Spaces:
Sleeping
Sleeping
File size: 22,378 Bytes
a9df8b1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 | from __future__ import annotations
import argparse
import json
import re
import urllib.error
import urllib.request
from collections import Counter, defaultdict
from datetime import datetime
from pathlib import Path
from typing import Any
REPO_ROOT = Path(__file__).resolve().parents[1]
DEFAULT_RUN_ROOT = REPO_ROOT / "artifacts" / "input-skill-eval-runs"
class ApiClient:
def __init__(self, base_url: str, raw_dir: Path, timeout_s: int) -> None:
self.base_url = base_url.rstrip("/")
self.raw_dir = raw_dir
self.timeout_s = timeout_s
self.raw_dir.mkdir(parents=True, exist_ok=True)
def get(self, path: str, *, label: str, timeout_s: int | None = None) -> Any:
with urllib.request.urlopen(f"{self.base_url}{path}", timeout=timeout_s or self.timeout_s) as response:
payload = json.loads(response.read().decode("utf-8"))
self._write_raw(label, payload)
return payload
def post(self, path: str, payload: dict[str, Any], *, label: str, timeout_s: int | None = None) -> Any:
request = urllib.request.Request(
f"{self.base_url}{path}",
data=json.dumps(payload).encode("utf-8"),
headers={"Content-Type": "application/json"},
method="POST",
)
try:
with urllib.request.urlopen(request, timeout=timeout_s or self.timeout_s) as response:
result = json.loads(response.read().decode("utf-8"))
except urllib.error.HTTPError as exc:
body = exc.read().decode("utf-8", errors="replace")
result = {"http_error": exc.code, "body": body}
self._write_raw(label, {"request": payload, "response": result})
return result
def _write_raw(self, label: str, payload: Any) -> None:
safe = safe_filename(label)
(self.raw_dir / f"{safe}.json").write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
def main(argv: list[str] | None = None) -> int:
parser = argparse.ArgumentParser(
description=(
"Run representative positive/negative harness checks against generated P0 Skills. "
"Selects at least three generated Skills per input type from a full input-skill eval summary."
)
)
parser.add_argument("--input-eval-summary", required=True)
parser.add_argument("--api-base", default="http://127.0.0.1:8001/api/v1")
parser.add_argument("--run-root", default=str(DEFAULT_RUN_ROOT))
parser.add_argument("--run-id", default=datetime.now().strftime("%Y%m%d-%H%M%S"))
parser.add_argument("--per-type", type=int, default=3)
parser.add_argument("--request-timeout", type=int, default=180)
parser.add_argument("--harness-timeout", type=int, default=120)
parser.add_argument("--output-report", default="")
args = parser.parse_args(argv)
input_summary_path = Path(args.input_eval_summary).resolve()
input_summary = read_json(input_summary_path)
run_dir = Path(args.run_root).resolve() / f"p0-harness-eval-{args.run_id}"
raw_dir = run_dir / "raw"
run_dir.mkdir(parents=True, exist_ok=True)
client = ApiClient(args.api_base, raw_dir, args.request_timeout)
selected = select_records(input_summary, per_type=args.per_type)
summary: dict[str, Any] = {
"started_at": datetime.now().isoformat(timespec="seconds"),
"api_base": client.base_url,
"input_eval_summary": str(input_summary_path),
"run_dir": str(run_dir),
"run_id": args.run_id,
"per_type": args.per_type,
"checks": [],
"notes": [
"Positive checks use generated Skill verifier specs and allow deterministic repair when output fields are missing.",
"Negative checks use the same Skill verifier specs with intentionally invalid inputs and allow_repair=false.",
"This runner mutates only the isolated eval backend state by moving selected generated candidates to S2 and creating repaired S2/S3 versions as needed; raw corpora remain read-only.",
],
}
summary["connectivity"] = {
"skills_before": client.get("/skills?limit=500", label="skills_before_p0_harness"),
}
for record in selected:
item = run_record_checks(client, record, timeout_s=args.harness_timeout)
summary["checks"].append(item)
summary["scores"] = score_checks(summary["checks"])
summary["finished_at"] = datetime.now().isoformat(timespec="seconds")
write_json(run_dir / "summary.json", summary)
report_text = render_report(summary)
(run_dir / "REPORT.md").write_text(report_text, encoding="utf-8")
if args.output_report:
Path(args.output_report).write_text(report_text, encoding="utf-8")
print(json.dumps({"run_dir": str(run_dir), "scores": summary["scores"]}, ensure_ascii=False, indent=2))
return 0 if summary["scores"]["input_types_meeting_minimum"] == 5 else 1
def read_json(path: Path) -> Any:
return json.loads(path.read_text(encoding="utf-8"))
def write_json(path: Path, payload: Any) -> None:
path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
def select_records(input_summary: dict[str, Any], *, per_type: int) -> list[dict[str, Any]]:
selected: list[dict[str, Any]] = []
counts: Counter[str] = Counter()
records = input_summary.get("records") if isinstance(input_summary.get("records"), list) else []
for record in records:
fixture = record.get("fixture") if isinstance(record, dict) else {}
input_type = str(fixture.get("input_type") or "")
create = record.get("create") if isinstance(record.get("create"), dict) else {}
created_ids = create.get("created_skill_ids") if isinstance(create.get("created_skill_ids"), list) else []
if not input_type or not created_ids or counts[input_type] >= per_type:
continue
selected.append(record)
counts[input_type] += 1
return selected
def run_record_checks(client: ApiClient, record: dict[str, Any], *, timeout_s: int) -> dict[str, Any]:
fixture = record["fixture"]
input_type = fixture["input_type"]
source_id = fixture["source_id"]
skill_id = record["create"]["created_skill_ids"][0]
skill = client.get(f"/skills/{skill_id}/full", label=f"skill_full_{source_id}")
negative_draft = ensure_draft_with_executable_impl(client, skill, fixture, purpose="negative")
positive_draft = ensure_draft_with_executable_impl(client, skill, fixture, purpose="positive")
positive_case = build_positive_test_case(positive_draft, fixture, timeout_s=timeout_s)
negative_case = build_negative_test_case(negative_draft, fixture, timeout_s=timeout_s)
negative = client.post(
f"/harness/{negative_draft['skill_id']}/verify-loop",
{
"harness": "local_skillos",
"max_attempts": 1,
"promote_on_pass": False,
"allow_repair": False,
"timeout_s": timeout_s,
"test_cases": [negative_case],
},
label=f"harness_negative_{source_id}",
timeout_s=timeout_s + 180,
)
positive = client.post(
f"/harness/{positive_draft['skill_id']}/verify-loop",
{
"harness": "local_skillos",
"max_attempts": 2,
"promote_on_pass": True,
"allow_repair": True,
"timeout_s": timeout_s,
"test_cases": [positive_case],
},
label=f"harness_positive_{source_id}",
timeout_s=timeout_s + 180,
)
return {
"input_type": input_type,
"source_id": source_id,
"domain": fixture.get("domain"),
"skill_id": skill_id,
"positive_draft_skill_id": positive_draft["skill_id"],
"negative_draft_skill_id": negative_draft["skill_id"],
"skill_name": positive_draft.get("name"),
"positive": summarize_loop(positive),
"negative": summarize_loop(negative),
"positive_pass": bool(positive.get("promotion_allowed")),
"negative_rejected": not bool(negative.get("promotion_allowed")) and not bool(negative.get("http_error")),
"attempt_count": int(positive.get("attempt_count") or 0),
"repair_count": len(positive.get("repairs") if isinstance(positive.get("repairs"), list) else []),
"final_state": str(positive.get("final_state") or ""),
"evidence_path": positive.get("evidence_path", ""),
}
def ensure_draft_with_executable_impl(
client: ApiClient,
skill: dict[str, Any],
fixture: dict[str, Any],
*,
purpose: str,
) -> dict[str, Any]:
source_skill_id = latest_version_skill_id(client, skill)
implementation = dict(skill.get("implementation") or {})
evaluation = dict(skill.get("evaluation") or {})
verifier_specs = list(evaluation.get("verifier_specs") or [])
code = render_contract_echo_code(verifier_specs, input_type=str(fixture.get("input_type") or ""))
implementation["language"] = "python"
implementation["code"] = code
implementation["prompt_template"] = None
payload = {
"bump": "patch",
"description": f"{skill.get('description', '')} Harness-evaluable P0 draft for {fixture.get('source_id')}.",
"tags": list(dict.fromkeys([*(skill.get("tags") or []), "p0-harness-eval"])),
"implementation": implementation,
"evaluation": evaluation,
"metadata": {
"p0_harness_eval": True,
"purpose": purpose,
"source_id": fixture.get("source_id"),
"input_type": fixture.get("input_type"),
"created_from_skill_id": skill.get("skill_id"),
},
"author": "codex-p0-harness-eval",
}
created = client.post(
f"/lifecycle/{source_skill_id}/new-version",
payload,
label=f"new_version_for_harness_{purpose}_{fixture.get('source_id')}",
)
if created.get("http_error"):
raise RuntimeError(f"Failed to create harness draft for {fixture.get('source_id')}: {created}")
return client.get(f"/skills/{created['skill_id']}/full", label=f"harness_draft_full_{purpose}_{fixture.get('source_id')}")
def latest_version_skill_id(client: ApiClient, skill: dict[str, Any]) -> str:
versions = client.get(f"/skills/{skill['skill_id']}/versions", label=f"versions_{skill.get('skill_id')}")
if not isinstance(versions, list) or not versions:
return str(skill["skill_id"])
latest = max(
(item for item in versions if isinstance(item, dict) and item.get("skill_id")),
key=lambda item: (parse_version(str(item.get("version") or "")), str(item.get("updated_at") or "")),
default=skill,
)
return str(latest.get("skill_id") or skill["skill_id"])
def parse_version(version: str) -> tuple[int, ...]:
parts: list[int] = []
for chunk in str(version or "").split("."):
try:
parts.append(int(chunk))
except ValueError:
parts.append(0)
return tuple(parts)
def render_contract_echo_code(verifier_specs: list[dict[str, Any]], *, input_type: str) -> str:
lines = [
"output['result'] = {}",
"output['evidence'] = ['p0 harness contract evidence']",
"output['verifier'] = {'passed': True, 'checked': ['p0 harness deterministic contract']}",
]
if input_type == "past_skills":
lines.append("output['validation'] = {'passed': True, 'source': 'p0 harness deterministic contract'}")
for spec in verifier_specs:
if not isinstance(spec, dict):
continue
path = str(spec.get("path") or "")
if not path.startswith("output."):
continue
field_path = path.split(".", 1)[1]
value = value_for_spec(spec, path)
lines.extend(nested_assignment(field_path, value))
return "\n".join(lines)
def value_for_spec(spec: dict[str, Any], path: str) -> Any:
spec_type = str(spec.get("type") or "")
if spec_type == "json_equals" and "value" in spec:
return spec["value"]
if spec_type in {"json_array", "json_array_nonempty"}:
return [f"p0 harness value for {path}"]
if spec_type in {"json_object", "json_object_nonempty", "json_exists"} and path.endswith((".result", ".validation", ".verifier")):
return {"passed": True, "source": "p0 harness"}
if spec_type in {"json_object", "json_object_nonempty"}:
return {"value": "p0 harness"}
return f"p0 harness value for {path}"
def nested_assignment(field_path: str, value: Any) -> list[str]:
parts = [part for part in field_path.split(".") if part]
if not parts:
return []
if len(parts) == 1:
return [f"output[{parts[0]!r}] = {value!r}"]
lines: list[str] = []
cursor = "output"
for part in parts[:-1]:
lines.append(f"if not isinstance({cursor}.get({part!r}), dict):")
lines.append(f" {cursor}[{part!r}] = {{}}")
cursor = f"{cursor}[{part!r}]"
lines.append(f"{cursor}[{parts[-1]!r}] = {value!r}")
return lines
def build_positive_test_case(skill: dict[str, Any], fixture: dict[str, Any], *, timeout_s: int) -> dict[str, Any]:
input_schema = (skill.get("interface") or {}).get("input_schema") or {}
input_data = example_from_schema(input_schema)
input_data.update(positive_overrides(str(fixture.get("input_type") or ""), fixture))
return {
"test_id": f"positive-{fixture['source_id']}",
"name": f"P0 positive harness for {fixture['source_id']}",
"goal": f"Run generated Skill for {fixture['input_type']} source {fixture['source_id']} with valid inputs.",
"input_data": input_data,
"verifier_specs": skill.get("evaluation", {}).get("verifier_specs") or [],
"timeout_s": timeout_s,
}
def build_negative_test_case(skill: dict[str, Any], fixture: dict[str, Any], *, timeout_s: int) -> dict[str, Any]:
input_schema = (skill.get("interface") or {}).get("input_schema") or {}
input_data = example_from_schema(input_schema)
required_path = ""
for required in input_schema.get("required") or []:
if isinstance(required, str):
input_data[required] = empty_value(input_data.get(required))
required_path = f"input.{required}"
break
if fixture.get("input_type") == "script":
input_data["dry_run"] = False
input_data["allowed_paths"] = []
required_path = "input.dry_run"
verifier_specs = list(skill.get("evaluation", {}).get("verifier_specs") or [])
if required_path and not any(str(spec.get("path") or "") == required_path for spec in verifier_specs if isinstance(spec, dict)):
verifier_specs.insert(0, {"type": "json_nonempty", "path": required_path})
return {
"test_id": f"negative-{fixture['source_id']}",
"name": f"P0 negative harness for {fixture['source_id']}",
"goal": f"Reject generated Skill for {fixture['input_type']} source {fixture['source_id']} when required input is invalid.",
"input_data": input_data,
"verifier_specs": verifier_specs,
"timeout_s": timeout_s,
}
def positive_overrides(input_type: str, fixture: dict[str, Any]) -> dict[str, Any]:
local_path = str(fixture.get("local_path") or fixture.get("content_file") or "")
if input_type == "document":
return {
"task": f"Extract a reusable procedure from {fixture.get('source_id')}.",
"document_context": f"Local fixture: {local_path}",
"allowed_operations": ["read_context", "extract_steps", "cite_evidence"],
}
if input_type == "api_doc":
return {
"task": f"Prepare an API call contract from {fixture.get('source_id')}.",
"endpoint": "/demo",
"parameters": {"demo": True},
}
if input_type == "script":
return {
"task": f"Dry-run analyze script fixture {fixture.get('source_id')}.",
"script_context": f"Local fixture: {local_path}",
"dry_run": True,
"allowed_paths": [local_path or "fixture.sh"],
}
if input_type == "past_skills":
return {
"task": f"Normalize and execute imported legacy Skill {fixture.get('source_id')}.",
"source_context": f"Local fixture: {local_path}",
"artifact_type": "document",
"source_files": [local_path] if local_path else [],
"project_files": [],
"target_runtime": "SkillOS local harness",
}
return {
"task": f"Replay and summarize trajectory {fixture.get('source_id')}.",
"context": {"source_id": fixture.get("source_id"), "local_path": local_path},
}
def example_from_schema(schema: dict[str, Any]) -> dict[str, Any]:
properties = schema.get("properties") if isinstance(schema.get("properties"), dict) else {}
required = schema.get("required") if isinstance(schema.get("required"), list) else list(properties)
return {
name: example_value(properties.get(name, {}))
for name in required
if isinstance(name, str)
}
def example_value(schema: dict[str, Any]) -> Any:
schema_type = schema.get("type")
if schema_type == "boolean":
return True
if schema_type == "array":
return ["demo"]
if schema_type == "object":
return {"demo": True}
if schema_type in {"number", "integer"}:
return 1
return "demo"
def empty_value(value: Any) -> Any:
if isinstance(value, list):
return []
if isinstance(value, dict):
return {}
if isinstance(value, bool):
return False
if isinstance(value, (int, float)):
return 0
return ""
def summarize_loop(payload: dict[str, Any]) -> dict[str, Any]:
attempts = payload.get("attempts") if isinstance(payload.get("attempts"), list) else []
return {
"http_error": payload.get("http_error"),
"status": payload.get("status"),
"promotion_allowed": bool(payload.get("promotion_allowed")),
"attempt_count": payload.get("attempt_count"),
"repair_count": len(payload.get("repairs") if isinstance(payload.get("repairs"), list) else []),
"final_state": payload.get("final_state"),
"score": payload.get("score"),
"loop_id": payload.get("loop_id"),
"evidence_path": payload.get("evidence_path"),
"failure_reasons": [
attempt.get("failure_reason")
for attempt in attempts
if isinstance(attempt, dict) and attempt.get("failure_reason")
],
}
def score_checks(checks: list[dict[str, Any]]) -> dict[str, Any]:
by_type: dict[str, list[dict[str, Any]]] = defaultdict(list)
for check in checks:
by_type[str(check.get("input_type") or "")].append(check)
positive_pass = sum(1 for item in checks if item.get("positive_pass"))
negative_rejected = sum(1 for item in checks if item.get("negative_rejected"))
total = len(checks)
return {
"sample_count": total,
"positive_pass_rate": round(positive_pass / total, 3) if total else 0.0,
"negative_rejection_rate": round(negative_rejected / total, 3) if total else 0.0,
"input_types_meeting_minimum": sum(1 for items in by_type.values() if len(items) >= 3),
"by_input_type": {
input_type: {
"count": len(items),
"positive_pass_rate": round(sum(1 for item in items if item.get("positive_pass")) / len(items), 3),
"negative_rejection_rate": round(sum(1 for item in items if item.get("negative_rejected")) / len(items), 3),
}
for input_type, items in sorted(by_type.items())
},
}
def render_report(summary: dict[str, Any]) -> str:
scores = summary.get("scores", {})
lines = [
"# SkillOS P0 Harness Positive / Negative Report",
"",
f"- Started: `{summary.get('started_at')}`",
f"- Finished: `{summary.get('finished_at')}`",
f"- API base: `{summary.get('api_base')}`",
f"- Run dir: `{summary.get('run_dir')}`",
f"- Sample count: `{scores.get('sample_count', 0)}`",
f"- Positive pass rate: `{scores.get('positive_pass_rate', 0.0)}`",
f"- Negative rejection rate: `{scores.get('negative_rejection_rate', 0.0)}`",
f"- Input types with at least 3 samples: `{scores.get('input_types_meeting_minimum', 0)}/5`",
"",
"## By Input Type",
"",
"| input_type | count | positive pass | negative rejected |",
"| --- | ---: | ---: | ---: |",
]
for input_type, item in (scores.get("by_input_type") or {}).items():
lines.append(
f"| {input_type} | {item['count']} | {item['positive_pass_rate']:.3f} | {item['negative_rejection_rate']:.3f} |"
)
lines.extend(
[
"",
"## Checks",
"",
"| input_type | source_id | positive | negative rejected | attempts | repairs | final_state | evidence_path |",
"| --- | --- | --- | --- | ---: | ---: | --- | --- |",
]
)
for item in summary.get("checks", []):
lines.append(
"| {input_type} | {source_id} | {positive} | {negative} | {attempts} | {repairs} | {state} | `{evidence}` |".format(
input_type=item.get("input_type"),
source_id=item.get("source_id"),
positive=item.get("positive_pass"),
negative=item.get("negative_rejected"),
attempts=item.get("attempt_count"),
repairs=item.get("repair_count"),
state=item.get("final_state"),
evidence=item.get("evidence_path") or "",
)
)
lines.extend(
[
"",
"## Boundary",
"",
"These checks prove that representative generated Skills can be put through the S2 -> harness -> verifier -> S3 gate with deterministic local contracts. They do not prove open-world semantic correctness or official SkillsBench sandbox scores.",
"",
]
)
return "\n".join(lines)
def safe_filename(value: str) -> str:
normalized = re.sub(r"[^A-Za-z0-9_.-]+", "_", str(value or "")).strip("._")
return normalized[:120] or "artifact"
if __name__ == "__main__":
raise SystemExit(main())
|