prompt.md integrity
+# -------------------------------------------------------------
+section("J. system_prompt.md integrity")
+
+sp = ROOT / "system_prompt.md"
+check(49, "system_prompt.md exists", sp.exists(), "")
+if sp.exists():
+    sp_text = sp.read_text(encoding="utf-8")
+    check(50, "system_prompt.md > 10,000 chars", len(sp_text) > 10000, f"actual: {len(sp_text)}")
+    check(51, "system_prompt.md contains orchestrator content (## Role or ## Goals)",
+          "## role" in sp_text.lower() or "## goals" in sp_text.lower(), "")
+    check(52, "system_prompt.md contains version header",
+          "agents-maker system_prompt.md" in sp_text, sp_text[:200])
+    check(53, "system_prompt.md contains [Companion] instruction",
+          "[Companion]" in sp_text, sp_text[:500])
+else:
+    for n in range(50, 54):
+        check(n, "system_prompt.md content check", False, "file missing")
+
+# -------------------------------------------------------------
+# K. generate_prompt.py — --compress with all phases
+# -------------------------------------------------------------
+section("K. --compress with all phases")
+
+compress_phases = ["task_framing", "implementation", "review_refinement", "handoff", "solution_design"]
+for i, ph in enumerate(compress_phases):
+    rc, out, err = run([PY, "tools/generate_prompt.py", "some task", "--phase", ph, "--compress"])
+    n = 54 + i
+    check(n, f"--compress --phase {ph} -> exits 0 + Output Policy in output",
+          rc == 0 and "## Output Policy" in out,
+          f"exit={rc} stderr={err[:100]}")
+
+# -------------------------------------------------------------
+# L. --full + --compress combined
+# -------------------------------------------------------------
+section("L. --full + --compress combined")
+
+rc, out, err = run([PY, "tools/generate_prompt.py", "add feature", "--full", "--compress"])
+check(59, "--full --compress combined -> exits 0", rc == 0, err[:200])
+check(60, "--full --compress -> contains both system prompt and Output Policy",
+      ("Orchestrator" in out or "orchestrator" in out) and "## Output Policy" in out, "")
+
+# -------------------------------------------------------------
+# M. generate_claude_agents.py — subagents + slash commands
+# -------------------------------------------------------------
+section("M. generate_claude_agents.py — .claude/ subagents + commands")
+
+rc, out, err = run([PY, "tools/generate_claude_agents.py", "--template", "--dry-run"])
+check(61, "generate_claude_agents --dry-run exits 0", rc == 0, err[:200])
+check(62, "dry-run reports 20 files (10 agents + 10 commands)",
+      out.count("agents/") >= 10 and out.count("commands/") >= 10, out[:300])
+
+claude_agents_dir = ROOT / "claude" / "agents"
+claude_cmds_dir = ROOT / "claude" / "commands"
+agent_files = sorted(claude_agents_dir.glob("*.md")) if claude_agents_dir.is_dir() else []
+cmd_files = sorted(claude_cmds_dir.glob("*.md")) if claude_cmds_dir.is_dir() else []
+check(63, "committed claude/ templates: 10 agents + 10 commands",
+      len(agent_files) == 10 and len(cmd_files) == 10,
+      f"agents={len(agent_files)} commands={len(cmd_files)}")
+
+pp = claude_agents_dir / "planpro.md"
+pp_text = pp.read_text(encoding="utf-8") if pp.exists() else ""
+check(64, "subagent planpro.md has valid frontmatter (name/tools/model)",
+      "name: planpro" in pp_text and "tools:" in pp_text and "model: inherit" in pp_text,
+      pp_text[:200])
+
+brain_cmd = claude_cmds_dir / "brain.md"
+bc_text = brain_cmd.read_text(encoding="utf-8") if brain_cmd.exists() else ""
+check(65, "command brain.md has description + $ARGUMENTS",
+      "description:" in bc_text and "$ARGUMENTS" in bc_text, bc_text[:200])
+
+# -------------------------------------------------------------
+# Summary
+# -------------------------------------------------------------
+total = PASS_COUNT + FAIL_COUNT
+print(f"\n{'=' * 60}")
+print(f"  RESULTS: {PASS_COUNT}/{total} tests passed | {FAIL_COUNT} failed")
+print('=' * 60)
+
+if FAIL_COUNT > 0:
+    print("\nFailed tests:")
+    for n, passed, desc, reason in RESULTS:
+        if not passed:
+            print(f"  [{n:02d}] {desc}")
+            if reason:
+                print(f"       {reason[:300]}")
+
+sys.exit(0 if FAIL_COUNT == 0 else 1)