@mmerterden/multi-agent-pipeline 8.6.0

package/pipeline/scripts/smoke-tasklist-ordering.sh

@@ -0,0 +1,111 @@
+#!/usr/bin/env bash
+# smoke-tasklist-ordering.sh — v8.3.1
+#
+# Enforces the TaskCreate ordering contract: every mode entry point doc on
+# Claude Code AND the Copilot full-inline orchestrator MUST carry the
+# explicit "TaskCreate calls MUST fire in strict phase-number order BEFORE
+# any TaskUpdate is applied" rule.
+#
+# Why this smoke exists: the native TaskList widget renders tiles in
+# TaskCreate creation order, not by phase-number metadata. An agent that
+# pre-creates "completed" tiles for skipped phases before Phase 0 starts
+# produces visually scrambled tile stacks (1✓ · 2✓ · 4✓ · 0▶ · 3☐) even
+# when the underlying tracker state is correct. Centralizing the rule in
+# refs/tracker-contract.md is necessary but not sufficient — the LLM
+# orchestrating each mode reads the mode entry doc, so the rule must
+# appear in every mode entry doc.
+#
+# Exit 0 = all required surfaces declare the rule, 1 = any missing.
+
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+
+pass=0
+fail=0
+failures=()
+record_pass() { pass=$((pass + 1)); printf '  \033[0;32mPASS\033[0m %s\n' "$1"; }
+record_fail() { fail=$((fail + 1)); failures+=("$1"); printf '  \033[0;31mFAIL\033[0m %s\n' "$1"; }
+
+printf '→ smoke-tasklist-ordering (v8.3.1): TaskCreate phase-number ordering contract\n'
+
+# 1. tracker-contract.md — primary source of the rule
+TRACKER_CONTRACT="$ROOT/pipeline/commands/multi-agent/refs/tracker-contract.md"
+if grep -qE 'TaskCreate ordering \(strict\)' "$TRACKER_CONTRACT" \
+   && grep -qE 'phase-number order' "$TRACKER_CONTRACT" \
+   && grep -qE 'BEFORE any TaskUpdate' "$TRACKER_CONTRACT"; then
+  record_pass "tracker-contract.md declares the canonical ordering rule"
+else
+  record_fail "tracker-contract.md missing the canonical ordering rule"
+fi
+
+# 2. Phase 0 init ref doc — the agent always reads this on Phase 0
+PHASE0="$ROOT/pipeline/commands/multi-agent/refs/phases/phase-0-init.md"
+if grep -qE '(strict|required).*(TaskCreate|ordering)' "$PHASE0" \
+   && grep -qE 'phase-number order' "$PHASE0"; then
+  record_pass "phase-0-init.md reinforces the ordering rule on Phase 0 entry"
+else
+  record_fail "phase-0-init.md missing the ordering reinforcement"
+fi
+
+# 3. Every Claude Code mode entry doc — the rule must appear in each
+MODE_ENTRY_DOCS=(
+  "pipeline/commands/multi-agent/dev.md"
+  "pipeline/commands/multi-agent/autopilot.md"
+  "pipeline/commands/multi-agent/local.md"
+  "pipeline/commands/multi-agent/local-autopilot.md"
+  "pipeline/commands/multi-agent/dev-autopilot.md"
+  "pipeline/commands/multi-agent/dev-local.md"
+  "pipeline/commands/multi-agent/dev-local-autopilot.md"
+)
+for rel in "${MODE_ENTRY_DOCS[@]}"; do
+  abs="$ROOT/$rel"
+  base="$(basename "$rel")"
+  if [ ! -f "$abs" ]; then
+    record_fail "$base missing from repo (inventory drift)"
+    continue
+  fi
+  if grep -qE 'TaskCreate ordering \(strict\)' "$abs" \
+     && grep -qE 'phase-number order' "$abs"; then
+    record_pass "$base declares the TaskCreate ordering rule"
+  else
+    record_fail "$base missing the TaskCreate ordering rule"
+  fi
+done
+
+# 4. Copilot full-inline orchestrator — Claude Code is single-CLI but the
+#    Copilot SKILL.md mirror must declare the same rule so the cross-CLI
+#    contract holds and any future Copilot-side TaskList equivalent inherits it.
+COPILOT_SKILL="$ROOT/pipeline/skills/shared/core/multi-agent/SKILL.md"
+if grep -qE 'TaskCreate ordering \(strict\)' "$COPILOT_SKILL" \
+   && grep -qE 'phase-number order' "$COPILOT_SKILL"; then
+  record_pass "Copilot multi-agent SKILL.md declares the TaskCreate ordering rule"
+else
+  record_fail "Copilot multi-agent SKILL.md missing the TaskCreate ordering rule"
+fi
+
+# 5. Negative check — no doc should still carry the old "pre-mark skipped
+#    phases as completed" wording without the new ordering reinforcement.
+LEGACY_ANTI_PATTERN='register the tile at startup, then mark it completed with .activeForm: "\[SKIPPED\]"'
+LEGACY_HITS=$(grep -rlE "$LEGACY_ANTI_PATTERN" "$ROOT/pipeline/commands/multi-agent" 2>/dev/null || true)
+if [ -z "$LEGACY_HITS" ]; then
+  record_pass "no legacy 'pre-mark skipped' wording lingering without the new rule"
+else
+  for h in $LEGACY_HITS; do
+    if grep -qE 'TaskCreate ordering \(strict\)' "$h"; then
+      record_pass "$(basename "$h") carries legacy pattern but ALSO the ordering rule"
+    else
+      record_fail "$(basename "$h") still carries legacy 'pre-mark skipped' wording without the ordering rule"
+    fi
+  done
+fi
+
+# Summary
+total=$((pass + fail))
+printf '\n→ smoke-tasklist-ordering: %d/%d passed\n' "$pass" "$total"
+if [ "$fail" -ne 0 ]; then
+  printf '\nFailures:\n'
+  for f in "${failures[@]}"; do printf '  - %s\n' "$f"; done
+  exit 1
+fi
+exit 0

package/pipeline/scripts/smoke-telemetry.sh

@@ -0,0 +1,147 @@
+#!/usr/bin/env bash
+# smoke-telemetry.sh — contract check for log-metric.sh + aggregate-metrics.mjs.
+#
+# Verifies:
+#   1. log-metric.sh appends valid JSONL (one line, parses, has required fields).
+#   2. Type coercion works (booleans, ints, strings).
+#   3. Concurrent appends don't interleave (10 parallel writers → 10 lines).
+#   4. aggregate-metrics.mjs handles missing file gracefully.
+#   5. aggregate-metrics.mjs computes accept_rate / reject_rate correctly.
+#   6. aggregate-metrics.mjs --json produces valid JSON.
+
+set -uo pipefail
+
+HERE="$(cd "$(dirname "$0")" && pwd)"
+ROOT="$(cd "$HERE/../.." && pwd)"
+TMP="$(mktemp -d -t multi-agent-smoke-XXXX)"
+trap 'rm -rf "$TMP"' EXIT
+
+PASS=0
+FAIL=0
+pass() { PASS=$((PASS+1)); echo "  ✓ $1"; }
+fail() { FAIL=$((FAIL+1)); echo "  ✗ $1"; }
+
+need() { command -v "$1" >/dev/null 2>&1 || { echo "error: $1 is required" >&2; exit 127; }; }
+need node
+need jq
+
+LOG="$ROOT/pipeline/scripts/log-metric.sh"
+AGG="$ROOT/pipeline/scripts/aggregate-metrics.mjs"
+export METRICS_FILE="$TMP/metrics.jsonl"
+
+echo "→ 1. log-metric writes a JSONL line with required fields"
+"$LOG" TASK-1 4 review.completed raw_count=8 accepted=3 deferred=2 rejected=3 approved=true
+LINE=$(tail -1 "$METRICS_FILE")
+if echo "$LINE" | jq -e '.ts and .task_id and .phase and .event and .details' >/dev/null; then
+  pass "JSONL line has all required fields"
+else
+  fail "missing required field in: $LINE"
+fi
+
+echo ""
+echo "→ 2. type coercion: bool, int, string"
+TYPES=$(echo "$LINE" | jq -r '.details | "approved=" + (.approved | type) + " accepted=" + (.accepted | type) + " raw_count=" + (.raw_count | type)')
+if [ "$TYPES" = "approved=boolean accepted=number raw_count=number" ]; then
+  pass "bool/int coerce correctly: $TYPES"
+else
+  fail "type coercion wrong: $TYPES"
+fi
+"$LOG" TASK-1 4 triage.edge_case case=high-rejection-rate
+STR_TYPE=$(tail -1 "$METRICS_FILE" | jq -r '.details.case | type')
+if [ "$STR_TYPE" = "string" ]; then
+  pass "non-numeric strings stay strings"
+else
+  fail "string coercion wrong: got $STR_TYPE"
+fi
+
+echo ""
+echo "→ 3. concurrent appends (10 parallel writers → 10 lines)"
+rm -f "$METRICS_FILE"
+PIDS=()
+for i in 1 2 3 4 5 6 7 8 9 10; do
+  "$LOG" TASK-$i 4 review.completed raw_count=$i accepted=0 deferred=0 rejected=0 approved=true &
+  PIDS+=($!)
+done
+FAILED=0
+for pid in "${PIDS[@]}"; do
+  wait "$pid" || FAILED=$((FAILED+1))
+done
+if [ "$FAILED" -gt 0 ]; then
+  fail "$FAILED of 10 background log-metric calls failed"
+fi
+LINE_COUNT=$(wc -l < "$METRICS_FILE" | tr -d ' ')
+if [ "$LINE_COUNT" = "10" ]; then
+  pass "10 concurrent writes produced 10 lines (no interleaving)"
+else
+  fail "expected 10 lines, got $LINE_COUNT"
+fi
+PARSE_OK=$(awk '{ print }' "$METRICS_FILE" | jq -e '.task_id' >/dev/null && echo ok || echo fail)
+if [ "$PARSE_OK" = "ok" ]; then
+  pass "all 10 lines parse as valid JSON"
+else
+  fail "concurrent append corrupted JSON"
+fi
+
+echo ""
+echo "→ 4. aggregator handles missing file gracefully"
+EXIT_OK=0
+node "$AGG" --json >/dev/null 2>&1 || EXIT_OK=$?
+# Above call uses default METRICS_FILE which IS set, so let's force missing.
+unset METRICS_FILE
+EXIT_MISSING=0
+METRICS_FILE="$TMP/does-not-exist.jsonl" node "$AGG" --json > "$TMP/missing-out.json" 2>&1 || EXIT_MISSING=$?
+if [ "$EXIT_MISSING" = "0" ] && jq -e '.ok == false' "$TMP/missing-out.json" >/dev/null 2>&1; then
+  pass "missing file → exit 0 + ok:false"
+else
+  fail "missing file handling wrong (exit=$EXIT_MISSING)"
+fi
+export METRICS_FILE="$TMP/metrics.jsonl"
+
+echo ""
+echo "→ 5. accept_rate and reject_rate computed correctly"
+rm -f "$METRICS_FILE"
+"$LOG" T 4 review.completed raw_count=10 accepted=4 deferred=2 rejected=4 approved=true
+"$LOG" T 4 review.completed raw_count=10 accepted=2 deferred=4 rejected=4 approved=true
+# Total: raw=20, accepted=6, deferred=6, rejected=8 → accept=0.30, reject=0.40
+SUMMARY=$(node "$AGG" --json)
+ACCEPT_RATE=$(echo "$SUMMARY" | jq -r '.triage_totals.accept_rate')
+REJECT_RATE=$(echo "$SUMMARY" | jq -r '.triage_totals.reject_rate')
+if [ "$ACCEPT_RATE" = "0.3" ] && [ "$REJECT_RATE" = "0.4" ]; then
+  pass "accept_rate=$ACCEPT_RATE reject_rate=$REJECT_RATE (expected 0.3 / 0.4)"
+else
+  fail "rates wrong: accept=$ACCEPT_RATE reject=$REJECT_RATE"
+fi
+
+echo ""
+echo "→ 6. --json output is valid JSON"
+if echo "$SUMMARY" | jq empty >/dev/null 2>&1; then
+  pass "--json output parses"
+else
+  fail "--json output is not valid JSON"
+fi
+
+echo ""
+echo "→ 7. cost_per_model section aggregates per-model totals"
+rm -f "$METRICS_FILE"
+"$LOG" T 4 review.completed model=opus duration_ms=12000 tokens_in=4500 tokens_out=600
+"$LOG" T 4 review.completed model=opus duration_ms=8000  tokens_in=3000 tokens_out=400
+"$LOG" T 4 review.completed model=sonnet duration_ms=5000 tokens_in=4500 tokens_out=350
+SUMMARY=$(node "$AGG" --json)
+OPUS_CALLS=$(echo "$SUMMARY" | jq -r '.cost_per_model.opus.calls')
+OPUS_TOKENS_IN=$(echo "$SUMMARY" | jq -r '.cost_per_model.opus.tokens_in')
+SONNET_DURATION=$(echo "$SUMMARY" | jq -r '.cost_per_model.sonnet.duration_ms')
+[ "$OPUS_CALLS" = "2" ] && pass "opus calls summed (=2)" || fail "opus calls expected 2, got $OPUS_CALLS"
+[ "$OPUS_TOKENS_IN" = "7500" ] && pass "opus tokens_in summed (=7500)" || fail "opus tokens_in expected 7500, got $OPUS_TOKENS_IN"
+[ "$SONNET_DURATION" = "5000" ] && pass "sonnet duration_ms summed (=5000)" || fail "sonnet duration_ms expected 5000, got $SONNET_DURATION"
+
+echo ""
+echo "→ 8. cost section absent when no model fields present"
+rm -f "$METRICS_FILE"
+"$LOG" T 0 init.completed phases=8
+SUMMARY=$(node "$AGG" --json)
+COST_KEYS=$(echo "$SUMMARY" | jq -r '.cost_per_model | keys | length')
+[ "$COST_KEYS" = "0" ] && pass "cost_per_model is empty when no model fields emitted" || fail "expected 0 keys, got $COST_KEYS"
+
+echo ""
+echo "══ telemetry smoke: $PASS passed, $FAIL failed ══"
+[ "$FAIL" -eq 0 ]

package/pipeline/scripts/smoke-test-gap.sh

@@ -0,0 +1,183 @@
+#!/usr/bin/env bash
+# smoke-test-gap.sh — v8.3.0
+#
+# Verifies the Phase 5 test-gap pipeline:
+#   1. Rules files load + parse for ios/android/python/node
+#   2. iOS fixture: scanner finds save/load/fetch as gaps; testRenders
+#      isn't a gap because it's the test, not source
+#   3. Android fixture: composable_fun symbol gets `important` severity
+#   4. Python fixture: detects async_def + top_def, ignores test files
+#   5. Node fixture: catches named_export Parser; tokenize is covered by
+#      paired test → no gap
+#   6. validate-test-gap.mjs accepts valid output, rejects malformed
+#   7. --severity-promote bumps suggestion → important
+#   8. Empty / wrong-stack diff exits 2
+#   9. phase-5-test.md ref doc declares the Test Gap Report step
+#  10. prefs.schema.json exposes testGap settings group
+#
+# Exit 0 = all pass, 1 = any failure.
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+SCAN="$ROOT/pipeline/scripts/test-gap-scan.mjs"
+VAL="$ROOT/pipeline/scripts/validate-test-gap.mjs"
+PHASE5="$ROOT/pipeline/commands/multi-agent/refs/phases/phase-5-test.md"
+PREFS="$ROOT/pipeline/schemas/prefs.schema.json"
+RULES_DIR="$ROOT/pipeline/scripts/test-gap-rules"
+FIX_IOS="$ROOT/pipeline/scripts/fixtures/diff-risk-ios.diff"
+FIX_AND="$ROOT/pipeline/scripts/fixtures/diff-risk-android.diff"
+FIX_PY="$ROOT/pipeline/scripts/fixtures/test-gap-python.diff"
+FIX_NODE="$ROOT/pipeline/scripts/fixtures/test-gap-node.diff"
+
+pass=0
+fail=0
+failures=()
+record_pass() { pass=$((pass + 1)); printf '  \033[0;32mPASS\033[0m %s\n' "$1"; }
+record_fail() { fail=$((fail + 1)); failures+=("$1"); printf '  \033[0;31mFAIL\033[0m %s\n' "$1"; }
+
+printf '→ smoke-test-gap (v8.3.0): Phase 5 test-gap detector contract\n'
+
+# --- 1: rules files exist + parse for each stack ---
+for stack in ios android python node; do
+  if jq -e '.stack and .sourceExtensions and .publicApiPatterns' "$RULES_DIR/$stack.json" >/dev/null 2>&1; then
+    record_pass "$stack rules parse and contain stack/extensions/patterns"
+  else
+    record_fail "$stack rules malformed"
+  fi
+done
+
+# --- 2: iOS fixture — KeychainStore.save/load + ApiClient.fetch detected ---
+out_ios=$(node "$SCAN" --diff "$FIX_IOS" --stack ios 2>/dev/null)
+gap_count=$(jq '.totals.gapCount' <<< "$out_ios")
+if [ "$gap_count" -ge 3 ]; then
+  record_pass "iOS: at least 3 gaps reported (got $gap_count)"
+else
+  record_fail "iOS: expected >=3 gaps, got $gap_count"
+fi
+if jq -e '.gaps[] | select(.symbol == "save")' <<< "$out_ios" >/dev/null 2>&1; then
+  record_pass "iOS: KeychainStore.save flagged as gap"
+else
+  record_fail "iOS: KeychainStore.save missing from gaps"
+fi
+if jq -e '.gaps[] | select(.symbol == "fetch")' <<< "$out_ios" >/dev/null 2>&1; then
+  record_pass "iOS: ApiClient.fetch flagged as gap"
+else
+  record_fail "iOS: ApiClient.fetch missing from gaps"
+fi
+# Test method itself shouldn't show up as a gap
+if jq -e '.gaps[] | select(.symbol == "testRenders")' <<< "$out_ios" >/dev/null 2>&1; then
+  record_fail "iOS: testRenders should not be a gap (it's already a test)"
+else
+  record_pass "iOS: test methods excluded from gap list"
+fi
+
+# --- 3: Android fixture — composable_fun → important severity ---
+out_and=$(node "$SCAN" --diff "$FIX_AND" --stack android 2>/dev/null)
+if jq -e '.gaps[] | select(.kind == "composable_fun" and .severity == "important")' <<< "$out_and" >/dev/null 2>&1; then
+  record_pass "Android: composable_fun gets 'important' severity"
+else
+  record_fail "Android: composable_fun severity wrong"
+fi
+if jq -e '.gaps[] | select(.symbol == "login")' <<< "$out_and" >/dev/null 2>&1; then
+  record_pass "Android: AuthRepository.login flagged"
+else
+  record_fail "Android: AuthRepository.login missing"
+fi
+
+# --- 4: Python fixture — async_def + top_def detected, test file ignored ---
+out_py=$(node "$SCAN" --diff "$FIX_PY" --stack python 2>/dev/null)
+if jq -e '.gaps[] | select(.kind == "async_def" and .symbol == "refund")' <<< "$out_py" >/dev/null 2>&1; then
+  record_pass "Python: async_def refund detected"
+else
+  record_fail "Python: async_def refund missing"
+fi
+if jq -e '.gaps[] | select(.symbol == "normalize_currency")' <<< "$out_py" >/dev/null 2>&1; then
+  record_pass "Python: normalize_currency detected"
+else
+  record_fail "Python: normalize_currency missing"
+fi
+if jq -e '.totals.testsScanned >= 1' <<< "$out_py" >/dev/null 2>&1; then
+  record_pass "Python: test file recognised in counter"
+else
+  record_fail "Python: testsScanned should include the new test_processor.py"
+fi
+
+# --- 5: Node fixture — Parser flagged, tokenize covered by paired test ---
+out_node=$(node "$SCAN" --diff "$FIX_NODE" --stack node 2>/dev/null)
+if jq -e '.gaps[] | select(.symbol == "Parser")' <<< "$out_node" >/dev/null 2>&1; then
+  record_pass "Node: Parser export flagged"
+else
+  record_fail "Node: Parser export missing from gaps"
+fi
+if jq -e '.gaps[] | select(.symbol == "tokenize")' <<< "$out_node" >/dev/null 2>&1; then
+  record_fail "Node: tokenize should be covered by added test, not flagged"
+else
+  record_pass "Node: tokenize correctly excluded (paired test mentions it)"
+fi
+
+# --- 6: Validator accepts/rejects ---
+set +e
+echo "$out_ios" | node "$VAL" - >/dev/null 2>&1
+rc_ok=$?
+echo '{"schemaVersion":"0.0.0","gaps":[]}' | node "$VAL" - >/dev/null 2>&1
+rc_bad=$?
+set -e
+if [ "$rc_ok" -eq 0 ]; then record_pass "validator accepts valid output"; else record_fail "validator rejected valid output"; fi
+if [ "$rc_bad" -ne 0 ]; then record_pass "validator rejects malformed"; else record_fail "validator should reject malformed"; fi
+
+# --- 7: --severity-promote ---
+out_prom=$(node "$SCAN" --diff "$FIX_IOS" --stack ios --severity-promote 2>/dev/null)
+if jq -e 'all(.gaps[]; .severity == "important")' <<< "$out_prom" >/dev/null 2>&1; then
+  record_pass "--severity-promote forces all gaps to 'important'"
+else
+  record_fail "--severity-promote did not promote severities"
+fi
+
+# --- 8: Empty diff exits 2 ---
+empty=$(mktemp)
+set +e
+node "$SCAN" --diff "$empty" --stack ios >/dev/null 2>&1
+rc=$?
+set -e
+rm -f "$empty"
+if [ "$rc" -eq 2 ]; then
+  record_pass "empty diff exits 2"
+else
+  record_fail "empty diff should exit 2 (got $rc)"
+fi
+
+# Wrong-stack: iOS diff scanned with python rules → no python sources → exit 2
+set +e
+node "$SCAN" --diff "$FIX_IOS" --stack python >/dev/null 2>&1
+rc=$?
+set -e
+if [ "$rc" -eq 2 ]; then
+  record_pass "iOS diff with python rules exits 2 (no matching sources)"
+else
+  record_fail "wrong-stack should exit 2 (got $rc)"
+fi
+
+# --- 9: phase-5-test.md declares the Test Gap Report step ---
+if grep -qE 'Test Gap Report|test-gap-scan\.mjs' "$PHASE5"; then
+  record_pass "phase-5-test.md declares Test Gap Report step"
+else
+  record_fail "phase-5-test.md missing Test Gap Report step"
+fi
+
+# --- 10: prefs schema exposes testGap group ---
+if jq -e '.properties.global.properties.testGap' "$PREFS" >/dev/null 2>&1; then
+  record_pass "prefs schema exposes testGap settings group"
+else
+  record_fail "prefs.schema.json missing global.testGap group"
+fi
+
+# --- Summary ---
+total=$((pass + fail))
+printf '\n→ smoke-test-gap: %d/%d passed\n' "$pass" "$total"
+if [ "$fail" -ne 0 ]; then
+  printf '\nFailures:\n'
+  for f in "${failures[@]}"; do printf '  - %s\n' "$f"; done
+  exit 1
+fi
+exit 0

package/pipeline/scripts/smoke-token-budget.sh

@@ -0,0 +1,67 @@
+#!/usr/bin/env bash
+# smoke-token-budget.sh — verify phase docs stay within token budget.
+# Reads token-budget.json and checks each phase doc's char count / 4.
+# Fails if any phase exceeds its max_tokens limit.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+PIPELINE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
+BUDGET_FILE="$PIPELINE_DIR/schemas/token-budget.json"
+PHASES_DIR="$PIPELINE_DIR/commands/multi-agent/refs/phases"
+
+if [ ! -f "$BUDGET_FILE" ]; then
+  echo "FAIL: token-budget.json not found at $BUDGET_FILE" >&2
+  exit 1
+fi
+
+command -v jq >/dev/null 2>&1 || { echo "jq is required" >&2; exit 64; }
+
+PASS=0
+FAIL=0
+WARN=0
+
+pass() { PASS=$((PASS+1)); echo "  ✓ $1"; }
+warn() { WARN=$((WARN+1)); echo "  ⚠ $1"; }
+fail() { FAIL=$((FAIL+1)); echo "  ✗ $1"; }
+
+TOTAL_TOKENS=0
+TOTAL_MAX=$(jq -r '.total_max_tokens' "$BUDGET_FILE")
+
+echo "→ Token budget check (chars/4 estimate)"
+echo ""
+
+for phase_key in $(jq -r '.phases | keys[]' "$BUDGET_FILE"); do
+  file="$PHASES_DIR/${phase_key}.md"
+  max_tokens=$(jq -r ".phases[\"$phase_key\"].max_tokens" "$BUDGET_FILE")
+  warn_tokens=$(jq -r ".phases[\"$phase_key\"].warn_tokens" "$BUDGET_FILE")
+
+  if [ ! -f "$file" ]; then
+    fail "$phase_key: file not found ($file)"
+    continue
+  fi
+
+  chars=$(wc -c < "$file" | tr -d ' ')
+  tokens=$(( (chars + 3) / 4 ))
+  TOTAL_TOKENS=$((TOTAL_TOKENS + tokens))
+
+  if [ "$tokens" -gt "$max_tokens" ]; then
+    fail "$phase_key: ${tokens} tokens > max ${max_tokens} (${chars} chars)"
+  elif [ "$tokens" -gt "$warn_tokens" ]; then
+    warn "$phase_key: ${tokens} tokens > warn ${warn_tokens} (within max ${max_tokens})"
+  else
+    pass "$phase_key: ${tokens} tokens ≤ ${warn_tokens} warn (max ${max_tokens})"
+  fi
+done
+
+echo ""
+echo "→ Total phase doc budget"
+if [ "$TOTAL_TOKENS" -gt "$TOTAL_MAX" ]; then
+  fail "total: ${TOTAL_TOKENS} tokens > max ${TOTAL_MAX}"
+else
+  pass "total: ${TOTAL_TOKENS} tokens ≤ max ${TOTAL_MAX}"
+fi
+
+echo ""
+echo "══ token-budget smoke: $((PASS)) passed, $((WARN)) warnings, $((FAIL)) failed ══"
+[ "$FAIL" -eq 0 ]

package/pipeline/scripts/smoke-tracker-contract.sh

@@ -0,0 +1,129 @@
+#!/usr/bin/env bash
+# smoke-tracker-contract.sh — v7.9.1
+#
+# Validates that every mode dispatch file (dev.md, dev-local.md, autopilot.md,
+# etc.) carries the mandatory phase-tracker.sh invocation contract.
+#
+# Without this, agent prompts can fall through to TaskCreate/TaskList UI which
+# does NOT track elapsed time or token counts — users see only static
+# checkboxes with no progress info. The 2026-04-27 regression that prompted
+# this smoke: 8 mode dispatch files shipped through v7.6 → v7.9 with zero
+# tracker references; agent went straight to Claude Code's plan UI.
+#
+# What this enforces:
+#   1. Each mode file references phase-tracker.sh at least N times (init +
+#      add + update + tokens + render at minimum)
+#   2. Each mode file links to refs/tracker-contract.md
+#   3. The shared contract file exists and documents the invocation pattern
+#
+# What this does NOT enforce:
+#   - Runtime — that the agent actually invokes the tracker. That's a contract
+#     test the orchestrator carries; this smoke catches the prompt-side gap.
+
+set -uo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+CMDS_DIR="$REPO_ROOT/pipeline/commands/multi-agent"
+CONTRACT_FILE="$CMDS_DIR/refs/tracker-contract.md"
+
+# Mode dispatch files that must reference the tracker contract.
+# Updated when new mode files ship; no auto-discovery (false positives on
+# helpers like _account-picker, status, log).
+MODE_FILES=(
+  dev.md
+  dev-local.md
+  dev-autopilot.md
+  dev-local-autopilot.md
+  autopilot.md
+  local.md
+  local-autopilot.md
+)
+
+# Minimum tracker references per file (init + at least one of the verbs).
+# Set conservatively — the section template emits 6 references; bump if
+# template is shortened in future.
+MIN_REFS=4
+
+PASS=0
+FAIL=0
+pass() { PASS=$((PASS + 1)); echo "  ✓ $1"; }
+fail() { FAIL=$((FAIL + 1)); echo "  ✗ $1"; }
+
+# ──────────────────────────────────────────────────────────────────────────
+echo "→ 1. Tracker contract reference file exists"
+if [ -f "$CONTRACT_FILE" ]; then
+  pass "refs/tracker-contract.md present"
+else
+  fail "refs/tracker-contract.md missing — mode files have nowhere to link to"
+fi
+
+# ──────────────────────────────────────────────────────────────────────────
+echo "→ 2. Contract file documents the full invocation pattern"
+if [ -f "$CONTRACT_FILE" ]; then
+  for verb in "init" "add" "update" "tokens" "render"; do
+    if grep -q "phase-tracker.sh $verb" "$CONTRACT_FILE"; then
+      pass "contract documents '$verb' verb"
+    else
+      fail "contract missing '$verb' verb in invocation pattern"
+    fi
+  done
+fi
+
+# ──────────────────────────────────────────────────────────────────────────
+echo "→ 3. Every mode file references phase-tracker.sh"
+for mode_file in "${MODE_FILES[@]}"; do
+  full="$CMDS_DIR/$mode_file"
+  if [ ! -f "$full" ]; then
+    fail "$mode_file missing"
+    continue
+  fi
+  refs=$(grep -c "phase-tracker" "$full" || true)
+  if [ "$refs" -ge "$MIN_REFS" ]; then
+    pass "$mode_file: $refs tracker references (≥ $MIN_REFS)"
+  else
+    fail "$mode_file: only $refs tracker references (< $MIN_REFS expected)"
+  fi
+done
+
+# ──────────────────────────────────────────────────────────────────────────
+echo "→ 4. Every mode file links to refs/tracker-contract.md"
+for mode_file in "${MODE_FILES[@]}"; do
+  full="$CMDS_DIR/$mode_file"
+  [ -f "$full" ] || continue
+  if grep -q "tracker-contract" "$full"; then
+    pass "$mode_file → links to tracker-contract.md"
+  else
+    fail "$mode_file missing link to refs/tracker-contract.md"
+  fi
+done
+
+# ──────────────────────────────────────────────────────────────────────────
+echo "→ 5. Anti-pattern guard — mode files warn against TaskCreate fallback"
+for mode_file in "${MODE_FILES[@]}"; do
+  full="$CMDS_DIR/$mode_file"
+  [ -f "$full" ] || continue
+  if grep -qE "TaskCreate|checkbox UI" "$full"; then
+    pass "$mode_file mentions anti-pattern (TaskCreate / checkbox UI)"
+  else
+    fail "$mode_file has no anti-pattern guard"
+  fi
+done
+
+# ──────────────────────────────────────────────────────────────────────────
+echo "→ 6. Phase 0 init pattern present in dispatch templates"
+# This catches the specific bug from 2026-04-27: tracker was never init'd
+# because dispatch files said "see refs" without inlining the call.
+for mode_file in "${MODE_FILES[@]}"; do
+  full="$CMDS_DIR/$mode_file"
+  [ -f "$full" ] || continue
+  if grep -q 'phase-tracker.sh init' "$full"; then
+    pass "$mode_file inlines 'phase-tracker.sh init' (refs-only is not enough)"
+  else
+    fail "$mode_file does not inline init call"
+  fi
+done
+
+# ──────────────────────────────────────────────────────────────────────────
+echo ""
+echo "══ tracker-contract smoke: $PASS passed, $FAIL failed ══"
+[ "$FAIL" -eq 0 ]