ops: make DAARWIZZ awareness canary static by default with optional runtime mode
This commit is contained in:
Apple
@@ -2,6 +2,7 @@
|
||||
set -euo pipefail
|
||||
|
||||
ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
|
||||
MODE="${DAARWIZZ_AWARENESS_MODE:-static}" # static | runtime
|
||||
ROUTER_URL="${ROUTER_URL:-http://127.0.0.1:9102}"
|
||||
REQUEST_TIMEOUT_SEC="${DAARWIZZ_AWARENESS_TIMEOUT_SEC:-35}"
|
||||
OUT_JSON="${DAARWIZZ_AWARENESS_OUT_JSON:-/tmp/daarwizz_awareness_audit.json}"
|
||||
@@ -11,10 +12,9 @@ if [[ -x "$ROOT/.venv-codex/bin/python" ]]; then
|
||||
PYTHON_BIN="$ROOT/.venv-codex/bin/python"
|
||||
fi
|
||||
|
||||
echo "[INFO] checking DAARWIZZ awareness for top-level agents"
|
||||
echo "[INFO] router=$ROUTER_URL timeout=${REQUEST_TIMEOUT_SEC}s"
|
||||
echo "[INFO] checking DAARWIZZ awareness (mode=$MODE)"
|
||||
|
||||
"$PYTHON_BIN" - "$ROOT" "$ROUTER_URL" "$REQUEST_TIMEOUT_SEC" "$OUT_JSON" <<'PY'
|
||||
"$PYTHON_BIN" - "$ROOT" "$MODE" "$ROUTER_URL" "$REQUEST_TIMEOUT_SEC" "$OUT_JSON" <<'PY'
|
||||
import json
|
||||
import re
|
||||
import sys
|
||||
@@ -25,86 +25,125 @@ from urllib import request as urlreq
|
||||
import yaml
|
||||
|
||||
root = Path(sys.argv[1])
|
||||
router = sys.argv[2].rstrip("/")
|
||||
timeout_s = int(sys.argv[3])
|
||||
out_json = Path(sys.argv[4])
|
||||
mode = sys.argv[2]
|
||||
router = sys.argv[3].rstrip("/")
|
||||
timeout_s = int(sys.argv[4])
|
||||
out_json = Path(sys.argv[5])
|
||||
|
||||
registry = yaml.safe_load((root / "config" / "agent_registry.yml").read_text(encoding="utf-8"))
|
||||
agents = [a for a in registry.get("agents", []) if a.get("class") == "top_level"]
|
||||
|
||||
prompt = (
|
||||
"Who is DAARWIZZ in DAARION network? "
|
||||
"Answer briefly in 1 sentence and mention that DAARWIZZ is the main orchestrator/mayor."
|
||||
)
|
||||
results = []
|
||||
|
||||
deny_patterns = [
|
||||
r"не знаю\s+daarwizz",
|
||||
r"don't know\s+daarwizz",
|
||||
r"do not know\s+daarwizz",
|
||||
r"no information about daarwizz",
|
||||
r"не\s+знаю\s+хто\s+це",
|
||||
]
|
||||
pos_patterns = [
|
||||
r"оркестратор",
|
||||
r"orchestrator",
|
||||
r"мер",
|
||||
r"mayor",
|
||||
r"entry point",
|
||||
r"вхідн",
|
||||
]
|
||||
if mode == "static":
|
||||
for idx, agent in enumerate(agents, 1):
|
||||
aid = agent["id"]
|
||||
prompt_file = agent.get("prompt_file", "")
|
||||
row = {"agent": aid, "status": "FAIL", "reason": "", "preview": ""}
|
||||
|
||||
rows = []
|
||||
for idx, agent in enumerate(agents, 1):
|
||||
agent_id = agent["id"]
|
||||
req_body = {"prompt": prompt, "max_tokens": 100, "temperature": 0.1}
|
||||
req = urlreq.Request(
|
||||
url=f"{router}/v1/agents/{agent_id}/infer",
|
||||
data=json.dumps(req_body).encode("utf-8"),
|
||||
headers={"Content-Type": "application/json"},
|
||||
method="POST",
|
||||
# DAARWIZZ itself can be defined without self-reference check.
|
||||
if aid == "daarwizz":
|
||||
row["status"] = "PASS"
|
||||
row["reason"] = "self_agent"
|
||||
results.append(row)
|
||||
print(f"[{idx:02d}/{len(agents)}] {aid}: PASS (self_agent)", flush=True)
|
||||
continue
|
||||
|
||||
if not prompt_file:
|
||||
row["reason"] = "missing_prompt_file"
|
||||
results.append(row)
|
||||
print(f"[{idx:02d}/{len(agents)}] {aid}: FAIL (missing_prompt_file)", flush=True)
|
||||
continue
|
||||
|
||||
prompt_path = root / "gateway-bot" / prompt_file
|
||||
if not prompt_path.exists():
|
||||
row["reason"] = f"prompt_not_found:{prompt_file}"
|
||||
results.append(row)
|
||||
print(f"[{idx:02d}/{len(agents)}] {aid}: FAIL ({row['reason']})", flush=True)
|
||||
continue
|
||||
|
||||
text = prompt_path.read_text(encoding="utf-8", errors="ignore")
|
||||
low = text.lower()
|
||||
row["preview"] = text[:180].replace("\n", " ")
|
||||
|
||||
if "daarwizz" in low:
|
||||
row["status"] = "PASS"
|
||||
row["reason"] = "prompt_mentions_daarwizz"
|
||||
else:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = "prompt_missing_daarwizz_reference"
|
||||
|
||||
results.append(row)
|
||||
print(f"[{idx:02d}/{len(agents)}] {aid}: {row['status']} ({row['reason']})", flush=True)
|
||||
|
||||
elif mode == "runtime":
|
||||
prompt = (
|
||||
"Who is DAARWIZZ in DAARION network? "
|
||||
"Answer briefly in 1 sentence and mention that DAARWIZZ is the main orchestrator/mayor."
|
||||
)
|
||||
deny_patterns = [
|
||||
r"не знаю\s+daarwizz",
|
||||
r"don't know\s+daarwizz",
|
||||
r"do not know\s+daarwizz",
|
||||
r"no information about daarwizz",
|
||||
r"не\s+знаю\s+хто\s+це",
|
||||
]
|
||||
pos_patterns = [r"оркестратор", r"orchestrator", r"мер", r"mayor", r"entry point", r"вхідн"]
|
||||
|
||||
row = {"agent": agent_id, "status": "FAIL", "reason": "", "preview": ""}
|
||||
try:
|
||||
with urlreq.urlopen(req, timeout=timeout_s) as resp:
|
||||
data = json.loads(resp.read().decode("utf-8", errors="ignore"))
|
||||
text = (data.get("response") or "").strip()
|
||||
low = text.lower()
|
||||
row["preview"] = text[:220].replace("\n", " ")
|
||||
for idx, agent in enumerate(agents, 1):
|
||||
aid = agent["id"]
|
||||
req_body = {"prompt": prompt, "max_tokens": 100, "temperature": 0.1}
|
||||
req = urlreq.Request(
|
||||
url=f"{router}/v1/agents/{aid}/infer",
|
||||
data=json.dumps(req_body).encode("utf-8"),
|
||||
headers={"Content-Type": "application/json"},
|
||||
method="POST",
|
||||
)
|
||||
|
||||
deny = any(re.search(p, low) for p in deny_patterns)
|
||||
pos = any(re.search(p, low) for p in pos_patterns)
|
||||
if deny:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = "explicit_deny"
|
||||
elif pos:
|
||||
row["status"] = "PASS"
|
||||
row["reason"] = "role_context_found"
|
||||
else:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = "no_role_context"
|
||||
except Exception as e:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = f"http_error:{type(e).__name__}"
|
||||
row = {"agent": aid, "status": "FAIL", "reason": "", "preview": ""}
|
||||
try:
|
||||
with urlreq.urlopen(req, timeout=timeout_s) as resp:
|
||||
data = json.loads(resp.read().decode("utf-8", errors="ignore"))
|
||||
text = (data.get("response") or "").strip()
|
||||
low = text.lower()
|
||||
row["preview"] = text[:220].replace("\n", " ")
|
||||
|
||||
rows.append(row)
|
||||
print(f"[{idx:02d}/{len(agents)}] {agent_id}: {row['status']} ({row['reason']})")
|
||||
time.sleep(0.05)
|
||||
deny = any(re.search(p, low) for p in deny_patterns)
|
||||
pos = any(re.search(p, low) for p in pos_patterns)
|
||||
if deny:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = "explicit_deny"
|
||||
elif pos:
|
||||
row["status"] = "PASS"
|
||||
row["reason"] = "role_context_found"
|
||||
else:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = "no_role_context"
|
||||
except Exception as e:
|
||||
row["status"] = "FAIL"
|
||||
row["reason"] = f"http_error:{type(e).__name__}"
|
||||
|
||||
results.append(row)
|
||||
print(f"[{idx:02d}/{len(agents)}] {aid}: {row['status']} ({row['reason']})", flush=True)
|
||||
time.sleep(0.05)
|
||||
else:
|
||||
raise SystemExit(f"[FAIL] unsupported mode: {mode}")
|
||||
|
||||
summary = {
|
||||
"total": len(rows),
|
||||
"pass": sum(1 for r in rows if r["status"] == "PASS"),
|
||||
"fail": sum(1 for r in rows if r["status"] == "FAIL"),
|
||||
"mode": mode,
|
||||
"total": len(results),
|
||||
"pass": sum(1 for r in results if r["status"] == "PASS"),
|
||||
"fail": sum(1 for r in results if r["status"] == "FAIL"),
|
||||
}
|
||||
|
||||
payload = {"summary": summary, "results": rows}
|
||||
payload = {"summary": summary, "results": results}
|
||||
out_json.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
|
||||
print(json.dumps(summary, ensure_ascii=False))
|
||||
print(f"[INFO] details: {out_json}")
|
||||
|
||||
print(json.dumps(summary, ensure_ascii=False), flush=True)
|
||||
print(f"[INFO] details: {out_json}", flush=True)
|
||||
|
||||
if summary["fail"] > 0:
|
||||
failed = [r["agent"] for r in rows if r["status"] == "FAIL"]
|
||||
failed = [r["agent"] for r in results if r["status"] == "FAIL"]
|
||||
raise SystemExit(f"[FAIL] DAARWIZZ awareness failed for: {', '.join(failed)}")
|
||||
|
||||
print("[OK] DAARWIZZ awareness canary passed")
|
||||
print("[OK] DAARWIZZ awareness canary passed", flush=True)
|
||||
PY
|
||||
|
||||
Reference in New Issue
Block a user