kodelyth-ecc 2.19.0 → 2.21.0

package/.cursor/skills/skill-comply/scripts/classifier.py

@@ -1,85 +0,0 @@
-"""Classify tool calls against compliance steps using LLM."""
-
-from __future__ import annotations
-
-import json
-import logging
-import subprocess
-from pathlib import Path
-
-logger = logging.getLogger(__name__)
-
-from scripts.parser import ComplianceSpec, ObservationEvent
-
-PROMPTS_DIR = Path(__file__).parent.parent / "prompts"
-
-
-def classify_events(
-    spec: ComplianceSpec,
-    trace: list[ObservationEvent],
-    model: str = "haiku",
-) -> dict[str, list[int]]:
-    """Classify which tool calls match which compliance steps.
-
-    Returns {step_id: [event_indices]} via a single LLM call.
-    """
-    if not trace:
-        return {}
-
-    steps_desc = "\n".join(
-        f"- {step.id}: {step.detector.description}"
-        for step in spec.steps
-    )
-
-    tool_calls = "\n".join(
-        f"[{i}] {event.tool}: input={event.input[:500]} output={event.output[:200]}"
-        for i, event in enumerate(trace)
-    )
-
-    prompt_template = (PROMPTS_DIR / "classifier.md").read_text()
-    prompt = (
-        prompt_template
-        .replace("{steps_description}", steps_desc)
-        .replace("{tool_calls}", tool_calls)
-    )
-
-    result = subprocess.run(
-        ["claude", "-p", prompt, "--model", model, "--output-format", "text"],
-        capture_output=True,
-        text=True,
-        timeout=60,
-    )
-
-    if result.returncode != 0:
-        raise RuntimeError(
-            f"classifier subprocess failed (rc={result.returncode}): "
-            f"{result.stderr[:500]}"
-        )
-
-    return _parse_classification(result.stdout)
-
-
-def _parse_classification(text: str) -> dict[str, list[int]]:
-    """Parse LLM classification output into {step_id: [event_indices]}."""
-    text = text.strip()
-    # Strip markdown fences
-    lines = text.splitlines()
-    if lines and lines[0].startswith("```"):
-        lines = lines[1:]
-    if lines and lines[-1].startswith("```"):
-        lines = lines[:-1]
-    cleaned = "\n".join(lines)
-
-    try:
-        parsed = json.loads(cleaned)
-        if not isinstance(parsed, dict):
-            logger.warning("Classifier returned non-dict JSON: %s", type(parsed).__name__)
-            return {}
-        return {
-            k: [int(i) for i in v]
-            for k, v in parsed.items()
-            if isinstance(v, list)
-        }
-    except (json.JSONDecodeError, ValueError, TypeError) as e:
-        logger.warning("Failed to parse classification output: %s", e)
-        return {}

package/.cursor/skills/skill-comply/scripts/grader.py

@@ -1,122 +0,0 @@
-"""Grade observation traces against compliance specs using LLM classification."""
-
-from __future__ import annotations
-
-from dataclasses import dataclass
-
-from scripts.classifier import classify_events
-from scripts.parser import ComplianceSpec, ObservationEvent, Step
-
-
-@dataclass(frozen=True)
-class StepResult:
-    step_id: str
-    detected: bool
-    evidence: tuple[ObservationEvent, ...]
-    failure_reason: str | None
-
-
-@dataclass(frozen=True)
-class ComplianceResult:
-    spec_id: str
-    steps: tuple[StepResult, ...]
-    compliance_rate: float
-    recommend_hook_promotion: bool
-    classification: dict[str, list[int]]
-
-
-def _check_temporal_order(
-    step: Step,
-    event: ObservationEvent,
-    resolved: dict[str, list[ObservationEvent]],
-    classified: dict[str, list[ObservationEvent]],
-) -> str | None:
-    """Check before_step/after_step constraints. Returns failure reason or None."""
-    if step.detector.after_step is not None:
-        after_events = resolved.get(step.detector.after_step, [])
-        if not after_events:
-            return f"after_step '{step.detector.after_step}' not yet detected"
-        latest_after = max(e.timestamp for e in after_events)
-        if event.timestamp <= latest_after:
-            return (
-                f"must occur after '{step.detector.after_step}' "
-                f"(last at {latest_after}), but found at {event.timestamp}"
-            )
-
-    if step.detector.before_step is not None:
-        # Look ahead using LLM classification results
-        before_events = resolved.get(step.detector.before_step)
-        if before_events is None:
-            before_events = classified.get(step.detector.before_step, [])
-        if before_events:
-            earliest_before = min(e.timestamp for e in before_events)
-            if event.timestamp >= earliest_before:
-                return (
-                    f"must occur before '{step.detector.before_step}' "
-                    f"(first at {earliest_before}), but found at {event.timestamp}"
-                )
-
-    return None
-
-
-def grade(
-    spec: ComplianceSpec,
-    trace: list[ObservationEvent],
-    classifier_model: str = "haiku",
-) -> ComplianceResult:
-    """Grade a trace against a compliance spec using LLM classification."""
-    sorted_trace = sorted(trace, key=lambda e: e.timestamp)
-
-    # Step 1: LLM classifies all events in one batch call
-    classification = classify_events(spec, sorted_trace, model=classifier_model)
-
-    # Convert indices to events
-    classified: dict[str, list[ObservationEvent]] = {
-        step_id: [sorted_trace[i] for i in indices if 0 <= i < len(sorted_trace)]
-        for step_id, indices in classification.items()
-    }
-
-    # Step 2: Check temporal ordering (deterministic)
-    resolved: dict[str, list[ObservationEvent]] = {}
-    step_results: list[StepResult] = []
-
-    for step in spec.steps:
-        candidates = classified.get(step.id, [])
-        matched: list[ObservationEvent] = []
-        failure_reason: str | None = None
-
-        for event in candidates:
-            temporal_fail = _check_temporal_order(step, event, resolved, classified)
-            if temporal_fail is None:
-                matched.append(event)
-                break
-            else:
-                failure_reason = temporal_fail
-
-        detected = len(matched) > 0
-        if detected:
-            resolved[step.id] = matched
-        elif failure_reason is None:
-            failure_reason = f"no matching event classified for step '{step.id}'"
-
-        step_results.append(StepResult(
-            step_id=step.id,
-            detected=detected,
-            evidence=tuple(matched),
-            failure_reason=failure_reason if not detected else None,
-        ))
-
-    required_ids = {s.id for s in spec.steps if s.required}
-    required_steps = [s for s in step_results if s.step_id in required_ids]
-    detected_required = sum(1 for s in required_steps if s.detected)
-    total_required = len(required_steps)
-
-    compliance_rate = detected_required / total_required if total_required > 0 else 0.0
-
-    return ComplianceResult(
-        spec_id=spec.id,
-        steps=tuple(step_results),
-        compliance_rate=compliance_rate,
-        recommend_hook_promotion=compliance_rate < spec.threshold_promote_to_hook,
-        classification=classification,
-    )

package/.cursor/skills/skill-comply/scripts/parser.py

@@ -1,107 +0,0 @@
-"""Parse observation traces (JSONL) and compliance specs (YAML)."""
-
-from __future__ import annotations
-
-import json
-from dataclasses import dataclass
-from pathlib import Path
-
-import yaml
-
-
-@dataclass(frozen=True)
-class ObservationEvent:
-    timestamp: str
-    event: str
-    tool: str
-    session: str
-    input: str
-    output: str
-
-
-@dataclass(frozen=True)
-class Detector:
-    description: str
-    after_step: str | None = None
-    before_step: str | None = None
-
-
-@dataclass(frozen=True)
-class Step:
-    id: str
-    description: str
-    required: bool
-    detector: Detector
-
-
-@dataclass(frozen=True)
-class ComplianceSpec:
-    id: str
-    name: str
-    source_rule: str
-    version: str
-    steps: tuple[Step, ...]
-    threshold_promote_to_hook: float
-
-
-def parse_trace(path: Path) -> list[ObservationEvent]:
-    """Parse a JSONL observation trace file into sorted events."""
-    if not path.is_file():
-        raise FileNotFoundError(f"Trace file not found: {path}")
-
-    text = path.read_text().strip()
-    if not text:
-        return []
-
-    events: list[ObservationEvent] = []
-    for i, line in enumerate(text.splitlines(), 1):
-        try:
-            raw = json.loads(line)
-        except json.JSONDecodeError as e:
-            raise ValueError(f"Invalid JSON at line {i}: {e}") from e
-        try:
-            events.append(ObservationEvent(
-                timestamp=raw["timestamp"],
-                event=raw["event"],
-                tool=raw["tool"],
-                session=raw["session"],
-                input=raw.get("input", ""),
-                output=raw.get("output", ""),
-            ))
-        except KeyError as e:
-            raise ValueError(f"Missing required field {e} at line {i}") from e
-
-    return sorted(events, key=lambda e: e.timestamp)
-
-
-def parse_spec(path: Path) -> ComplianceSpec:
-    """Parse a YAML compliance spec file."""
-    if not path.is_file():
-        raise FileNotFoundError(f"Spec file not found: {path}")
-    raw = yaml.safe_load(path.read_text())
-
-    steps: list[Step] = []
-    for s in raw["steps"]:
-        d = s["detector"]
-        steps.append(Step(
-            id=s["id"],
-            description=s["description"],
-            required=s["required"],
-            detector=Detector(
-                description=d["description"],
-                after_step=d.get("after_step"),
-                before_step=d.get("before_step"),
-            ),
-        ))
-
-    if "scoring" not in raw:
-        raise KeyError("Missing 'scoring' section in compliance spec")
-
-    return ComplianceSpec(
-        id=raw["id"],
-        name=raw["name"],
-        source_rule=raw["source_rule"],
-        version=raw["version"],
-        steps=tuple(steps),
-        threshold_promote_to_hook=raw["scoring"]["threshold_promote_to_hook"],
-    )

package/.cursor/skills/skill-comply/scripts/report.py

@@ -1,170 +0,0 @@
-"""Generate Markdown compliance reports."""
-
-from __future__ import annotations
-
-from datetime import datetime, timezone
-from pathlib import Path
-
-from scripts.grader import ComplianceResult
-from scripts.parser import ComplianceSpec, ObservationEvent
-from scripts.scenario_generator import Scenario
-
-
-def generate_report(
-    skill_path: Path,
-    spec: ComplianceSpec,
-    results: list[tuple[str, ComplianceResult, list[ObservationEvent]]],
-    scenarios: list[Scenario] | None = None,
-) -> str:
-    """Generate a Markdown compliance report.
-
-    Args:
-        skill_path: Path to the skill file that was tested.
-        spec: The compliance spec used for grading.
-        results: List of (scenario_level_name, ComplianceResult, observations) tuples.
-        scenarios: Original scenario definitions with prompts.
-    """
-    now = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
-    overall = _overall_compliance(results)
-    threshold = spec.threshold_promote_to_hook
-
-    lines: list[str] = []
-    lines.append(f"# skill-comply Report: {skill_path.name}")
-    lines.append(f"Generated: {now}")
-    lines.append("")
-
-    # Summary
-    lines.append("## Summary")
-    lines.append("")
-    lines.append(f"| Metric | Value |")
-    lines.append(f"|--------|-------|")
-    lines.append(f"| Skill | `{skill_path}` |")
-    lines.append(f"| Spec | {spec.id} |")
-    lines.append(f"| Scenarios | {len(results)} |")
-    lines.append(f"| Overall Compliance | {overall:.0%} |")
-    lines.append(f"| Threshold | {threshold:.0%} |")
-
-    promote_steps = _steps_to_promote(spec, results, threshold)
-    if promote_steps:
-        step_names = ", ".join(promote_steps)
-        lines.append(f"| Recommendation | **Promote {step_names} to hooks** |")
-    else:
-        lines.append(f"| Recommendation | All steps above threshold — no hook promotion needed |")
-    lines.append("")
-
-    # Expected Behavioral Sequence
-    lines.append("## Expected Behavioral Sequence")
-    lines.append("")
-    lines.append("| # | Step | Required | Description |")
-    lines.append("|---|------|----------|-------------|")
-    for i, step in enumerate(spec.steps, 1):
-        req = "Yes" if step.required else "No"
-        lines.append(f"| {i} | {step.id} | {req} | {step.detector.description} |")
-    lines.append("")
-
-    # Scenario Results
-    lines.append("## Scenario Results")
-    lines.append("")
-    lines.append("| Scenario | Compliance | Failed Steps |")
-    lines.append("|----------|-----------|----------------|")
-    for level_name, result, _obs in results:
-        failed = [s.step_id for s in result.steps if not s.detected
-                  and any(sp.id == s.step_id and sp.required for sp in spec.steps)]
-        failed_str = ", ".join(failed) if failed else "—"
-        lines.append(f"| {level_name} | {result.compliance_rate:.0%} | {failed_str} |")
-    lines.append("")
-
-    # Scenario Prompts
-    if scenarios:
-        lines.append("## Scenario Prompts")
-        lines.append("")
-        for s in scenarios:
-            lines.append(f"### {s.level_name} (Level {s.level})")
-            lines.append("")
-            for prompt_line in s.prompt.splitlines():
-                lines.append(f"> {prompt_line}")
-            lines.append("")
-
-    # Hook Promotion Recommendations (optional/advanced)
-    if promote_steps:
-        lines.append("## Advanced: Hook Promotion Recommendations (optional)")
-        lines.append("")
-        for step_id in promote_steps:
-            rate = _step_compliance_rate(step_id, results)
-            step = next(s for s in spec.steps if s.id == step_id)
-            lines.append(
-                f"- **{step_id}** (compliance {rate:.0%}): {step.description}"
-            )
-        lines.append("")
-
-    # Per-scenario details with timeline
-    lines.append("## Detail")
-    lines.append("")
-    for level_name, result, observations in results:
-        lines.append(f"### {level_name} (Compliance: {result.compliance_rate:.0%})")
-        lines.append("")
-        lines.append("| Step | Required | Detected | Reason |")
-        lines.append("|------|----------|----------|--------|")
-        for sr in result.steps:
-            req = "Yes" if any(
-                sp.id == sr.step_id and sp.required for sp in spec.steps
-            ) else "No"
-            det = "YES" if sr.detected else "NO"
-            reason = sr.failure_reason or "—"
-            lines.append(f"| {sr.step_id} | {req} | {det} | {reason} |")
-        lines.append("")
-
-        # Timeline: show what the agent actually did
-        if observations:
-            # Build reverse index: event_index → step_id
-            index_to_step: dict[int, str] = {}
-            for step_id, indices in result.classification.items():
-                for idx in indices:
-                    index_to_step[idx] = step_id
-
-            lines.append(f"**Tool Call Timeline ({len(observations)} calls)**")
-            lines.append("")
-            lines.append("| # | Tool | Input | Output | Classified As |")
-            lines.append("|---|------|-------|--------|------|")
-            for i, obs in enumerate(observations):
-                step_label = index_to_step.get(i, "—")
-                input_summary = obs.input[:100].replace("|", "\\|").replace("\n", " ")
-                output_summary = obs.output[:50].replace("|", "\\|").replace("\n", " ")
-                lines.append(
-                    f"| {i} | {obs.tool} | {input_summary} | {output_summary} | {step_label} |"
-                )
-            lines.append("")
-
-    return "\n".join(lines)
-
-
-def _overall_compliance(results: list[tuple[str, ComplianceResult, list[ObservationEvent]]]) -> float:
-    if not results:
-        return 0.0
-    return sum(r.compliance_rate for _, r, _obs in results) / len(results)
-
-
-def _step_compliance_rate(
-    step_id: str,
-    results: list[tuple[str, ComplianceResult, list[ObservationEvent]]],
-) -> float:
-    detected = sum(
-        1 for _, r, _obs in results
-        for s in r.steps if s.step_id == step_id and s.detected
-    )
-    return detected / len(results) if results else 0.0
-
-
-def _steps_to_promote(
-    spec: ComplianceSpec,
-    results: list[tuple[str, ComplianceResult, list[ObservationEvent]]],
-    threshold: float,
-) -> list[str]:
-    promote = []
-    for step in spec.steps:
-        if not step.required:
-            continue
-        rate = _step_compliance_rate(step.id, results)
-        if rate < threshold:
-            promote.append(step.id)
-    return promote

package/.cursor/skills/skill-comply/scripts/run.py

@@ -1,127 +0,0 @@
-"""CLI entry point for skill-comply."""
-
-from __future__ import annotations
-
-import argparse
-import logging
-import sys
-from pathlib import Path
-from typing import Any
-
-import yaml
-
-from scripts.grader import grade
-from scripts.report import generate_report
-from scripts.runner import run_scenario
-from scripts.scenario_generator import generate_scenarios
-from scripts.spec_generator import generate_spec
-
-logger = logging.getLogger(__name__)
-
-
-def main() -> None:
-    logging.basicConfig(level=logging.INFO, format="%(message)s")
-
-    parser = argparse.ArgumentParser(
-        description="skill-comply: Measure skill compliance rates",
-    )
-    parser.add_argument(
-        "skill",
-        type=Path,
-        help="Path to skill/rule file to test",
-    )
-    parser.add_argument(
-        "--model",
-        default="sonnet",
-        help="Model for scenario execution (default: sonnet)",
-    )
-    parser.add_argument(
-        "--gen-model",
-        default="haiku",
-        help="Model for spec/scenario generation (default: haiku)",
-    )
-    parser.add_argument(
-        "--dry-run",
-        action="store_true",
-        help="Generate spec and scenarios without executing",
-    )
-    parser.add_argument(
-        "--output",
-        type=Path,
-        default=None,
-        help="Output report path (default: results/<skill-name>.md)",
-    )
-
-    args = parser.parse_args()
-
-    if not args.skill.is_file():
-        logger.error("Error: Skill file not found: %s", args.skill)
-        sys.exit(1)
-
-    results_dir = Path(__file__).parent.parent / "results"
-    results_dir.mkdir(exist_ok=True)
-
-    # Step 1: Generate compliance spec
-    logger.info("[1/4] Generating compliance spec from %s...", args.skill.name)
-    spec = generate_spec(args.skill, model=args.gen_model)
-    logger.info("       %d steps extracted", len(spec.steps))
-
-    # Step 2: Generate scenarios
-    spec_yaml = yaml.dump({
-        "steps": [
-            {"id": s.id, "description": s.description, "required": s.required}
-            for s in spec.steps
-        ]
-    })
-    logger.info("[2/4] Generating scenarios (3 prompt strictness levels)...")
-    scenarios = generate_scenarios(args.skill, spec_yaml, model=args.gen_model)
-    logger.info("       %d scenarios generated", len(scenarios))
-
-    for s in scenarios:
-        logger.info("       - %s: %s", s.level_name, s.description[:60])
-
-    if args.dry_run:
-        logger.info("\n[dry-run] Spec and scenarios generated. Skipping execution.")
-        logger.info("\nSpec: %s (%d steps)", spec.id, len(spec.steps))
-        for step in spec.steps:
-            marker = "*" if step.required else " "
-            logger.info("  [%s] %s: %s", marker, step.id, step.description)
-        return
-
-    # Step 3: Execute scenarios
-    logger.info("[3/4] Executing scenarios (model=%s)...", args.model)
-    graded_results: list[tuple[str, Any, list[Any]]] = []
-
-    for scenario in scenarios:
-        logger.info("       Running %s...", scenario.level_name)
-        run = run_scenario(scenario, model=args.model)
-        result = grade(spec, list(run.observations))
-        graded_results.append((scenario.level_name, result, list(run.observations)))
-        logger.info("       %s: %.0f%%", scenario.level_name, result.compliance_rate * 100)
-
-    # Step 4: Generate report
-    skill_name = args.skill.parent.name if args.skill.stem == "SKILL" else args.skill.stem
-    output_path = args.output or results_dir / f"{skill_name}.md"
-    logger.info("[4/4] Generating report...")
-
-    report = generate_report(args.skill, spec, graded_results, scenarios=scenarios)
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    output_path.write_text(report)
-    logger.info("       Report saved to %s", output_path)
-
-    # Summary
-    if not graded_results:
-        logger.warning("No scenarios were executed.")
-        return
-    overall = sum(r.compliance_rate for _, r, _obs in graded_results) / len(graded_results)
-    logger.info("\n%s", "=" * 50)
-    logger.info("Overall Compliance: %.0f%%", overall * 100)
-    if overall < spec.threshold_promote_to_hook:
-        logger.info(
-            "Recommendation: Some steps have low compliance. "
-            "Consider promoting them to hooks. See the report for details."
-        )
-
-
-if __name__ == "__main__":
-    main()