diff --git a/ci/agents/pr-review/mistral-agent-sync.sh b/ci/agents/pr-review/mistral-agent-sync.sh
new file mode 100644
index 00000000..ae61d1e1
--- /dev/null
+++ b/ci/agents/pr-review/mistral-agent-sync.sh
@@ -0,0 +1,49 @@
+#!/bin/sh
+# Create or update the listed Mistral agent "pr-review" from mistral-prompt.md, so the
+# agent definition in Mistral Studio follows git. Prints the agent id.
+#
+# Run it after changing mistral-prompt.md, the model or the connector:
+#   MISTRAL_API_KEY=... sh ci/agents/pr-review/mistral-agent-sync.sh
+#
+# Env: MISTRAL_API_KEY (required), MISTRAL_AGENT_NAME, MISTRAL_MODEL, MISTRAL_CONNECTOR_ID,
+#      MISTRAL_API_BASE, MISTRAL_MAX_TOKENS.
+set -eu
+HERE="$(cd "$(dirname "$0")" && pwd)"
+: "${MISTRAL_API_KEY:?MISTRAL_API_KEY is not set}"
+NAME="${MISTRAL_AGENT_NAME:-pr-review}"
+MODEL="${MISTRAL_MODEL:-mistral-large-4-0}"
+CONNECTOR="${MISTRAL_CONNECTOR_ID:-01a12adf-cdc3-7562-b32b-b1fa96be244c}"
+BASE="${MISTRAL_API_BASE:-https://api.mistral.ai/v1}"
+MAX_TOKENS="${MISTRAL_MAX_TOKENS:-64000}"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+
+# The key goes to curl through a config file, not argv.
+umask 077
+printf 'header = "Authorization: Bearer %s"\n' "$MISTRAL_API_KEY" > "$WORK/auth"
+api() { # method path [datafile] -> body on stdout, fails on HTTP >= 300
+  if [ -n "${3:-}" ]; then set -- "$1" "$2" -d @"$3"; else set -- "$1" "$2"; fi
+  m="$1"; p="$2"; shift 2
+  code=$(curl -s -m 60 -K "$WORK/auth" -H 'Content-Type: application/json' -X "$m" "$@" \
+    -o "$WORK/resp" -w '%{http_code}' "$BASE$p")
+  if [ "$code" -ge 300 ]; then echo "Mistral API $m $p -> HTTP $code: $(head -c 400 "$WORK/resp")" >&2; return 1; fi
+  cat "$WORK/resp"
+}
+
+HASH="$(sha256sum "$HERE/mistral-prompt.md" | cut -c1-12)"
+jq -n --arg name "$NAME" --arg model "$MODEL" --arg cid "$CONNECTOR" --argjson max "$MAX_TOKENS" \
+  --arg desc "PR review, read-only Gitea connector. prompt sha256:$HASH" \
+  --rawfile instructions "$HERE/mistral-prompt.md" \
+  '{name: $name, model: $model, description: $desc, instructions: $instructions,
+    tools: [{type: "connector", connector_id: $cid}],
+    completion_args: {temperature: 0.1, max_tokens: $max}}' > "$WORK/body.json"
+
+ID="$(api GET "/agents?page_size=100" | jq -r --arg n "$NAME" '[.[]? | select(.name == $n) | .id][0] // empty')"
+if [ -n "$ID" ]; then
+  api PATCH "/agents/$ID" "$WORK/body.json" > /dev/null
+  echo "updated $NAME (prompt sha256:$HASH)" >&2
+else
+  ID="$(api POST "/agents" "$WORK/body.json" | jq -r '.id')"
+  echo "created $NAME (prompt sha256:$HASH)" >&2
+fi
+echo "$ID"
diff --git a/ci/agents/pr-review/mistral-prompt.md b/ci/agents/pr-review/mistral-prompt.md
index 9c9d0112..11c7d6f2 100644
--- a/ci/agents/pr-review/mistral-prompt.md
+++ b/ci/agents/pr-review/mistral-prompt.md
@@ -27,6 +27,21 @@ HARD RULES
    training data is not evidence for how current releases behave.
 5. Before you write "no other code does X", search for it.
 
+## Mandatory checks
+Do these for every PR, in addition to reading the diff and the changed files:
+1. Read the pull request itself for `mergeable`, its base branch and whether the base moved. Read each changed file
+   at the base branch tip as well as at the head, and list recent commits on the base branch that touch those files.
+   State plainly if the change is already on the base branch, superseded by another PR, or conflicts with it. That is
+   a blocking finding.
+2. Read every issue the PR links to.
+3. Search the repository for other code that does the same thing before you write that nothing else does.
+
+## Efficiency
+Request all independent reads in the same step instead of one at a time (the diff, the changed files at head and
+base, neighbouring files, README or CLAUDE.md together). Do not repeat a call whose result you already have. There is
+no cap on tool calls; stop reading when every mandatory check is done and every suspicion is settled or listed as
+unverified.
+
 ## What to look for, in priority order
 1. Correctness: real bugs, wrong conditionals, unhandled errors, race conditions, resource leaks, incorrect API
    usage, broken backward compatibility, functions whose return value is not what the caller assumes.
diff --git a/ci/agents/pr-review/mistral-review.sh b/ci/agents/pr-review/mistral-review.sh
index ae82a279..5148bbf0 100644
--- a/ci/agents/pr-review/mistral-review.sh
+++ b/ci/agents/pr-review/mistral-review.sh
@@ -1,45 +1,49 @@
 #!/bin/sh
-# Review one PR with a Mistral agent that reads the repository through the Gitea
-# connector registered in Mistral Studio (read-only credential, see mistral-prompt.md).
+# Review one PR with the listed Mistral agent "pr-review". The agent holds the prompt, the
+# model and the read-only Gitea connector (kept in sync from git by mistral-agent-sync.sh).
 #
 # Usage: mistral-review.sh <workdir> <owner/repo> <pr number> <head sha>
-# Writes <workdir>/report.md. Exits non-zero, and leaves no report, when the request
-# fails or the agent called a tool that is not read-only.
+# Writes <workdir>/report.md and <workdir>/model.txt. Exits non-zero, and leaves no report,
+# when the request fails or the agent called a tool that is not read-only.
 #
-# Env: MISTRAL_API_KEY (required), MISTRAL_MODEL, MISTRAL_CONNECTOR_ID, MISTRAL_API,
-#      MISTRAL_MAX_TOKENS, MISTRAL_TIMEOUT.
+# Env: MISTRAL_API_KEY (required), MISTRAL_AGENT_ID, MISTRAL_API_BASE, MISTRAL_TIMEOUT.
 set -eu
 D="$1"; SLUG="$2"; NUM="$3"; HEAD="$4"
 HERE="$(cd "$(dirname "$0")" && pwd)"
 : "${MISTRAL_API_KEY:?MISTRAL_API_KEY is not set}"
-MODEL="${MISTRAL_MODEL:-mistral-large-4-0}"
-CONNECTOR="${MISTRAL_CONNECTOR_ID:-01a12adf-cdc3-7562-b32b-b1fa96be244c}"
-API="${MISTRAL_API:-https://api.mistral.ai/v1/conversations}"
-MAX_TOKENS="${MISTRAL_MAX_TOKENS:-64000}"
+AGENT="${MISTRAL_AGENT_ID:-ag_01a12b392002750b8ad2b4692a9de03c}"
+BASE="${MISTRAL_API_BASE:-https://api.mistral.ai/v1}"
 TIMEOUT="${MISTRAL_TIMEOUT:-1500}"
 
-rm -f "$D/report.md"
-jq -n --arg model "$MODEL" --arg connector "$CONNECTOR" --argjson max "$MAX_TOKENS" \
-  --arg task "Review pull request #$NUM of $SLUG. Head commit: $HEAD." \
-  --rawfile prompt "$HERE/mistral-prompt.md" \
-  '{model: $model, inputs: ($task + "\n\n" + $prompt), store: false,
-    tools: [{type: "connector", connector_id: $connector}],
-    completion_args: {temperature: 0.1, max_tokens: $max}}' > "$D/mistral-request.json"
-
+rm -f "$D/report.md" "$D/model.txt"
 # The key goes to curl through a config file, not argv.
 umask 077
 printf 'header = "Authorization: Bearer %s"\n' "$MISTRAL_API_KEY" > "$D/.curl-auth"
+
+# Model name for the review header, and a warning when the agent's prompt is not the repo's.
+if curl -sf -m 60 -K "$D/.curl-auth" -o "$D/agent.json" "$BASE/agents/$AGENT"; then
+  jq -r '.model' "$D/agent.json" > "$D/model.txt"
+  HASH="$(sha256sum "$HERE/mistral-prompt.md" | cut -c1-12)"
+  jq -e --arg h "sha256:$HASH" '.description | contains($h)' "$D/agent.json" > /dev/null \
+    || echo "WARNING: agent $AGENT was not built from this mistral-prompt.md; run mistral-agent-sync.sh."
+else
+  echo "WARNING: could not read agent $AGENT."
+fi
+
+jq -n --arg agent "$AGENT" --arg task "Review pull request #$NUM of $SLUG. Head commit: $HEAD." \
+  '{agent_id: $agent, inputs: $task, store: false}' > "$D/mistral-request.json"
+
 code=000
 for TRY in 1 2; do
   code=$(curl -s -m "$TIMEOUT" -K "$D/.curl-auth" -H 'Content-Type: application/json' \
-    -d @"$D/mistral-request.json" -o "$D/mistral-response.json" -w '%{http_code}' "$API" || echo 000)
+    -d @"$D/mistral-request.json" -o "$D/mistral-response.json" -w '%{http_code}' "$BASE/conversations" || echo 000)
   [ "$code" = 200 ] && break
   echo "Mistral request attempt $TRY failed (HTTP $code): $(head -c 300 "$D/mistral-response.json" 2>&1 || true)"
 done
 rm -f "$D/.curl-auth"
 [ "$code" = 200 ] || exit 1
 
-echo "Mistral agent ($MODEL) made $(sh "$HERE/parse-response.sh" calls < "$D/mistral-response.json") tool calls."
+echo "Mistral agent $AGENT made $(sh "$HERE/parse-response.sh" calls < "$D/mistral-response.json") tool calls."
 BAD=$(sh "$HERE/parse-response.sh" violations < "$D/mistral-response.json")
 if [ -n "$BAD" ]; then
   echo "The agent called tools that are not read-only; no review is posted:"
diff --git a/ci/agents/pr-review/run.sh b/ci/agents/pr-review/run.sh
index e9878440..55039d49 100755
--- a/ci/agents/pr-review/run.sh
+++ b/ci/agents/pr-review/run.sh
@@ -147,14 +147,15 @@ while IFS= read -r ITEM; do
   #     --settings "$PROMPTS/settings.json" \
   #     -p ) || echo "(review agent exited non-zero)"
 
-  # --- Review with a Mistral agent (read-only Gitea connector, see mistral-review.sh) ---
-  MODEL="${MISTRAL_MODEL:-mistral-large-4-0}"
-  echo "Review model: $MODEL"
+  # --- Review with the listed Mistral agent "pr-review" (see mistral-review.sh) ---
   if ! sh "$PROMPTS/mistral-review.sh" "$D" "$SLUG" "$NUM" "$HEAD"; then
     echo "Mistral review failed for $SLUG#$NUM."
     REVIEW_FAILED=1
     continue
   fi
+  MODEL="Mistral agent"
+  [ -s "$D/model.txt" ] && MODEL="$(cat "$D/model.txt")"
+  echo "Review model: $MODEL"
 
   if [ ! -s "$D/report.md" ]; then
     echo "Agent wrote no report for $SLUG#$NUM — skipping post."