@onlooker-community/ecosystem 0.34.0 → 0.43.0

package/test/bats/librarian-lesson-transform.bats

@@ -220,13 +220,27 @@ _storage_setup() {
   [ "$status" -eq 0 ]
 }
 
-@test "seen finds an artifact already promoted into the approved pool" {
+@test "seen ignores the approved pool, which cannot identify an artifact" {
+  # This replaces a test that wrote {artifact_id: ...} into approved/ and
+  # asserted seen found it. That shape is unreachable — a pool entry is
+  # ZLesson, a strictObject with no artifact_id — so the old test fed the scan
+  # a file promote could never produce and reported coverage for a branch that
+  # never matched anything real. It was the reason the dead scan survived.
+  #
+  # ecosystem-d0m settled this: proposals/ is the sole dedup source. The
+  # end-to-end tripwire lives in librarian-lesson-promote.bats, against a real
+  # promoted entry; this one pins that a pool-shaped file is simply not
+  # consulted.
   _storage_setup
   librarian_lesson_storage_init "$PROJECT_KEY"
-  jq -n '{artifact_id: "01KZ45MKAM734ZS7JK24D2DK0R"}' \
+  jq -n '{id: "01KZ45MKGQ7QZWMABQ4H12SHSV", schema_version: 2, claim: "c",
+          rationale: "r", evidence: [], applies_to: [], visibility: "public",
+          consensus: {judges: 2, agreed: 2, decided_at: "2026-01-01T00:00:00Z"},
+          status: "active", superseded_by: null, source: "s",
+          author_key: "k", promoted_at: "2026-01-01T00:00:00Z"}' \
     > "${LESSONS_DIR}/approved/01KZ45MKGQ7QZWMABQ4H12SHSV.json"
   run librarian_lesson_seen "$PROJECT_KEY" "01KZ45MKAM734ZS7JK24D2DK0R"
-  [ "$status" -eq 0 ]
+  [ "$status" -eq 1 ]
 }
 
 @test "seen still finds a declined artifact when declined.jsonl has a truncated trailing line" {

package/test/bats/librarian-session-end.bats

@@ -42,10 +42,17 @@ setup() {
   # based on the artifact's summary contents.
   STUB_BIN="${BATS_TEST_TMPDIR}/bin"
   mkdir -p "$STUB_BIN"
+  # Every invocation is recorded so budget tests can assert on how many model
+  # calls a scan actually made — the only evidence that distinguishes an
+  # analyzer skipped by the budget from one that ran and produced nothing.
+  export CLAUDE_CALL_LOG="${BATS_TEST_TMPDIR}/claude-calls"
+  : > "$CLAUDE_CALL_LOG"
+
   cat > "${STUB_BIN}/claude" <<'STUB'
 #!/usr/bin/env bash
 # Read the prompt from stdin and decide which classifier response to emit.
 prompt=$(cat)
+[[ -n "${CLAUDE_CALL_LOG:-}" ]] && echo call >> "$CLAUDE_CALL_LOG"
 if [[ "$prompt" == *"prefer-functional-stub"* ]]; then
   printf '%s' '{"type":"feedback","title":"Prefer functional patterns","body":"User prefers functional patterns over class-based.\n\n**Why:** Stated explicitly during code review.\n**How to apply:** Default to plain functions and composition.","confidence":0.84}'
 elif [[ "$prompt" == *"compliance-stub"* ]]; then
@@ -176,3 +183,157 @@ _hook_input() {
   grep '"event_type":"librarian.scan.complete"' "$ONLOOKER_EVENTS_LOG" \
     | jq -e '.payload.outcome == "empty" and .payload.candidates_proposed == 0 and .payload.candidates_dropped >= 1' >/dev/null
 }
+
+# ---------------------------------------------------------------------------
+# Stage 5 aggregate budget (ecosystem-qwi).
+#
+# Each lesson transform carries a 20s ceiling of its own, but nothing bounded
+# KEPT_COUNT of them end to end, so a backlog could hold SessionEnd open for
+# minutes. These assert on model call counts, because that is what separates
+# "the budget skipped the loop" from "the loop ran and found nothing".
+# ---------------------------------------------------------------------------
+
+_llm_calls() {
+  wc -l < "$CLAUDE_CALL_LOG" | tr -d ' '
+}
+
+_settings() {
+  cat > "${PROJECT_REPO}/.claude/settings.json"
+}
+
+# Carries the marker phrase the durability filter wants AND a version token,
+# without which the lesson pregate rejects it before any model call and the
+# budget tests could not tell the two states apart.
+_seed_lessonable() {
+  _seed_artifact "decisions" "01LESSONBUDGETARTIFACT000" \
+    "User prefers functional patterns prefer-functional-stub" \
+    "User explicitly said: always prefer plain functions over classes. Pinned vite 5.4.21 to avoid the regression."
+}
+
+@test "stage 5 runs a lesson transform when the budget allows" {
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  # One classifier call plus one lesson call.
+  [ "$(_llm_calls)" -ge 2 ]
+}
+
+# A zero budget trips on the first iteration, so stage 5 spends nothing. The
+# classifier call still happens, which is what makes this a measurement of the
+# stage 5 guard rather than of the scan being disabled.
+@test "a zero budget skips stage 5 entirely" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  [ "$(_llm_calls)" = "1" ]
+}
+
+# The guard must fail toward skipping stage 5, never toward skipping the
+# watermark advance — an unadvanced watermark re-scans the same backlog every
+# session, turning a one-time cost into a permanent one.
+@test "the watermark still advances when the budget trips" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  [ -f "${LIBRARIAN_DIR}/last_scan.json" ] || return 1
+  jq -e '.scanned_at | test("^[0-9]{4}-[0-9]{2}-[0-9]{2}T")' "${LIBRARIAN_DIR}/last_scan.json" >/dev/null
+}
+
+@test "scan.complete is still emitted when the budget trips" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  grep '"event_type":"librarian.scan.complete"' "$ONLOOKER_EVENTS_LOG" \
+    | jq -e '.payload.outcome == "ok" or .payload.outcome == "empty"' >/dev/null
+}
+
+# Classification is upstream of stage 5, so a tripped lesson budget must not
+# cost the user their proposals.
+@test "a tripped lesson budget does not discard classifier proposals" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  grep '"event_type":"librarian.scan.complete"' "$ONLOOKER_EVENTS_LOG" \
+    | jq -e '.payload.candidates_proposed >= 1' >/dev/null
+}
+
+@test "the hook still exits 0 when the budget trips" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ]
+}
+
+# ---------------------------------------------------------------------------
+# Budget telemetry reaches the bus (ecosystem-1p1).
+#
+# librarian_emit swallows a validation failure and returns 0, appending to the
+# log only when the emitter accepted the payload. So presence on the bus is
+# proof the event validated — which is exactly the property that was missing:
+# the classifier gate's outcome:"budget_exceeded" was rejected by the schema
+# and silently dropped for as long as the gate has existed.
+# ---------------------------------------------------------------------------
+
+_scan_complete() {
+  grep '"event_type":"librarian.scan.complete"' "$ONLOOKER_EVENTS_LOG" | tail -1
+}
+
+@test "a truncated stage 5 reports how many artifacts it skipped" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  _scan_complete | jq -e '.payload.lessons_skipped >= 1' >/dev/null
+}
+
+# lessons_skipped accompanies a healthy outcome rather than replacing it: the
+# scan completed, only the one stage stopped early.
+@test "the skip count rides alongside a normal outcome" {
+  echo '{"librarian":{"lesson_transform":{"total_budget_ms":0}}}' | _settings
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  _scan_complete | jq -e '.payload.outcome == "ok" or .payload.outcome == "empty"' >/dev/null
+}
+
+@test "an untruncated scan reports zero skipped" {
+  _seed_lessonable
+  run bash -c "printf '%s' '$(_hook_input)' | '$HOOK'"
+  [ "$status" -eq 0 ] || return 1
+  _scan_complete | jq -e '.payload.lessons_skipped == 0' >/dev/null
+}
+
+# The regression guard for the original bug. This payload is what the classifier
+# gate builds verbatim; before schema 2.14.0 it failed validation and never
+# appeared on the bus at all.
+@test "the classifier gate's budget_exceeded payload reaches the bus" {
+  # shellcheck disable=SC1091
+  export _LIBRARIAN_EVENT_JS="${REPO_ROOT}/scripts/lib/onlooker-event.mjs"
+  source "${PLUGIN_ROOT}/scripts/lib/librarian-emit.sh"
+  mkdir -p "$(dirname "$ONLOOKER_EVENTS_LOG")"
+
+  librarian_emit "librarian.scan.complete" "sess-budget" "$(jq -cn \
+    '{ outcome: "budget_exceeded", duration_ms: 1200, candidates_proposed: 0,
+       candidates_dropped: 3, artifact_count_in_window: 5 }')"
+
+  _scan_complete | jq -e '.payload.outcome == "budget_exceeded"' >/dev/null
+}
+
+@test "an outcome outside the enum is still refused" {
+  export _LIBRARIAN_EVENT_JS="${REPO_ROOT}/scripts/lib/onlooker-event.mjs"
+  # shellcheck disable=SC1091
+  source "${PLUGIN_ROOT}/scripts/lib/librarian-emit.sh"
+  mkdir -p "$(dirname "$ONLOOKER_EVENTS_LOG")"
+
+  librarian_emit "librarian.scan.complete" "sess-bad" \
+    '{"outcome":"gave_up","duration_ms":5}'
+
+  # A refused payload never reaches the log, so the file may not exist at all.
+  # grep on a missing file errors rather than printing 0, which is why this
+  # asserts absence directly instead of comparing a count.
+  ! grep -q '"outcome":"gave_up"' "$ONLOOKER_EVENTS_LOG" 2>/dev/null
+}

package/test/bats/librarian-session-start.bats

@@ -83,7 +83,7 @@ _seed_proposal() {
   [ "$status" -eq 0 ]
   local ctx
   ctx=$(echo "$output" | jq -r '.hookSpecificOutput.additionalContext')
-  [[ "$ctx" == *"Librarian has 3 pending memory promotion proposals"* ]]
+  [[ "$ctx" == *"Librarian has 3 pending memory promotion proposals"* ]] || return 1
   [[ "$ctx" == *"/librarian review"* ]]
 }
 
@@ -94,7 +94,7 @@ _seed_proposal() {
   [ "$status" -eq 0 ]
   local ctx
   ctx=$(echo "$output" | jq -r '.hookSpecificOutput.additionalContext')
-  [[ "$ctx" == *"Librarian has 1 pending memory promotion proposal"* ]]
+  [[ "$ctx" == *"Librarian has 1 pending memory promotion proposal"* ]] || return 1
   [[ "$ctx" != *"proposals."* ]]
 }
 
@@ -124,3 +124,37 @@ _seed_proposal() {
   ctx=$(echo "$output" | jq -r '.hookSpecificOutput.additionalContext')
   [[ "$ctx" == *"Librarian has 3+ pending memory promotion proposals"* ]]
 }
+
+@test "session-start surfaces a pending lesson count as its own line" {
+	# Reuse this file's existing project/hook setup, then seed one pending
+	# lesson through the storage lib.
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/librarian-storage.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/librarian-ulid.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/librarian-lesson-storage.sh"
+	librarian_lesson_storage_init "$PROJECT_KEY"
+	librarian_lesson_write_proposal "$PROJECT_KEY" \
+		"$(jq -cn '{claim: "c", rationale: "r"}')" "01KZ45MKAM734ZS7JK24D2DK0R" >/dev/null
+
+	run bash -c "printf '%s' '$(_input)' | '$HOOK'"
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"lesson"* ]] || return 1
+}
+
+@test "session-start surfaces lessons even with zero memory proposals" {
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/librarian-storage.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/librarian-ulid.sh"
+	# shellcheck disable=SC1091
+	source "${PLUGIN_ROOT}/scripts/lib/librarian-lesson-storage.sh"
+	librarian_lesson_storage_init "$PROJECT_KEY"
+	librarian_lesson_write_proposal "$PROJECT_KEY" \
+		"$(jq -cn '{claim: "c", rationale: "r"}')" "01KZ45MKAM734ZS7JK24D2DK0R" >/dev/null
+
+	run bash -c "printf '%s' '$(_input)' | '$HOOK'"
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"lesson"* ]] || return 1
+}

package/test/bats/lineage-config.bats

@@ -43,6 +43,6 @@ setup() {
 	lineage_config_load ""
 	run lineage_config_ignore_globs
 	[ "$status" -eq 0 ]
-	[[ "$output" == *"node_modules"* ]]
+	[[ "$output" == *"node_modules"* ]] || return 1
 	[[ "$output" == *".lock"* ]]
 }

package/test/bats/lineage-redact.bats

@@ -21,7 +21,7 @@ setup() {
 	local tok out
 	tok="sk-ant-$(printf 'a%.0s' $(seq 1 28))"
 	out=$(printf '%s' "key=\"${tok}\"" | lineage_redact 4000 true)
-	[[ "$out" == *"[REDACTED:secret]"* ]]
+	[[ "$out" == *"[REDACTED:secret]"* ]] || return 1
 	[[ "$out" != *"$tok"* ]]
 }
 
@@ -29,7 +29,7 @@ setup() {
 	local tok out
 	tok="ghp_$(printf '0%.0s' $(seq 1 36))"
 	out=$(printf '%s' "$tok" | lineage_redact 4000 true)
-	[[ "$out" == *"[REDACTED:secret]"* ]]
+	[[ "$out" == *"[REDACTED:secret]"* ]] || return 1
 	[[ "$out" != *"$tok"* ]]
 }
 
@@ -37,8 +37,8 @@ setup() {
 	local val out
 	val=$(printf 'x%.0s' $(seq 1 24))
 	out=$(printf '%s' "API_TOKEN=${val}" | lineage_redact 4000 true)
-	[[ "$out" == *"API_TOKEN="* ]]
-	[[ "$out" == *"[REDACTED:secret]"* ]]
+	[[ "$out" == *"API_TOKEN="* ]] || return 1
+	[[ "$out" == *"[REDACTED:secret]"* ]] || return 1
 	[[ "$out" != *"$val"* ]]
 }
 
@@ -52,7 +52,7 @@ setup() {
 @test "content longer than the cap is truncated with a marker" {
 	local out
 	out=$(printf '%s' "abcdefghij" | lineage_redact 4 true)
-	[[ "$out" == "abcd"* ]]
+	[[ "$out" == "abcd"* ]] || return 1
 	[[ "$out" == *"truncated 6 chars"* ]]
 }
 

package/test/bats/session-tracker.bats

@@ -90,7 +90,7 @@ seed_start_ms_ago() {
 @test "session_tracker_now_ms returns 13-digit epoch milliseconds" {
   local ms
   ms=$(session_tracker_now_ms)
-  [[ "$ms" =~ ^[0-9]+$ ]]
+  [[ "$ms" =~ ^[0-9]+$ ]] || return 1
   [ "${#ms}" -eq 13 ]
 }
 
@@ -231,7 +231,7 @@ seed_start_ms_ago() {
   seed_start_ms_ago "$sid" 3000
   local dur
   dur=$(session_tracker_duration_ms "$sid")
-  [[ "$dur" =~ ^[0-9]+$ ]]
+  [[ "$dur" =~ ^[0-9]+$ ]] || return 1
   [ "$dur" -ge 3000 ]
 }
 
@@ -285,7 +285,7 @@ seed_start_ms_ago() {
 
   local out
   out=$(session_tracker_build_duration_context "$sid")
-  [[ "$out" == *"turn 3"* ]]
+  [[ "$out" == *"turn 3"* ]] || return 1
   [[ "$out" == *"2m 34s"* ]]
 }
 
@@ -294,7 +294,7 @@ seed_start_ms_ago() {
   rm -f "${ONLOOKER_SESSION_TRACKERS_DIR}/${sid}"
   local out
   out=$(session_tracker_build_duration_context "$sid")
-  [[ "$out" == *"turn 1"* ]]
+  [[ "$out" == *"turn 1"* ]] || return 1
   [[ "$out" == *"0s"* ]]
 }