Files
tkmind_go/scripts/test_providers.sh
T

58 lines
1.7 KiB
Bash
Raw Normal View History

2025-10-18 17:35:29 -04:00
#!/bin/bash
2026-02-12 21:12:11 +11:00
LIB_DIR="$(cd "$(dirname "$0")" && pwd)"
source "$LIB_DIR/test_providers_lib.sh"
2026-02-12 21:12:11 +11:00
echo "Mode: normal (direct tool calls)"
echo ""
2025-10-18 17:35:29 -04:00
2026-02-12 21:12:11 +11:00
GOOSE_BIN=$(build_goose)
BUILTINS="developer"
2025-10-18 17:35:29 -04:00
mkdir -p target
TEST_CONTENT="test-content-abc123"
TEST_FILE="./target/test-content.txt"
echo "$TEST_CONTENT" > "$TEST_FILE"
2026-02-03 11:21:59 -05:00
run_test() {
2026-02-12 21:12:11 +11:00
local provider="$1" model="$2" result_file="$3" output_file="$4"
2026-02-03 11:21:59 -05:00
local testdir=$(mktemp -d)
local prompt
if is_agentic_provider "$provider"; then
cp "$TEST_FILE" "$testdir/test-content.txt"
prompt="read ./test-content.txt and output its contents exactly"
else
2026-02-12 21:12:11 +11:00
echo "$TEST_CONTENT" > "$testdir/input.txt"
prompt="Use the text_editor view command to read ./input.txt, then output this file's contents in UPPERCASE. Do NOT use any other tool in Developer"
fi
2026-02-03 11:21:59 -05:00
(
export GOOSE_PROVIDER="$provider"
export GOOSE_MODEL="$model"
2026-02-12 21:12:11 +11:00
cd "$testdir" && "$GOOSE_BIN" run --text "$prompt" --with-builtin "$BUILTINS" 2>&1
2026-02-03 11:21:59 -05:00
) > "$output_file" 2>&1
if is_agentic_provider "$provider"; then
if grep -qi "$TEST_CONTENT" "$output_file"; then
2026-02-12 21:12:11 +11:00
echo "success|test content found by model" > "$result_file"
else
2026-02-12 21:12:11 +11:00
echo "failure|test content not found by model" > "$result_file"
fi
2026-02-03 11:21:59 -05:00
else
if ! grep -qE "(text_editor \| developer)|(▸.*text_editor.*developer)" "$output_file"; then
2026-02-12 21:12:11 +11:00
echo "failure|model did not use text_editor tool" > "$result_file"
elif ! grep -q "TEST-CONTENT-ABC123" "$output_file"; then
echo "failure|model did not return uppercased file content" > "$result_file"
2026-02-03 11:21:59 -05:00
else
2026-02-12 21:12:11 +11:00
echo "success|model read and uppercased file content" > "$result_file"
2026-02-03 11:21:59 -05:00
fi
fi
2026-02-12 21:12:11 +11:00
rm -rf "$testdir"
}
2026-02-12 21:12:11 +11:00
build_test_cases
run_test_cases run_test
report_results