diff --git a/scripts/dev-lead-fix-ci.sh b/scripts/dev-lead-fix-ci.sh
index 090338620..2331c7e5d 100755
--- a/scripts/dev-lead-fix-ci.sh
+++ b/scripts/dev-lead-fix-ci.sh
@@ -9,6 +9,7 @@ set -euo pipefail
 
 source "$(dirname "$0")/engine.sh"
 
+INTENT_TYPE="fix-ci"
 PR_NUMBER="${PR_NUMBER:-}"
 HEAD_SHA="${HEAD_SHA:-}"
 CHECKS_JSON="${CHECKS_JSON:-[]}"
@@ -202,7 +203,7 @@ main() {
     echo "  [fix-ci] cycle $cycle/$MAX_CI_CYCLES"
 
     local engine_rc=0
-    run_writer_with_fallback "$prompt_file" || engine_rc=$?
+    run_writer_with_fallback "$prompt_file" "$(model_for_intent "$INTENT_TYPE")" || engine_rc=$?
 
     if [ "$engine_rc" -ne 0 ]; then
       if [ "$engine_rc" -eq 2 ]; then
diff --git a/scripts/dev-lead-fix-issue.sh b/scripts/dev-lead-fix-issue.sh
index f3e89572d..09dc15e12 100755
--- a/scripts/dev-lead-fix-issue.sh
+++ b/scripts/dev-lead-fix-issue.sh
@@ -5,6 +5,7 @@ set -euo pipefail
 
 source "$(dirname "$0")/engine.sh"
 
+INTENT_TYPE="fix-issue"
 ISSUE_NUMBER="${ISSUE_NUMBER:-}"
 REPO="${REPO:-${GITHUB_REPOSITORY:-}}"
 DEV_LEAD_DRY_RUN="${DEV_LEAD_DRY_RUN:-false}"
@@ -54,7 +55,7 @@ main() {
   local pre_engine_sha
   pre_engine_sha=$(git rev-parse HEAD)
 
-  if ! run_writer_with_fallback "$prompt_file"; then
+  if ! run_writer_with_fallback "$prompt_file" "$(model_for_intent "$INTENT_TYPE")"; then
     echo "::error::Engine failed to implement issue #${ISSUE_NUMBER}"
     rm -f "$prompt_file"
     exit 1
diff --git a/scripts/dev-lead-fix-reviews.sh b/scripts/dev-lead-fix-reviews.sh
index d09465451..b92d80964 100755
--- a/scripts/dev-lead-fix-reviews.sh
+++ b/scripts/dev-lead-fix-reviews.sh
@@ -33,6 +33,7 @@ fi
 
 build_and_run() {
   local template_name="$1"
+  local model="${2:-$ENGINE_ACTION_MODEL}"
   local prompt_file="/tmp/dev-lead-${template_name}-prompt-$$.md"
   # Export required vars then envsubst
   envsubst < "${PROMPTS_DIR}/${template_name}.md" > "$prompt_file"
@@ -44,7 +45,7 @@ build_and_run() {
   fi
 
   local rc=0
-  run_writer_with_fallback "$prompt_file" || rc=$?
+  run_writer_with_fallback "$prompt_file" "$model" || rc=$?
   rm -f "$prompt_file"
   return "$rc"
 }
@@ -257,7 +258,7 @@ case "$INTENT_TYPE" in
       --jq '.data.repository.pullRequest.reviewThreads.nodes | map(select(.isResolved == false))' 2>/dev/null || echo "[]")
     export OPEN_THREADS_JSON
     rc=0
-    build_and_run "fix-reviews" || rc=$?
+    build_and_run "fix-reviews" "$(model_for_intent "fix-reviews")" || rc=$?
     [ "$rc" -eq 2 ] && handle_rate_limit "fix-reviews"
     if [ "$rc" -eq 0 ]; then
       if commit_and_push "fix-reviews"; then
@@ -272,7 +273,7 @@ case "$INTENT_TYPE" in
     export PR_NUMBER PR_URL="https://github.com/${REPO}/pull/${PR_NUMBER}"
     export REPO ACTOR="${ACTOR:-}" COMMENT_BODY="${COMMENT_BODY:-}" HEAD_SHA
     rc=0
-    build_and_run "fix-bot-comment" || rc=$?
+    build_and_run "fix-bot-comment" "$(model_for_intent "fix-bot-comment")" || rc=$?
     [ "$rc" -eq 2 ] && handle_rate_limit "fix-bot-comment"
     if [ "$rc" -eq 0 ]; then
       if commit_and_push "fix-bot-comment"; then
@@ -288,7 +289,7 @@ case "$INTENT_TYPE" in
     export PR_URL="https://github.com/${REPO}/pull/${PR_NUMBER}"
     export REPO ACTOR="${ACTOR:-}" USER_INSTRUCTION="${USER_INSTRUCTION:-}" PR_DESCRIPTION="${PR_DESCRIPTION:-}"
     rc=0
-    build_and_run "human" || rc=$?
+    build_and_run "human" "$(model_for_intent "human")" || rc=$?
     [ "$rc" -eq 2 ] && handle_rate_limit "human"
     if [ "$rc" -eq 0 ]; then
       if commit_and_push "human"; then
@@ -317,7 +318,7 @@ case "$INTENT_TYPE" in
       --jq '.data.repository.pullRequest.reviewThreads.nodes | map(select(.isResolved == false))' 2>/dev/null || echo "[]")
     export OPEN_THREADS_JSON BASE_REF="${BASE_REF:-main}"
     rc=0
-    build_and_run "human-pr" || rc=$?
+    build_and_run "human-pr" "$(model_for_intent "human-pr")" || rc=$?
     [ "$rc" -eq 2 ] && handle_rate_limit "human-pr"
     if [ "$rc" -eq 0 ]; then
       if commit_and_push "human-pr"; then
@@ -344,7 +345,7 @@ case "$INTENT_TYPE" in
     CONFLICTING_FILES=$(git merge-tree "$(git merge-base HEAD "origin/${BASE_REF}")" HEAD "origin/${BASE_REF}" 2>/dev/null | grep "^changed in both" | awk '{print $NF}' || true)
     export CONFLICTING_FILES
     rc=0
-    build_and_run "rebase" || rc=$?
+    build_and_run "rebase" "$(model_for_intent "rebase")" || rc=$?
     [ "$rc" -eq 2 ] && handle_rate_limit "rebase"
     if [ "$rc" -eq 0 ]; then
       if commit_and_push "rebase"; then
diff --git a/scripts/engine.sh b/scripts/engine.sh
index df0155eb2..8a2c60e3d 100644
--- a/scripts/engine.sh
+++ b/scripts/engine.sh
@@ -4,11 +4,16 @@ set -euo pipefail
 # Supports: claude, gemini, copilot
 #
 # Sourced by review-one-pr.sh — provides:
-#   run_triage <prompt_file>       — no-tool tier (stdout capture)
-#   run_agentic <prompt_file> <model>  — full-tool tier (stdout)
-#   run_duck <prompt_file> <model>     — cross-engine adversarial (stdout)
+#   run_triage <prompt_file>           — no-tool tier (review-one-pr.sh only)
+#   run_agentic <prompt_file> <model>  — full-tool tier (review-one-pr.sh only)
+#   run_duck <prompt_file> <model>     — adversarial cross-engine (review-one-pr.sh only)
 #   ENGINE_* env vars for model names and labels
 #   DUCK_ENGINE / DUCK_MODEL for rubber-duck cross-engine review
+#
+# Sourced by dev-lead scripts — provides:
+#   model_for_intent <intent>          — select model tier by intent complexity
+#   run_writer <prompt_file> [model]   — write-capable agent run
+#   run_writer_with_fallback <prompt_file> [model]  — run with engine fallback
 
 REVIEW_ENGINE="${REVIEW_ENGINE:-claude}"
 export REVIEW_ENGINE
@@ -29,8 +34,13 @@ DUCK_TIMEOUT_SEC="${DUCK_TIMEOUT_SEC:-300}"
 RETRY_MAX_ATTEMPTS="${RETRY_MAX_ATTEMPTS:-2}"   # total attempts including first
 RETRY_BASE_DELAY_SEC="${RETRY_BASE_DELAY_SEC:-5}"
 
-set_engine_config() {
-  case "$REVIEW_ENGINE" in
+# _setup_engine_vars <engine>
+# Sets all ENGINE_* model vars for the given engine name.
+# Called once at source time and again by run_writer_with_fallback when
+# switching engines mid-fallback, so each invocation gets compatible model IDs.
+_setup_engine_vars() {
+  local eng="${1:-$REVIEW_ENGINE}"
+  case "$eng" in
     claude)
       ENGINE_TRIAGE_MODEL="claude-haiku-4-5-20251001"
       ENGINE_DEEP_MODEL="claude-sonnet-4-6"
@@ -44,13 +54,13 @@ set_engine_config() {
       DUCK_MODEL="o4-mini"
       ;;
     gemini)
-      ENGINE_TRIAGE_MODEL="auto"
-      ENGINE_DEEP_MODEL="auto"
-      ENGINE_AUDIT_MODEL="auto"
-      ENGINE_ACTION_MODEL="auto"
-      ENGINE_SINGLE_MODEL="auto"
-      ENGINE_LABEL="auto (Gemini 2.5/3 balanced)"
-      ENGINE_SINGLE_LABEL="single-reviewer mode: auto"
+      ENGINE_TRIAGE_MODEL="gemini-2.0-flash"
+      ENGINE_DEEP_MODEL="gemini-1.5-pro"
+      ENGINE_AUDIT_MODEL="gemini-1.5-pro"
+      ENGINE_ACTION_MODEL="gemini-1.5-pro"
+      ENGINE_SINGLE_MODEL="gemini-1.5-pro"
+      ENGINE_LABEL="triage: gemini-2.0-flash → deep: gemini-1.5-pro + duck: sonnet 4.6 → audit: gemini-1.5-pro"
+      ENGINE_SINGLE_LABEL="single-reviewer mode: gemini-1.5-pro"
       # Cross-engine rubber duck: use Claude for diversity
       DUCK_ENGINE="claude"
       DUCK_MODEL="claude-sonnet-4-6"
@@ -75,19 +85,17 @@ set_engine_config() {
       DUCK_MODEL="claude-sonnet-4-6"
       ;;
     *)
-      echo "::error::Unknown REVIEW_ENGINE='$REVIEW_ENGINE' (expected: claude, gemini, or copilot)"
-      exit 1
+      echo "::error::Unknown REVIEW_ENGINE='$eng' (expected: claude, gemini, or copilot)"
+      return 1
       ;;
   esac
-
   export ENGINE_TRIAGE_MODEL ENGINE_DEEP_MODEL ENGINE_AUDIT_MODEL
   export ENGINE_ACTION_MODEL ENGINE_SINGLE_MODEL
   export ENGINE_LABEL ENGINE_SINGLE_LABEL
   export DUCK_ENGINE DUCK_MODEL
 }
 
-# Initial config
-set_engine_config
+_setup_engine_vars "$REVIEW_ENGINE" || exit 1
 echo "    engine: $REVIEW_ENGINE ($ENGINE_LABEL)"
 
 # is_rate_limited <text>
@@ -449,25 +457,52 @@ parse_reset_time() {
   printf '%s' "${iso:-}" > /tmp/dev-lead-rate-limit-reset
 }
 
-# run_writer <prompt_file> [model]
+# model_for_intent <intent>
+# Returns an engine-agnostic tier key: "triage", "action", or "deep".
+# run_writer resolves the key to the correct model ID for the active engine
+# via ENGINE_TRIAGE_MODEL / ENGINE_ACTION_MODEL / ENGINE_DEEP_MODEL, which are
+# re-initialized per engine by run_writer_with_fallback so fallback engines
+# always get a compatible model ID.
+#   triage — human-pr, fix-bot-comment: lightweight checks, no agentic writes needed
+#   action — fix-reviews, fix-ci, rebase: write operations on known-scope diffs
+#   deep   — fix-issue, human: full agentic feature work
+model_for_intent() {
+  case "${1:-}" in
+    human-pr|fix-bot-comment)   echo "triage" ;;
+    fix-reviews|fix-ci|rebase)  echo "action" ;;
+    fix-issue|human)            echo "deep"   ;;
+    *)                          echo "action" ;;
+  esac
+}
+
+# run_writer <prompt_file> [tier_or_model]
 # Full write-access mode for applying code fixes.
-# When DEV_LEAD_DRY_RUN=true: builds prompt but does NOT call engine; exits 0.
-# Exit codes: 0=success, 1=non-retriable failure, 2=rate-limited
+# tier_or_model: a tier key ("triage", "action", "deep") or a literal model ID.
+#   Tier keys are resolved to the current engine's model via ENGINE_*_MODEL vars,
+#   which run_writer_with_fallback keeps in sync when switching engines.
+# When DEV_LEAD_DRY_RUN=true: logs the prompt size but does NOT call engine; exits 0.
+# Exit codes: 0=success, 1=non-retriable failure, 2=rate-limited or unavailable
 # On exit 2, writes parsed reset timestamp to /tmp/dev-lead-rate-limit-reset.
 run_writer() {
   local prompt_file="$1"
-  local model="${2:-$ENGINE_ACTION_MODEL}"
+  local tier_or_model="${2:-action}"
+
+  # Resolve tier key → current engine's model ID.
+  local model
+  case "$tier_or_model" in
+    triage) model="$ENGINE_TRIAGE_MODEL" ;;
+    deep)   model="$ENGINE_DEEP_MODEL"   ;;
+    action) model="$ENGINE_ACTION_MODEL" ;;
+    *)      model="$tier_or_model"       ;;  # pass-through for literal model strings
+  esac
 
   if [ "${DEV_LEAD_DRY_RUN:-false}" = "true" ]; then
-    echo "  [dry-run] run_writer: would invoke $REVIEW_ENGINE with prompt $(wc -l < "$prompt_file") lines"
+    echo "  [dry-run] run_writer: would invoke $REVIEW_ENGINE ($model) with prompt $(wc -l < "$prompt_file") lines"
     return 0
   fi
 
   # Capture stdout to a temp file so is_rate_limited can inspect it, while
-  # still streaming output to the caller. The old approach read from
-  # /tmp/dev-lead-writer-stderr which was never written (claude --print outputs
-  # to stdout, not stderr), so is_rate_limited never fired and fallback engines
-  # were never tried.
+  # still streaming output to the caller.
   local _tmp rc=0
   _tmp=$(mktemp)
 
@@ -487,15 +522,13 @@ run_writer() {
         < "$prompt_file" | tee "$_tmp" || rc=${PIPESTATUS[0]}
       ;;
     copilot)
-      # Copilot (gh copilot suggest) is text-only — falls back to Claude for write ops
-      echo "::warning::Copilot engine is text-only; falling back to Claude for write operations" >&2
-      local saved="$REVIEW_ENGINE"
-      REVIEW_ENGINE="claude" timeout "$ACTION_TIMEOUT_SEC" claude --print \
-        --model "$model" \
-        --permission-mode acceptEdits \
-        --allowed-tools "Bash,Read,Write,Edit,Grep,Glob" \
-        < "$prompt_file" | tee "$_tmp" || rc=${PIPESTATUS[0]}
-      REVIEW_ENGINE="$saved"
+      # GitHub Models REST API is text-only — no Write/Edit tool access.
+      # Returning exit 2 signals "write engine unavailable" so run_writer_with_fallback
+      # exhausts the chain and the retry cron re-triggers when a write-capable
+      # engine (claude, gemini) is available again.
+      rm -f "$_tmp"
+      echo "::notice::copilot engine is text-only; write operations unavailable — triggering retry" >&2
+      return 2
       ;;
   esac
 
@@ -509,36 +542,43 @@ run_writer() {
   return "$rc"
 }
 
-# run_writer_with_fallback <prompt_file> [model]
-# Tries primary engine, falls back through claude → gemini → copilot on rate-limit.
-# Only rate-limit (exit 2) triggers fallback; other failures propagate immediately.
+# run_writer_with_fallback <prompt_file> [tier_or_model]
+# Tries primary engine, falls back through claude → gemini → copilot on rate-limit (exit 2).
+# Re-initializes ENGINE_* vars for each engine so model IDs are always compatible.
+# Exit 2 (rate-limited) and exit 127 (engine binary not installed) both trigger fallback.
+# Other non-zero exits propagate immediately as real failures.
 run_writer_with_fallback() {
   local prompt_file="$1"
+  local tier_or_model="${2:-action}"
   local engines=("$REVIEW_ENGINE")
 
   for e in claude gemini copilot; do
     [ "$e" != "$REVIEW_ENGINE" ] && engines+=("$e")
   done
 
+  local orig_engine="$REVIEW_ENGINE"
   for engine in "${engines[@]}"; do
-    local saved="$REVIEW_ENGINE"
+    # Re-initialize ENGINE_* vars for this engine so model tier resolution
+    # in run_writer returns a compatible model ID for the active provider.
     export REVIEW_ENGINE="$engine"
-    # Re-evaluate model names for the new engine
-    set_engine_config
+    _setup_engine_vars "$engine"
     local rc=0
-    # Don't pass 'model' argument; run_writer will use the updated $ENGINE_ACTION_MODEL
-    run_writer "$prompt_file" || rc=$?
-    export REVIEW_ENGINE="$saved"
-    # Restore original config for subsequent PRs in the same session
-    set_engine_config
-    [ "$rc" -eq 0 ] && return 0
-    if [ "$rc" -eq 2 ]; then
-      echo "::warning::$engine rate-limited, trying next engine" >&2
+    run_writer "$prompt_file" "$tier_or_model" || rc=$?
+    if [ "$rc" -eq 0 ]; then
+      export REVIEW_ENGINE="$orig_engine"; _setup_engine_vars "$orig_engine"
+      return 0
+    fi
+    if [ "$rc" -eq 2 ] || [ "$rc" -eq 127 ]; then
+      # exit 2: rate-limited or text-only engine unavailable for writes
+      # exit 127: engine binary not installed in this environment
+      echo "::warning::$engine unavailable (exit $rc), trying next engine" >&2
       continue
     fi
+    export REVIEW_ENGINE="$orig_engine"; _setup_engine_vars "$orig_engine"
     return "$rc"
   done
 
+  export REVIEW_ENGINE="$orig_engine"; _setup_engine_vars "$orig_engine"
   echo "::error::All engines rate-limited or unavailable" >&2
   return 2
 }
diff --git a/tests/dev-lead/unit/test_engine_fallback.bats b/tests/dev-lead/unit/test_engine_fallback.bats
index 440d950b7..b25221edd 100644
--- a/tests/dev-lead/unit/test_engine_fallback.bats
+++ b/tests/dev-lead/unit/test_engine_fallback.bats
@@ -92,16 +92,10 @@ STUBEOF
 }
 
 @test "fallback: all rate-limited → returns 2" {
-  # All engines exit 2
+  # claude exits 2 (rate-limited), gemini exits 2 (rate-limited),
+  # copilot returns 2 directly in run_writer (text-only, write ops unsupported).
   _make_stub "claude" 2
   _make_stub "gemini" 2
-  # copilot falls back to claude internally, so we only need claude and gemini
-  # Create a copilot stub that also fails with 2
-  cat > "$STUB_BIN_DIR/copilot" <<'STUBEOF'
-#!/usr/bin/env bash
-exit 2
-STUBEOF
-  chmod +x "$STUB_BIN_DIR/copilot"
   _source_engine "claude"
 
   run run_writer_with_fallback "$TEST_PROMPT"
@@ -122,6 +116,39 @@ STUBEOF
   [ "$status" -eq 1 ]
 }
 
+@test "fallback: engine not installed (exit 127) → tries next engine" {
+  # Simulates gemini not installed: exit 127 should trigger fallback, not abort
+  _make_stub "claude" 2
+  _make_stub "gemini" 127
+  _source_engine "claude"
+
+  run run_writer_with_fallback "$TEST_PROMPT"
+
+  # claude rate-limited (2), gemini not installed (127), copilot text-only (2) → all exhausted
+  [ "$status" -eq 2 ]
+}
+
+@test "fallback: engine not installed (127) → next engine succeeds" {
+  # claude rate-limited, gemini not installed, but the next engine after gemini succeeds
+  # In the actual fallback order: claude → gemini → copilot. Since copilot returns 2
+  # (text-only) in run_writer, we verify with a scenario where gemini is "not found"
+  # and a recording stub shows that claude was tried after gemini failed.
+  _make_stub "gemini" 2
+  local record_file
+  record_file="$(mktemp)"
+  _make_recording_stub "claude" 127 "$record_file"
+  _source_engine "gemini"
+
+  run run_writer_with_fallback "$TEST_PROMPT"
+
+  # gemini (primary) rate-limited, claude not installed (127 → treated as unavailable),
+  # copilot text-only (2) → all engines exhausted
+  [ "$status" -eq 2 ]
+  # claude was tried (even though it exited 127, it was reached)
+  [ -s "$record_file" ]
+  rm -f "$record_file"
+}
+
 @test "fallback: fallback engine order is claude → gemini → copilot" {
   # Primary = gemini (exits 2), then fallback order should try claude first
   _make_stub "gemini" 2
diff --git a/tests/dev-lead/unit/test_engine_writer.bats b/tests/dev-lead/unit/test_engine_writer.bats
index b749b9833..bf7ec3fe0 100644
--- a/tests/dev-lead/unit/test_engine_writer.bats
+++ b/tests/dev-lead/unit/test_engine_writer.bats
@@ -90,16 +90,18 @@ _source_engine() {
   [ "$status" -eq 0 ]
 }
 
-@test "writer: copilot falls back to claude in run_writer (internal)" {
-  # When REVIEW_ENGINE=copilot, run_writer internally falls back to claude
+@test "writer: copilot returns exit 2 for write operations (text-only engine)" {
+  # Copilot is text-only and cannot write files. run_writer returns exit 2 so
+  # run_writer_with_fallback exhausts the chain and the retry cron re-triggers.
+  # claude must NOT be called — the old fallback-to-claude bug must stay fixed.
+  export COPILOT_API_MODEL="openai/o4-mini"
   _source_engine "copilot"
-  export STUB_ENGINE_EXIT=0
   export DEV_LEAD_DRY_RUN=false
+  rm -f "$STUB_BIN_DIR/claude"
 
   run run_writer "$TEST_PROMPT"
 
-  # Should succeed because the internal claude stub is present
-  [ "$status" -eq 0 ]
+  [ "$status" -eq 2 ]
 }
 
 @test "writer: run_writer dry-run logs prompt line count" {
@@ -178,7 +180,7 @@ STUB
 @test "writer: run_writer_with_fallback retries all engines and returns 2 when all rate-limited" {
   _source_engine "claude"
   export DEV_LEAD_DRY_RUN=false
-  # All engines output rate-limit text and exit 1 → run_writer returns 2 for each
+  # claude and gemini stubs output rate-limit text and exit 1 → run_writer returns 2
   for engine in claude gemini; do
     cat > "$STUB_BIN_DIR/$engine" << 'STUB'
 #!/usr/bin/env bash
@@ -187,8 +189,7 @@ exit 1
 STUB
     chmod +x "$STUB_BIN_DIR/$engine"
   done
-  # Add copilot stub (falls back to claude internally)
-  cp "$STUB_BIN_DIR/claude" "$STUB_BIN_DIR/copilot"
+  # copilot returns exit 2 directly in run_writer (text-only, no stub needed)
 
   run run_writer_with_fallback "$TEST_PROMPT"
 
@@ -240,3 +241,64 @@ STUB
   result=$(cat /tmp/dev-lead-rate-limit-reset)
   [ -z "$result" ]
 }
+
+# ── model_for_intent tests ─────────────────────────────────────────────────────
+
+@test "model_for_intent: human-pr → triage tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "human-pr")" = "triage" ]
+}
+
+@test "model_for_intent: fix-bot-comment → triage tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "fix-bot-comment")" = "triage" ]
+}
+
+@test "model_for_intent: fix-reviews → action tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "fix-reviews")" = "action" ]
+}
+
+@test "model_for_intent: fix-ci → action tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "fix-ci")" = "action" ]
+}
+
+@test "model_for_intent: rebase → action tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "rebase")" = "action" ]
+}
+
+@test "model_for_intent: fix-issue → deep tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "fix-issue")" = "deep" ]
+}
+
+@test "model_for_intent: human → deep tier" {
+  _source_engine "claude"
+  [ "$(model_for_intent "human")" = "deep" ]
+}
+
+@test "model_for_intent: unknown intent → action tier (default)" {
+  _source_engine "claude"
+  [ "$(model_for_intent "unknown-intent")" = "action" ]
+}
+
+@test "model_for_intent: empty intent → action tier (default)" {
+  _source_engine "claude"
+  [ "$(model_for_intent "")" = "action" ]
+}
+
+@test "model_for_intent: tier resolves to correct model for claude engine" {
+  _source_engine "claude"
+  [ "$(model_for_intent "human-pr")" = "triage" ]
+  # run_writer resolves "triage" → ENGINE_TRIAGE_MODEL for the current engine
+  [ "$ENGINE_TRIAGE_MODEL" = "claude-haiku-4-5-20251001" ]
+}
+
+@test "model_for_intent: tier key is engine-agnostic (same key for gemini)" {
+  _source_engine "gemini"
+  [ "$(model_for_intent "human-pr")" = "triage" ]
+  # ENGINE_TRIAGE_MODEL is now gemini-2.0-flash for this engine
+  [ "$ENGINE_TRIAGE_MODEL" = "gemini-2.0-flash" ]
+}
diff --git a/tests/dev-lead/unit/test_fix_ci.bats b/tests/dev-lead/unit/test_fix_ci.bats
index 97295bcf6..2d605a077 100644
--- a/tests/dev-lead/unit/test_fix_ci.bats
+++ b/tests/dev-lead/unit/test_fix_ci.bats
@@ -19,6 +19,23 @@ setup() {
   export PATH="$STUB_BIN_DIR:$PATH"
   export STUB_BIN_DIR
 
+  # Stub curl so copilot_chat returns a 429 rate-limit response for GitHub Models API.
+  # This makes the copilot fallback behave as rate-limited without a real API call.
+  cat > "$STUB_BIN_DIR/curl" << 'CURLEOF'
+#!/usr/bin/env bash
+# Detect GitHub Models API calls and return a synthetic 429 response.
+for arg in "$@"; do
+  if [[ "$arg" == *"models.github.ai"* ]]; then
+    echo '{"error":{"message":"rate limit exceeded","type":"rate_limit"}}'
+    echo "429"
+    exit 0
+  fi
+done
+# Non-API curl calls: pass through to system curl
+exec /usr/bin/curl "$@"
+CURLEOF
+  chmod +x "$STUB_BIN_DIR/curl"
+
   # Default environment for fix-ci
   export PR_NUMBER="42"
   export HEAD_SHA="abc123def456"
@@ -27,6 +44,8 @@ setup() {
   export REVIEW_ENGINE="claude"
   export DEV_LEAD_DRY_RUN="true"
   export GH_STUB_COMMENT_BODY=""
+  export COPILOT_GITHUB_TOKEN="stub-token"
+  export COPILOT_API_MODEL="openai/o4-mini"
   # Point to project root for prompts
   cd "$SCRIPT_DIR"
 }
diff --git a/tests/dev-lead/unit/test_fix_reviews.bats b/tests/dev-lead/unit/test_fix_reviews.bats
index acecf127c..7d7be49d7 100644
--- a/tests/dev-lead/unit/test_fix_reviews.bats
+++ b/tests/dev-lead/unit/test_fix_reviews.bats
@@ -18,6 +18,21 @@ setup() {
   export PATH="$STUB_BIN_DIR:$PATH"
   export STUB_BIN_DIR
 
+  # Stub curl so copilot_chat returns a 429 rate-limit response for GitHub Models API.
+  # This makes the copilot fallback behave as rate-limited without a real API call.
+  cat > "$STUB_BIN_DIR/curl" << 'CURLEOF'
+#!/usr/bin/env bash
+for arg in "$@"; do
+  if [[ "$arg" == *"models.github.ai"* ]]; then
+    echo '{"error":{"message":"rate limit exceeded","type":"rate_limit"}}'
+    echo "429"
+    exit 0
+  fi
+done
+exec /usr/bin/curl "$@"
+CURLEOF
+  chmod +x "$STUB_BIN_DIR/curl"
+
   # Default env
   export PR_NUMBER="54"
   export HEAD_SHA="ddd444eee555"
@@ -27,6 +42,8 @@ setup() {
   export GITHUB_REPOSITORY="petry-projects/.github-private"
   export BASE_REF="main"
   export ACTOR="donpetry"
+  export COPILOT_GITHUB_TOKEN="stub-token"
+  export COPILOT_API_MODEL="openai/o4-mini"
 
   # Install a graphql-aware gh stub
   cat > "$STUB_BIN_DIR/gh" <<'GHEOF'