Self-contain CLIProxyAPI install + ieee-fraud schema cleanup

Cliproxyapi:
- install.sh fetches binary from upstream releases (vN linux-amd64)
- _vendor/LICENSE is the only committed file (binary + config.yaml +
proxy.log are .gitignored — install.sh produces them)
- The Python shim already had config.example.yaml at the parent level

ieee-fraud-detection manifest:
- files now lists train/val/test_features.csv (with identity merged in)
+ sample_submission, instead of raw transaction/identity tables
- description spells out the temporal val split (last 20% by TransactionDT)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.gitignore

@@ -31,3 +31,8 @@ runs/
 agents/**/runs/
 agents/**/_vendor/
 .claude/
+
+# CLIProxyAPI vendored
+agents/cliproxyapi/_vendor/cli-proxy-api
+agents/cliproxyapi/_vendor/config.yaml
+agents/cliproxyapi/_vendor/proxy.log

agents/cliproxyapi/install.sh

@@ -0,0 +1,51 @@
+#!/usr/bin/env bash
+# Install CLIProxyAPI binary + scaffold config under agents/cliproxyapi/_vendor/.
+# Re-run any time to upgrade. Linux x86_64 only here; adjust BINARY_NAME for
+# other platforms (see github.com/router-for-me/CLIProxyAPI/releases).
+#
+# Override the release with: CLIPROXYAPI_VERSION=v0.5.2 bash install.sh
+set -euo pipefail
+
+VERSION="${CLIPROXYAPI_VERSION:-v0.5.2}"
+HERE="$(cd "$(dirname "$0")" && pwd)"
+DEST="${HERE}/_vendor"
+BINARY_NAME="cli-proxy-api-linux-amd64"
+URL="https://github.com/router-for-me/CLIProxyAPI/releases/download/${VERSION}/${BINARY_NAME}"
+
+mkdir -p "${DEST}"
+cd "${DEST}"
+
+if [[ -x ./cli-proxy-api ]]; then
+    echo "cli-proxy-api already present at ${DEST}/cli-proxy-api (delete to force re-download)"
+else
+    echo "Downloading ${URL}"
+    curl -fL --retry 3 -o cli-proxy-api "${URL}"
+    chmod +x cli-proxy-api
+fi
+
+if [[ ! -f config.yaml ]]; then
+    cp ../config.example.yaml config.yaml
+    echo "Copied agents/cliproxyapi/config.example.yaml → _vendor/config.yaml"
+    echo "  → Edit the 'api-keys:' list before first use."
+fi
+
+cat <<EOF
+
+Installed CLIProxyAPI ${VERSION} under ${DEST}
+
+Next steps:
+  1. Edit ${DEST}/config.yaml — set the 'api-keys:' list to a secret of your
+     choosing (the agent runners read CLIPROXYAPI_KEY which must match).
+  2. OAuth-login each upstream you intend to use, e.g.:
+       cd ${DEST}
+       ./cli-proxy-api --config config.yaml login claude
+       ./cli-proxy-api --config config.yaml login codex
+     The resulting token JSONs land in ~/.cli-proxy-api/ by default.
+  3. Start the proxy in the background:
+       cd ${DEST}
+       nohup ./cli-proxy-api --config config.yaml > proxy.log 2>&1 &
+  4. (Optional) Start the max_tokens-clamp passthrough used by aibuildai:
+       cd ${HERE%/cliproxyapi}/..   # repo root
+       .venv/bin/python -m agents.cliproxyapi.clamp_proxy &
+  5. export CLIPROXYAPI_KEY=<the api-key you set in step 1>
+EOF

datasets/manifest.yaml

@@ -2,17 +2,14 @@ ieee-fraud-detection:
   hf_repo: graphtestbed/ieee-fraud-detection
   hf_revision: main
   files:
-    train_transaction:
-      filename: train_transaction.csv
-      sha256: TBD
-    train_identity:
-      filename: train_identity.csv
+    train_features:
+      filename: train_features.csv
       sha256: TBD
-    test_transaction:
-      filename: test_transaction.csv
+    val_features:
+      filename: val_features.csv
       sha256: TBD
-    test_identity:
-      filename: test_identity.csv
+    test_features:
+      filename: test_features.csv
       sha256: TBD
     sample_submission:
       filename: sample_submission.csv
@@ -28,10 +25,16 @@ ieee-fraud-detection:
     - auc_pr
     - f1
   description: 'Predict the probability that an online transaction is fraudulent.
-    Source: Kaggle competition ieee-fraud-detection (Vesta). Backend: kaggle — server
-    forwards your CSV to Kaggle''s grading API (kaggle competitions submit -c ieee-fraud-detection)
-    and returns Kaggle''s publicScore as primary, privateScore as secondary. Submission
-    uses Kaggle''s full hidden test split (506,691 rows). Scoring takes 1–5 min —
+    Source: Kaggle competition ieee-fraud-detection (Vesta). The agent sees
+    train/val/test features that already merge transaction + identity tables on
+    TransactionID (left join). The val split is the last 20% of train by
+    TransactionDT (temporal), so use it for HPO. Test is Kaggle''s 506,691-row
+    hidden split — predictions are forwarded to Kaggle for scoring.
+
+
+    Backend: kaggle — server forwards your CSV to Kaggle''s grading API
+    (kaggle competitions submit -c ieee-fraud-detection) and returns Kaggle''s
+    publicScore as primary, privateScore as secondary. Scoring takes 1–5 min —
     be patient.