@onlooker-community/ecosystem 0.9.0 → 0.14.0

package/test/bats/tribunal-aggregate.bats

@@ -0,0 +1,77 @@
+#!/usr/bin/env bats
+
+setup() {
+	source "${BATS_TEST_DIRNAME}/../helpers/setup.bash"
+	setup_test_env
+
+	PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal"
+	export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-aggregate.sh"
+}
+
+VERDICTS='[{"judge_id":"a","score":0.8},{"judge_id":"b","score":0.6},{"judge_id":"c","score":0.4}]'
+
+@test "mean of [0.8, 0.6, 0.4] is 0.6" {
+	local v
+	v=$(tribunal_aggregate "mean" "$VERDICTS")
+	awk -v v="$v" 'BEGIN { exit !(v > 0.59 && v < 0.61) }'
+}
+
+@test "median of three is the middle" {
+	local v
+	v=$(tribunal_aggregate "median" "$VERDICTS")
+	awk -v v="$v" 'BEGIN { exit !(v > 0.59 && v < 0.61) }'
+}
+
+@test "median of four averages the two middle scores" {
+	local four='[{"judge_id":"a","score":0.2},{"judge_id":"b","score":0.4},{"judge_id":"c","score":0.6},{"judge_id":"d","score":0.8}]'
+	local v
+	v=$(tribunal_aggregate "median" "$four")
+	awk -v v="$v" 'BEGIN { exit !(v > 0.49 && v < 0.51) }'
+}
+
+@test "min picks the lowest score" {
+	local v
+	v=$(tribunal_aggregate "min" "$VERDICTS")
+	awk -v v="$v" 'BEGIN { exit !(v > 0.39 && v < 0.41) }'
+}
+
+@test "weighted_mean degrades to mean in v0.1" {
+	local v
+	v=$(tribunal_aggregate "weighted_mean" "$VERDICTS")
+	awk -v v="$v" 'BEGIN { exit !(v > 0.59 && v < 0.61) }'
+}
+
+@test "unknown method falls back to mean with warning on stderr" {
+	run bash -c '
+		source "${REPO_ROOT}/plugins/tribunal/scripts/lib/tribunal-aggregate.sh"
+		tribunal_aggregate "lottery" "[{\"score\":0.5},{\"score\":0.7}]" 2>&1
+	'
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"unknown method lottery"* ]]
+}
+
+@test "empty verdicts aggregates to 0" {
+	local v
+	v=$(tribunal_aggregate "mean" "[]")
+	[ "$v" = "0" ]
+}
+
+@test "disagreement of identical scores is 0" {
+	local d
+	d=$(tribunal_disagreement '[{"score":0.7},{"score":0.7}]')
+	awk -v d="$d" 'BEGIN { exit !(d < 0.01) }'
+}
+
+@test "disagreement of [0.2, 0.8] is 0.6" {
+	local d
+	d=$(tribunal_disagreement '[{"score":0.2},{"score":0.8}]')
+	awk -v d="$d" 'BEGIN { exit !(d > 0.59 && d < 0.61) }'
+}
+
+@test "disagreement of single verdict is 0" {
+	local d
+	d=$(tribunal_disagreement '[{"score":0.7}]')
+	[ "$d" = "0" ]
+}

package/test/bats/tribunal-config.bats

@@ -0,0 +1,86 @@
+#!/usr/bin/env bats
+
+setup() {
+	source "${BATS_TEST_DIRNAME}/../helpers/setup.bash"
+	setup_test_env
+
+	PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal"
+	export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-config.sh"
+}
+
+@test "config defaults from plugin config.json: enabled" {
+	tribunal_config_load ""
+	run tribunal_config_enabled
+	[ "$status" -eq 0 ]
+}
+
+@test "stop_hook defaults to disabled" {
+	tribunal_config_load ""
+	run tribunal_config_stop_hook_enabled
+	[ "$status" -ne 0 ]
+}
+
+@test "user-level settings.json overlay can disable tribunal" {
+	mkdir -p "${HOME}/.claude"
+	printf '%s\n' '{"tribunal":{"enabled":false}}' > "${HOME}/.claude/settings.json"
+	tribunal_config_load ""
+	run tribunal_config_enabled
+	[ "$status" -ne 0 ]
+}
+
+@test "repo-level settings.json overrides user-level" {
+	mkdir -p "${HOME}/.claude"
+	printf '%s\n' '{"tribunal":{"enabled":false}}' > "${HOME}/.claude/settings.json"
+	local repo="${BATS_TEST_TMPDIR}/repo"
+	mkdir -p "${repo}/.claude"
+	printf '%s\n' '{"tribunal":{"enabled":true}}' > "${repo}/.claude/settings.json"
+	tribunal_config_load "$repo"
+	run tribunal_config_enabled
+	[ "$status" -eq 0 ]
+}
+
+@test "default judge_types is standard + adversarial" {
+	tribunal_config_load ""
+	local types
+	types=$(tribunal_config_get_json '.tribunal.session.judge_types')
+	[ "$types" = '["standard","adversarial"]' ]
+}
+
+@test "default gate_policy is majority" {
+	tribunal_config_load ""
+	local v
+	v=$(tribunal_config_get '.tribunal.session.gate_policy')
+	[ "$v" = "majority" ]
+}
+
+@test "judge model falls back to tribunal.judges.model when no per-type override" {
+	tribunal_config_load ""
+	local m
+	m=$(tribunal_config_judge_model "standard")
+	[ "$m" = "claude-opus-4-7" ]
+}
+
+@test "per-judge-type model override wins over fallback" {
+	mkdir -p "${HOME}/.claude"
+	printf '%s\n' '{"tribunal":{"judges":{"security":{"model":"claude-opus-4-7-deep"}}}}' > "${HOME}/.claude/settings.json"
+	tribunal_config_load ""
+	local m
+	m=$(tribunal_config_judge_model "security")
+	[ "$m" = "claude-opus-4-7-deep" ]
+	# Other types still fall through to the default.
+	m=$(tribunal_config_judge_model "standard")
+	[ "$m" = "claude-opus-4-7" ]
+}
+
+@test "deep-merge preserves unset defaults" {
+	mkdir -p "${HOME}/.claude"
+	printf '%s\n' '{"tribunal":{"session":{"max_iterations":7}}}' > "${HOME}/.claude/settings.json"
+	tribunal_config_load ""
+	local mi gp
+	mi=$(tribunal_config_get '.tribunal.session.max_iterations')
+	gp=$(tribunal_config_get '.tribunal.session.gate_policy')
+	[ "$mi" = "7" ]
+	[ "$gp" = "majority" ]
+}

package/test/bats/tribunal-events.bats

@@ -0,0 +1,209 @@
+#!/usr/bin/env bats
+
+# Validates every emitted tribunal.* event against @onlooker-community/schema
+# v2.1.0+. Builds a single event via the canonical emitter and asserts the
+# resulting JSONL line passes validate().
+
+setup() {
+	source "${BATS_TEST_DIRNAME}/../helpers/setup.bash"
+	setup_test_env
+
+	PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal"
+	export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
+	export ONLOOKER_EVENTS_LOG="${ONLOOKER_DIR}/logs/onlooker-events.jsonl"
+	mkdir -p "$(dirname "$ONLOOKER_EVENTS_LOG")"
+
+	# tribunal-events.sh looks up onlooker-event.mjs relative to its plugin
+	# root, but tests set CLAUDE_PLUGIN_ROOT to plugins/tribunal — point the
+	# wrapper at the ecosystem copy directly so it does not have to walk.
+	export _ONLOOKER_EVENT_JS="${REPO_ROOT}/scripts/lib/onlooker-event.mjs"
+
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-events.sh"
+
+	export CLAUDE_SESSION_ID="bats-session-$$"
+}
+
+# Re-validate the latest event in the log against the schema.
+_validate_latest_event() {
+	local last
+	last=$(tail -n 1 "$ONLOOKER_EVENTS_LOG")
+	[ -n "$last" ] || return 1
+	printf '%s' "$last" | ONLOOKER_DIR="$ONLOOKER_DIR" \
+		node "${REPO_ROOT}/scripts/lib/onlooker-event.mjs" validate >/dev/null
+}
+
+TASK_ID="01J000000000000000000000TS"
+ITER_ID="01J000000000000000000000IT"
+JUDGE_ID="01J000000000000000000000JJ"
+
+@test "session.start validates" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" '{
+		task_id: $t,
+		judge_types: ["standard","adversarial"],
+		gate_policy: "majority",
+		score_threshold: 0.75,
+		max_iterations: 3,
+		actor_model_id: "claude-sonnet-4-6",
+		judge_model_ids: ["claude-opus-4-7","claude-opus-4-7"],
+		meta_model_id: "claude-opus-4-7"
+	}')
+	tribunal_emit_event "tribunal.session.start" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "iteration.start validates" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" \
+		'{task_id: $t, iteration_id: $i, iteration_number: 0, trigger: "initial"}')
+	tribunal_emit_event "tribunal.iteration.start" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "actor.start + actor.complete validate" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" \
+		'{task_id: $t, iteration_id: $i, iteration_number: 0, actor_model_id: "claude-sonnet-4-6"}')
+	tribunal_emit_event "tribunal.actor.start" "$p"
+	_validate_latest_event
+
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" \
+		'{task_id: $t, success: true, duration_ms: 4200, iteration_id: $i, iteration_number: 0, artifact_kind: "patch", actor_model_id: "claude-sonnet-4-6"}')
+	tribunal_emit_event "tribunal.actor.complete" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "jury.empaneled validates" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" --arg j "$JUDGE_ID" '{
+		task_id: $t,
+		iteration_id: $i,
+		judges: [
+			{judge_id: $j, judge_type: "standard", model_id: "claude-opus-4-7"},
+			{judge_id: ($j+"X"), judge_type: "adversarial", model_id: "claude-opus-4-7"}
+		],
+		panel_size: 2
+	}')
+	tribunal_emit_event "tribunal.jury.empaneled" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "judge.start + verdict validate" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" --arg j "$JUDGE_ID" \
+		'{task_id: $t, iteration_id: $i, judge_id: $j, judge_type: "standard", judge_model_id: "claude-opus-4-7"}')
+	tribunal_emit_event "tribunal.judge.start" "$p"
+	_validate_latest_event
+
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" --arg j "$JUDGE_ID" '{
+		task_id: $t,
+		score: 0.82,
+		passed: true,
+		judge_type: "standard",
+		iteration_id: $i,
+		judge_id: $j,
+		judge_model_id: "claude-opus-4-7",
+		criteria_evaluated: ["correctness","completeness","clarity"],
+		strengths_count: 3,
+		weaknesses_count: 1,
+		confidence: 0.85,
+		feedback_summary: "looks fine"
+	}')
+	tribunal_emit_event "tribunal.verdict" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "meta.start + meta.complete validate (with bias_types)" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" \
+		'{task_id: $t, iteration_id: $i, meta_model_id: "claude-opus-4-7", verdicts_reviewed: 2}')
+	tribunal_emit_event "tribunal.meta.start" "$p"
+	_validate_latest_event
+
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" '{
+		task_id: $t,
+		verdict_quality: "questionable",
+		bias_detected: true,
+		bias_types: ["verbosity","sycophancy"],
+		override_recommendation: "re-evaluate",
+		confidence: 0.7,
+		iteration_id: $i,
+		meta_model_id: "claude-opus-4-7"
+	}')
+	tribunal_emit_event "tribunal.meta.complete" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "consensus.reached validates" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" --arg j "$JUDGE_ID" '{
+		task_id: $t,
+		iteration_id: $i,
+		aggregated_score: 0.7,
+		passed: true,
+		aggregation_method: "weighted_mean",
+		judges: [{judge_id: $j, score: 0.8},{judge_id: ($j+"X"), score: 0.6}]
+	}')
+	tribunal_emit_event "tribunal.consensus.reached" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "dissent.recorded validates with re-evaluate resolution" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" --arg j "$JUDGE_ID" '{
+		task_id: $t,
+		iteration_id: $i,
+		disagreement_score: 0.5,
+		judges: [
+			{judge_id: $j, score: 0.85, passed: true},
+			{judge_id: ($j+"X"), score: 0.35, passed: false}
+		],
+		resolution: "re-evaluate"
+	}')
+	tribunal_emit_event "tribunal.dissent.recorded" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "gate.passed and gate.blocked validate" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" \
+		'{task_id: $t, iteration_id: $i, final_score: 0.82, iteration_number: 0, judges_consulted: 2}')
+	tribunal_emit_event "tribunal.gate.passed" "$p"
+	_validate_latest_event
+
+	p=$(jq -n --arg t "$TASK_ID" --arg i "$ITER_ID" '{
+		task_id: $t,
+		iteration_id: $i,
+		reason: "low_score",
+		final_score: 0.42,
+		iteration_number: 0,
+		will_retry: true,
+		retry_iteration_number: 1
+	}')
+	tribunal_emit_event "tribunal.gate.blocked" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "session.complete validates with exhausted_iterations outcome" {
+	local p
+	p=$(jq -n --arg t "$TASK_ID" \
+		'{task_id: $t, outcome: "exhausted_iterations", final_score: 0.55, iterations_used: 3, total_duration_ms: 28000}')
+	tribunal_emit_event "tribunal.session.complete" "$p"
+	run _validate_latest_event
+	[ "$status" -eq 0 ]
+}
+
+@test "emission fails loudly on bogus event_type (schema rejects)" {
+	run tribunal_emit_event "tribunal.no.such.event" '{"task_id":"x"}'
+	[ "$status" -ne 0 ]
+}

package/test/bats/tribunal-gate.bats

@@ -0,0 +1,95 @@
+#!/usr/bin/env bats
+
+setup() {
+	source "${BATS_TEST_DIRNAME}/../helpers/setup.bash"
+	setup_test_env
+
+	PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal"
+	export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-gate.sh"
+}
+
+ALL_PASSED='[{"judge_id":"a","score":0.85,"passed":true},{"judge_id":"b","score":0.80,"passed":true}]'
+ONE_FAILED='[{"judge_id":"a","score":0.85,"passed":true},{"judge_id":"b","score":0.40,"passed":false}]'
+ALL_FAILED='[{"judge_id":"a","score":0.30,"passed":false},{"judge_id":"b","score":0.40,"passed":false}]'
+NO_META='{}'
+
+@test "strict: all judges pass + score >= threshold → passed" {
+	local out
+	out=$(tribunal_gate_decide "strict" "$ALL_PASSED" "0.82" "0.75" "$NO_META" "0.05" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "true" ]
+}
+
+@test "strict: one judge fails → blocked with dissent_unresolved or low_score" {
+	local out
+	out=$(tribunal_gate_decide "strict" "$ONE_FAILED" "0.62" "0.75" "$NO_META" "0.45" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+	# dissent + no meta override → dissent_unresolved
+	[ "$(printf '%s' "$out" | jq -r '.reason')" = "dissent_unresolved" ]
+}
+
+@test "majority: more than half pass + score clears → passed" {
+	local three='[{"judge_id":"a","score":0.9,"passed":true},{"judge_id":"b","score":0.8,"passed":true},{"judge_id":"c","score":0.4,"passed":false}]'
+	local out
+	out=$(tribunal_gate_decide "majority" "$three" "0.78" "0.75" "$NO_META" "0.20" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "true" ]
+}
+
+@test "majority: split 1-1 with low score → blocked low_score" {
+	local out
+	out=$(tribunal_gate_decide "majority" "$ONE_FAILED" "0.62" "0.75" "$NO_META" "0.20" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+	[ "$(printf '%s' "$out" | jq -r '.reason')" = "low_score" ]
+}
+
+@test "unanimous: identical to strict when count > 1" {
+	local out
+	out=$(tribunal_gate_decide "unanimous" "$ALL_PASSED" "0.82" "0.75" "$NO_META" "0.05" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "true" ]
+
+	out=$(tribunal_gate_decide "unanimous" "$ONE_FAILED" "0.62" "0.75" "$NO_META" "0.05" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+}
+
+@test "meta_override accept beats failing jury" {
+	local meta='{"override_recommendation":"accept","bias_detected":false}'
+	local out
+	out=$(tribunal_gate_decide "meta_override" "$ALL_FAILED" "0.30" "0.75" "$meta" "0.10" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "true" ]
+}
+
+@test "meta_override reject blocks even with passing jury" {
+	local meta='{"override_recommendation":"reject","bias_detected":false}'
+	local out
+	out=$(tribunal_gate_decide "meta_override" "$ALL_PASSED" "0.82" "0.75" "$meta" "0.05" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+	[ "$(printf '%s' "$out" | jq -r '.reason')" = "meta_override" ]
+}
+
+@test "bias_detected + meta says reject → bias_detected reason" {
+	local meta='{"override_recommendation":"reject","bias_detected":true,"bias_types":["verbosity"]}'
+	local out
+	out=$(tribunal_gate_decide "majority" "$ONE_FAILED" "0.60" "0.75" "$meta" "0.45" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+	[ "$(printf '%s' "$out" | jq -r '.reason')" = "bias_detected" ]
+}
+
+@test "dissent above threshold + no meta override → dissent_unresolved" {
+	local meta='{"bias_detected":false}'
+	local out
+	out=$(tribunal_gate_decide "majority" "$ONE_FAILED" "0.62" "0.50" "$meta" "0.45" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+	[ "$(printf '%s' "$out" | jq -r '.reason')" = "dissent_unresolved" ]
+}
+
+@test "score clears threshold but jury says no → meta_override or dissent reason" {
+	# All judges marked passed=false but aggregated_score is above threshold
+	# (contrived to exercise the "score_ok + jury_fail" branch).
+	local odd='[{"judge_id":"a","score":0.9,"passed":false},{"judge_id":"b","score":0.8,"passed":false}]'
+	local meta='{"override_recommendation":"reject","bias_detected":false}'
+	local out
+	out=$(tribunal_gate_decide "majority" "$odd" "0.85" "0.75" "$meta" "0.10" "0.25")
+	[ "$(printf '%s' "$out" | jq -r '.passed')" = "false" ]
+	[ "$(printf '%s' "$out" | jq -r '.reason')" = "meta_override" ]
+}

package/test/bats/tribunal-jury.bats

@@ -0,0 +1,80 @@
+#!/usr/bin/env bats
+
+setup() {
+	source "${BATS_TEST_DIRNAME}/../helpers/setup.bash"
+	setup_test_env
+
+	PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal"
+	export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-config.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-ulid.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-jury.sh"
+
+	tribunal_config_load ""
+}
+
+@test "default empanel produces standard + adversarial" {
+	local panel
+	panel=$(tribunal_jury_empanel '["standard","adversarial"]')
+	[ "$(printf '%s' "$panel" | jq 'length')" = "2" ]
+	[ "$(printf '%s' "$panel" | jq -r '.[0].judge_type')" = "standard" ]
+	[ "$(printf '%s' "$panel" | jq -r '.[1].judge_type')" = "adversarial" ]
+}
+
+@test "each panel member gets a distinct judge_id" {
+	local panel
+	panel=$(tribunal_jury_empanel '["standard","adversarial","security"]')
+	local distinct
+	distinct=$(printf '%s' "$panel" | jq -r '[.[].judge_id] | unique | length')
+	[ "$distinct" = "3" ]
+}
+
+@test "panel members get model from config" {
+	local panel m
+	panel=$(tribunal_jury_empanel '["standard"]')
+	m=$(printf '%s' "$panel" | jq -r '.[0].model')
+	[ "$m" = "claude-opus-4-7" ]
+}
+
+@test "maintainability degrades to standard with warning" {
+	run bash -c '
+		source "${REPO_ROOT}/plugins/tribunal/scripts/lib/tribunal-config.sh"
+		source "${REPO_ROOT}/plugins/tribunal/scripts/lib/tribunal-ulid.sh"
+		source "${REPO_ROOT}/plugins/tribunal/scripts/lib/tribunal-jury.sh"
+		CLAUDE_PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal" tribunal_config_load ""
+		tribunal_jury_empanel "[\"maintainability\"]" 2>&1
+	'
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"degrading to standard"* ]]
+	[[ "$output" == *"standard"* ]]
+}
+
+@test "meta type is refused in jury panel" {
+	local panel
+	panel=$(tribunal_jury_empanel '["standard","meta"]' 2>/dev/null)
+	[ "$(printf '%s' "$panel" | jq 'length')" = "1" ]
+	[ "$(printf '%s' "$panel" | jq -r '.[0].judge_type')" = "standard" ]
+}
+
+@test "subagent mapping is canonical per judge_type" {
+	local panel
+	panel=$(tribunal_jury_empanel '["standard","security","adversarial"]')
+	[ "$(printf '%s' "$panel" | jq -r '.[0].subagent')" = "tribunal-judge-standard" ]
+	[ "$(printf '%s' "$panel" | jq -r '.[1].subagent')" = "tribunal-judge-security" ]
+	[ "$(printf '%s' "$panel" | jq -r '.[2].subagent')" = "tribunal-judge-adversarial" ]
+}
+
+@test "to_schema_judges strips internal subagent field" {
+	local panel schema
+	panel=$(tribunal_jury_empanel '["standard"]')
+	schema=$(tribunal_jury_to_schema_judges "$panel")
+	# subagent must NOT appear in the schema-shape output
+	[ "$(printf '%s' "$schema" | jq -r '.[0] | has("subagent")')" = "false" ]
+	# but judge_id, judge_type, model_id must
+	[ "$(printf '%s' "$schema" | jq -r '.[0] | has("judge_id")')" = "true" ]
+	[ "$(printf '%s' "$schema" | jq -r '.[0] | has("judge_type")')" = "true" ]
+	[ "$(printf '%s' "$schema" | jq -r '.[0] | has("model_id")')" = "true" ]
+}

package/test/bats/tribunal-rubric.bats

@@ -0,0 +1,119 @@
+#!/usr/bin/env bats
+
+setup() {
+	source "${BATS_TEST_DIRNAME}/../helpers/setup.bash"
+	setup_test_env
+
+	PLUGIN_ROOT="${REPO_ROOT}/plugins/tribunal"
+	export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-config.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/tribunal-rubric.sh"
+
+	tribunal_config_load ""
+}
+
+@test "default rubric loads from builtins" {
+	tribunal_rubric_load ""
+	local r id
+	r=$(tribunal_rubric_get "default")
+	[ -n "$r" ]
+	id=$(printf '%s' "$r" | jq -r '.id')
+	[ "$id" = "default" ]
+}
+
+@test "default rubric id resolves to 'default'" {
+	local id
+	id=$(tribunal_rubric_default_id)
+	[ "$id" = "default" ]
+}
+
+@test "default rubric passes validation" {
+	tribunal_rubric_load ""
+	local r
+	r=$(tribunal_rubric_get "default")
+	run tribunal_rubric_validate "$r"
+	[ "$status" -eq 0 ]
+}
+
+@test "project rubric override by id replaces builtin" {
+	local repo="${BATS_TEST_TMPDIR}/repo"
+	mkdir -p "${repo}/.claude"
+	cat > "${repo}/.claude/tribunal.json" <<'EOF'
+{
+  "rubrics": [
+    {
+      "id": "default",
+      "criteria": [
+        { "name": "tests", "weight": 1.0, "min_pass": 0.9 }
+      ],
+      "score_threshold": 0.9,
+      "max_iterations": 5,
+      "judge_types": ["standard"],
+      "gate_policy": "strict",
+      "aggregation_method": "min"
+    }
+  ]
+}
+EOF
+	tribunal_rubric_load "$repo"
+	local r mi gp
+	r=$(tribunal_rubric_get "default")
+	mi=$(printf '%s' "$r" | jq -r '.max_iterations')
+	gp=$(printf '%s' "$r" | jq -r '.gate_policy')
+	[ "$mi" = "5" ]
+	[ "$gp" = "strict" ]
+}
+
+@test "named rubric from project file is reachable by id" {
+	local repo="${BATS_TEST_TMPDIR}/repo"
+	mkdir -p "${repo}/.claude"
+	cat > "${repo}/.claude/tribunal.json" <<'EOF'
+{
+  "rubrics": [
+    {
+      "id": "security-tight",
+      "criteria": [
+        { "name": "security", "weight": 1.0, "min_pass": 0.95 }
+      ],
+      "score_threshold": 0.95,
+      "max_iterations": 3,
+      "judge_types": ["standard", "security"],
+      "gate_policy": "unanimous",
+      "aggregation_method": "min"
+    }
+  ]
+}
+EOF
+	tribunal_rubric_load "$repo"
+	local r
+	r=$(tribunal_rubric_get "security-tight")
+	[ -n "$r" ]
+	[ "$(printf '%s' "$r" | jq -r '.id')" = "security-tight" ]
+}
+
+@test "missing rubric id returns empty" {
+	tribunal_rubric_load ""
+	local r
+	r=$(tribunal_rubric_get "does-not-exist")
+	[ -z "$r" ]
+}
+
+@test "validate rejects weights summing != 1" {
+	local r='{"id":"bad","criteria":[{"name":"a","weight":0.4,"min_pass":0.5},{"name":"b","weight":0.4,"min_pass":0.5}],"score_threshold":0.75,"max_iterations":3,"judge_types":["standard"],"gate_policy":"majority","aggregation_method":"mean"}'
+	run tribunal_rubric_validate "$r"
+	[ "$status" -ne 0 ]
+}
+
+@test "validate rejects invalid gate_policy" {
+	local r='{"id":"bad","criteria":[{"name":"a","weight":1.0,"min_pass":0.5}],"score_threshold":0.75,"max_iterations":3,"judge_types":["standard"],"gate_policy":"democracy","aggregation_method":"mean"}'
+	run tribunal_rubric_validate "$r"
+	[ "$status" -ne 0 ]
+}
+
+@test "validate rejects out-of-range score_threshold" {
+	local r='{"id":"bad","criteria":[{"name":"a","weight":1.0,"min_pass":0.5}],"score_threshold":1.5,"max_iterations":3,"judge_types":["standard"],"gate_policy":"majority","aggregation_method":"mean"}'
+	run tribunal_rubric_validate "$r"
+	[ "$status" -ne 0 ]
+}