crewly 1.20.273 → 1.20.275

package/config/roles/team-leader/tl-addon.md

@@ -87,6 +87,13 @@ bash {{TL_SKILLS_PATH}}/assign-ticket/execute.sh --project {{PROJECT_PATH}} --id
 ```
 Use when: A specific worker should take a specific ticket of the project backlog. The worker gets a linked WorkItem and is dispatched; when you verify that WorkItem the ticket moves to done.
 
+#### 11. propose-tier-change — Propose model tiers and routing (only when "Optimize usage" is on)
+```bash
+bash {{TL_SKILLS_PATH}}/propose-tier-change/execute.sh --member "<name>" --tier weak --reason "<why>"
+bash {{TL_SKILLS_PATH}}/propose-tier-change/execute.sh --submit
+```
+Use when: your weekly usage review arrives (`[MODEL TIER REVIEW]`), or you see a member's work does not need its tier. One owner card per review; nothing changes until the owner applies it.
+
 #### Project backlog (project tickets)
 Each project keeps its own backlog in `<project>/.crewly/tickets/` (one markdown file per ticket, tracked in git). Use the `project-tickets` skill to list, read, create and update tickets. Workers pick up `ready` tickets by themselves when idle; you decide what is `ready`, you may assign (`assign-ticket`), and you review the work through the normal verification item. Worker-created tickets land in `backlog` until you or the owner make them ready.
 

package/config/skills/agent/core/ask-owner/SKILL.md

@@ -67,9 +67,14 @@ options. Use plain Yes/No only for a single question with one possible action.
 
 What happens:
 
-- The card is posted by your own Slack bot:
-  - a ticket ask goes in the ticket's thread;
-  - any other ask goes where your current work is, else as a new thread in your team channel.
+- The card is posted by your own Slack bot, in the conversation it belongs to:
+  - the owner thread you are talking in right now (the owner is waiting there);
+  - else the ticket's thread (ticket asks), else your current work's thread;
+  - top-level in your team channel only when there is no thread at all.
+- Asking a question you already have an open card for, in the same thread / ticket, posts nothing
+  new: you get the open card back with `"reused": true`. Wait for its answer.
+- When the owner answers one card, Crewly closes your other open cards in that thread that ask the
+  same thing (the card reads `✓ Closed — answered in D-7`); you are told which ones.
 - The owner can answer by tapping a button, reacting (✅ default, ❌ "no", ⏰ tomorrow, 🚫 skip),
   replying in the thread, or using the dashboard.
 - You receive `[DECISION D-7] The owner chose "…"`. Act on it. Do not ask again.
@@ -85,13 +90,22 @@ What happens:
   transcript or the file link. Listen to it (transcribe-audio) and act on it.
 - Do not also message the owner about it. The card is the question.
 
-Withdraw a question you no longer need:
+Withdraw a card you no longer need — and **right away when a card should not exist**: you posted it
+by mistake, it was already answered in the thread, or Crewly made a card from a sentence in your
+reply that was not a question for the owner (a line of a script, interview questions for a client,
+a draft you quoted):
 
 ```bash
-bash execute.sh --cancel D-7 --reason "already answered in the thread"
+bash execute.sh --withdraw D-7 --reason "not a question for you — interview question for the client"
+bash execute.sh --mine        # your open cards: id, question, thread
 ```
 
-The card then reads `Closed — already answered in the thread`.
+(`--cancel` is the same.) The Slack card closes and loses its buttons: `✓ Closed — <reason>`. Do not
+leave a wrong card open for the owner to ask about.
+
+Crewly turns only the **last** question of your reply into a card, and only when it is a decision for
+the owner. A question in the middle of a reply, a numbered list of questions, or questions written
+for someone else get no card. Several decisions at once? Ask each with its own `ask-owner`.
 
 A rejected ask (vague, missing options, etc.) comes back with an error that says what to fix.
 

package/config/skills/agent/core/ask-owner/execute.sh

@@ -13,15 +13,17 @@ Usage:
     --option "Send Monday — after the review call" --option "Hold — wait for legal" \
     --default "Hold" [--deadline 2026-10-02T12:00] [--ticket APP-12 --project P] \
     [--sensitive email|publish|deploy|spend]
-  bash execute.sh --cancel D-7 [--reason "already answered in the thread"]
-                                       Withdraw a question you no longer need (the card shows the reason)
+  bash execute.sh --withdraw D-7 [--reason "posted by mistake"]
+                                       Withdraw a card you no longer need, or one posted by mistake
+                                       (--cancel is the same). The Slack card closes and loses its buttons.
+  bash execute.sh --mine               List your open cards (id, question, where)
   bash execute.sh --status D-7         Read a card: answered or not, and what the owner chose.
                                        Check this before acting on anything you asked about.
   bash execute.sh '{"question":"…","options":["A","B"],"default":"A"}'
 EOF_USAGE
 }
 
-QUESTION=""; OPTIONS_JSON="[]"; DEFAULT_OPT=""; DEADLINE=""; SENSITIVE=""; TICKET=""; PROJECT=""; CANCEL=""; REASON=""; STATUS_ID=""
+QUESTION=""; OPTIONS_JSON="[]"; DEFAULT_OPT=""; DEADLINE=""; SENSITIVE=""; TICKET=""; PROJECT=""; CANCEL=""; REASON=""; STATUS_ID=""; MINE=""
 
 if [[ $# -gt 0 && ${1:0:1} == '{' ]]; then
   J="$1"; shift
@@ -32,7 +34,7 @@ if [[ $# -gt 0 && ${1:0:1} == '{' ]]; then
   SENSITIVE=$(printf '%s' "$J" | jq -r '.sensitive // empty')
   TICKET=$(printf '%s' "$J" | jq -r '.ticket // empty')
   PROJECT=$(printf '%s' "$J" | jq -r '.project // empty')
-  CANCEL=$(printf '%s' "$J" | jq -r '.cancel // empty')
+  CANCEL=$(printf '%s' "$J" | jq -r '.cancel // .withdraw // empty')
   REASON=$(printf '%s' "$J" | jq -r '.reason // empty')
 fi
 
@@ -46,7 +48,8 @@ while [[ $# -gt 0 ]]; do
     --sensitive)    [ $# -ge 2 ] || error_exit "--sensitive requires a value"; SENSITIVE="$2"; shift 2 ;;
     --ticket|--id)  [ $# -ge 2 ] || error_exit "--ticket requires a value"; TICKET="$2"; shift 2 ;;
     --project|-p)   [ $# -ge 2 ] || error_exit "--project requires a value"; PROJECT="$2"; shift 2 ;;
-    --cancel)       [ $# -ge 2 ] || error_exit "--cancel requires a decision id"; CANCEL="$2"; shift 2 ;;
+    --cancel|--withdraw) [ $# -ge 2 ] || error_exit "$1 requires a decision id"; CANCEL="$2"; shift 2 ;;
+    --mine)         MINE=1; shift ;;
     --reason)       [ $# -ge 2 ] || error_exit "--reason requires a value"; REASON="$2"; shift 2 ;;
     --status)       [ $# -ge 2 ] || error_exit "--status requires a decision id"; STATUS_ID="$2"; shift 2 ;;
     --help|-h)      print_usage; exit 0 ;;
@@ -69,6 +72,15 @@ if [ -n "$STATUS_ID" ]; then
   exit 0
 fi
 
+if [ -n "$MINE" ]; then
+  [ -n "${CREWLY_SESSION_NAME:-}" ] || error_exit "--mine needs CREWLY_SESSION_NAME (your session name)"
+  api_call GET "/decisions?status=open" | jq --arg me "$CREWLY_SESSION_NAME" '{success, decisions: [(.data // [])[] | select(.asker == $me) | {
+      id, status, question, kind: (.kind // "ask-owner"), createdAt,
+      thread: (if .card then "\(.card.slackChannelId):\(.card.threadTs // .card.messageTs)" else null end)
+    }]}'
+  exit 0
+fi
+
 if [ -n "$CANCEL" ]; then
   CANCEL_BODY=$(jq -n --arg r "$REASON" 'if $r != "" then {note: $r} else {} end')
   api_call POST "/decisions/$(printf '%s' "$CANCEL" | jq -sRr @uri)/cancel" "$CANCEL_BODY" | jq '{success, decision: (.data | {id, status})}'
@@ -85,4 +97,4 @@ BODY=$(jq -n --arg q "$QUESTION" --argjson o "$OPTIONS_JSON" --arg d "$DEFAULT_O
    + (if $t != "" then {ticket: $t} else {} end)
    + (if $p != "" then {project: $p} else {} end)')
 api_call POST "/decisions" "$BODY" \
-  | jq '{success, decision: (.data | {id, asker, status, deadline, posted: (.card != null), postError})}'
+  | jq '{success, decision: (.data | {id, asker, status, deadline, posted: (.card != null), postError, reused: (.reused // false)})}'

package/config/skills/agent/core/ask-owner/execute.test.sh

@@ -26,6 +26,11 @@ class H(BaseHTTPRequestHandler):
             return self.reply(200, {'success': True, 'data': {'id': 'D-11', 'status': 'open', 'options': [{'key': 'a', 'label': 'Reply with this draft'}, {'key': 'b', 'label': 'Change the wording'}]}})
         if self.path.startswith('/api/decisions/D-12'):
             return self.reply(200, {'success': True, 'data': {'id': 'D-12', 'status': 'resolved', 'chosenKey': 'a', 'answeredVia': 'button', 'resolvedAt': 't', 'options': [{'key': 'a', 'label': 'Reply with this draft'}, {'key': 'b', 'label': 'Change the wording'}]}})
+        if self.path.startswith('/api/decisions?status=open'):
+            return self.reply(200, {'success': True, 'data': [
+                {'id': 'D-20', 'asker': 'dev-ann', 'status': 'open', 'question': 'Ship it?', 'kind': 'reply_question', 'createdAt': 't', 'card': {'slackChannelId': 'C1', 'messageTs': '2.2', 'threadTs': '1.1'}},
+                {'id': 'D-21', 'asker': 'dev-bob', 'status': 'open', 'question': 'Other?', 'createdAt': 't'},
+                {'id': 'D-22', 'asker': 'dev-ann', 'status': 'open', 'question': 'Top-level?', 'createdAt': 't', 'card': {'slackChannelId': 'C1', 'messageTs': '3.3'}}]})
         return self.reply(200, {'success': True})
     def do_POST(self):
         n = int(self.headers.get('content-length', '0')); body = json.loads(self.rfile.read(n).decode() or '{}')
@@ -51,7 +56,7 @@ OUT=$(run --question "Send the partner email on Monday?" --option "Send Monday
 check "ask: path" "$(last .path)" "/api/decisions"
 check "ask: agent header" "$(last .agent)" "dev-ann"
 check "ask: body" "$(last '.body | tostring')" '{"question":"Send the partner email on Monday?","options":["Send Monday — after review","Hold"],"default":"Hold","sensitive":"email","ticket":"APP-12","project":"p1"}'
-check "ask: output" "$(printf '%s' "$OUT" | jq -c '.decision')" '{"id":"D-7","asker":"dev-ann","status":"open","deadline":"x","posted":true,"postError":null}'
+check "ask: output" "$(printf '%s' "$OUT" | jq -c '.decision')" '{"id":"D-7","asker":"dev-ann","status":"open","deadline":"x","posted":true,"postError":null,"reused":false}'
 
 run '{"question":"Blue or green for the logo?","options":["Blue","Green"],"default":"wait"}' >/dev/null
 check "ask json: body" "$(last '.body | tostring')" '{"question":"Blue or green for the logo?","options":["Blue","Green"],"default":"wait"}'
@@ -66,6 +71,19 @@ check "cancel: no reason → empty body" "$(last .body | jq -c .)" '{}'
 run --cancel D-7 --reason "already answered in the thread" >/dev/null
 check "cancel: reason sent as note" "$(last .body.note)" "already answered in the thread"
 
+# --withdraw: the same as --cancel (a card posted by mistake)
+OUT=$(run --withdraw D-7 --reason "posted by mistake")
+check "withdraw: path" "$(last .path)" "/api/decisions/D-7/cancel"
+check "withdraw: reason sent as note" "$(last .body.note)" "posted by mistake"
+check "withdraw: output" "$(printf '%s' "$OUT" | jq -c '.decision')" '{"id":"D-7","status":"cancelled"}'
+run '{"withdraw":"D-7"}' >/dev/null
+check "withdraw json: path" "$(last .path)" "/api/decisions/D-7/cancel"
+
+# --mine: only the caller's open cards, with their thread
+OUT=$(run --mine)
+check "mine: path" "$(last .path)" "/api/decisions?status=open"
+check "mine: own cards only" "$(printf '%s' "$OUT" | jq -c '[.decisions[] | {id, kind, thread}]')" '[{"id":"D-20","kind":"reply_question","thread":"C1:1.1"},{"id":"D-22","kind":"ask-owner","thread":"C1:3.3"}]'
+
 # --status: an agent reads a card before acting (2026-10-03 phantom owner input)
 OUT=$(run --status D-11)
 check "status: path" "$(last .path)" "/api/decisions/D-11"

package/config/skills/agent/core/reply/SKILL.md

@@ -115,3 +115,14 @@ tells the system that you are the one replying.
   command to run — nothing is dropped silently, and nothing is filed as
   status for the orchestrator.
 - `reply-channel`, `reply-chat` and `slack-post` keep working.
+
+## Questions in your reply
+
+- Crewly turns the **last** question of your reply into a tap-to-answer card when it is a decision
+  for the owner ("这样写可以吗？", "Ship it today?"). Put the one question you need answered last.
+- Questions in the middle, numbered lists of questions, quoted drafts and questions written for
+  someone else (interview / survey questions, a script) get no card. Need several decisions? Ask
+  each with `ask-owner`.
+- A card that should not exist (it picked up a line that was not a question for the owner): withdraw
+  it at once — `ask-owner --withdraw D-12 --reason "not a question for you"` (`ask-owner --mine`
+  lists your open cards). The card closes and loses its buttons.

package/config/skills/team-leader/propose-tier-change/SKILL.md

@@ -0,0 +1,62 @@
+---
+name: Propose Tier Change (TL)
+description: "Propose moving a member to another model tier (strong / mid / weak) or a task routing rule, then send all proposals to the owner as one card. Only while the team's 'Optimize usage' is on. Nothing changes until the owner says yes."
+version: 1.0.0
+category: management
+skillType: claude-skill
+assignableRoles:
+  - team-leader
+triggers:
+  - propose tier change
+  - model tier review
+  - optimize usage
+tags:
+  - agent
+  - management
+  - model
+  - cost
+execution:
+  type: script
+  script:
+    file: execute.sh
+    interpreter: bash
+    timeoutMs: 30000
+---
+
+# Propose Tier Change (TL Version)
+
+Members run on model tiers: **strong** (hard reasoning, design, reviewing others), **mid** (normal implementation and writing), **weak** (routine work: polling, checks, formatting, sorting, first-pass triage, summaries). Each tier maps to a model per runtime (Claude Code: opus / sonnet / haiku).
+
+While the team's **Optimize usage** setting is on, you get a usage review every week (turns, context, output, estimated cost and work handled per member). Propose what should change, then submit. The owner gets **one card** with all proposals and taps Apply or Keep as is. You are told the answer.
+
+## Usage
+
+```bash
+# 1. One call per member that should move
+bash {{TL_SKILLS_PATH}}/propose-tier-change/execute.sh --member "Ella" --tier weak --reason "only polls the inbox and sorts tickets; nothing was sent back"
+
+# 2. Routing rules for incoming work (optional)
+bash {{TL_SKILLS_PATH}}/propose-tier-change/execute.sh --routing "polling / formatting / sorting -> Ella"
+
+# 3. Send everything to the owner (one card). With nothing proposed, this just closes the review.
+bash {{TL_SKILLS_PATH}}/propose-tier-change/execute.sh --submit
+```
+
+`--clear` drops the draft. Proposing the same member again replaces its earlier proposal.
+
+## Rules
+
+- Never change a tier or model any other way; only the owner's Apply does it.
+- Do not lower a member whose work was often sent back. You stay at mid or above.
+- A member with a fixed model loses it when the owner applies its tier (the card says so).
+- A draft you do not submit is sent to the owner automatically after two hours.
+- After a member moves down, Crewly watches its next work items; if they are sent back clearly more often, the owner is asked whether to move it back.
+
+## Output
+
+`{"success":true,"data":{"added":"change","change":"Ella: mid -> weak","draft":{...}}}` / `{"success":true,"data":{"submitted":true,"decisionId":"D-12",...}}`
+
+## Related Skills
+
+- `start-agent` / `stop-agent` — restart an idle member so an approved tier applies now
+- `delegate-task` — route work by tier

package/config/skills/team-leader/propose-tier-change/execute.sh

@@ -0,0 +1,64 @@
+#!/bin/bash
+# propose-tier-change — a team lead proposes model-tier changes and task
+# routing rules for its team (crewly#1173). Proposals collect in a draft;
+# --submit sends ONE owner card with all of them. Nothing changes until the
+# owner taps Apply. See specs/2026-10-08-model-tiers.md.
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/../_common/lib.sh"
+
+print_usage() {
+  cat <<'EOF_USAGE'
+Usage:
+  bash execute.sh --member "Ella" --tier weak --reason "only polls the inbox and sorts tickets"
+  bash execute.sh --routing "polling / formatting / sorting -> Ella"
+  bash execute.sh --submit          Send the draft to the owner (one card). With nothing proposed, closes the review.
+  bash execute.sh --clear           Drop the draft without sending it.
+  bash execute.sh '{"member":"Ella","tier":"weak","reason":"..."}'
+Tiers: strong (hard reasoning, design, review), mid (normal work), weak (routine: polling, checks, formatting, sorting, triage).
+EOF_USAGE
+}
+
+MEMBER=""; TIER=""; REASON=""; ROUTING=""; SUBMIT=""; CLEAR=""
+
+if [[ $# -gt 0 && ${1:0:1} == '{' ]]; then
+  J="$1"; shift
+  MEMBER=$(printf '%s' "$J" | jq -r '.member // empty')
+  TIER=$(printf '%s' "$J" | jq -r '.tier // empty')
+  REASON=$(printf '%s' "$J" | jq -r '.reason // empty')
+  ROUTING=$(printf '%s' "$J" | jq -r '.routing // empty')
+  [ "$(printf '%s' "$J" | jq -r '.submit // false')" = "true" ] && SUBMIT=1
+  [ "$(printf '%s' "$J" | jq -r '.clear // false')" = "true" ] && CLEAR=1
+fi
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --member|-m)  [ $# -ge 2 ] || error_exit "--member requires a value"; MEMBER="$2"; shift 2 ;;
+    --tier|-t)    [ $# -ge 2 ] || error_exit "--tier requires a value"; TIER=$(printf '%s' "$2" | tr '[:upper:]' '[:lower:]'); shift 2 ;;
+    --reason|-r)  [ $# -ge 2 ] || error_exit "--reason requires a value"; REASON="$2"; shift 2 ;;
+    --routing)    [ $# -ge 2 ] || error_exit "--routing requires a value"; ROUTING="$2"; shift 2 ;;
+    --submit)     SUBMIT=1; shift ;;
+    --clear)      CLEAR=1; shift ;;
+    --help|-h)    print_usage; exit 0 ;;
+    *) error_exit "Unknown option: $1 (see --help)" ;;
+  esac
+done
+
+if [ -n "$CLEAR" ]; then
+  BODY='{"clear":true}'
+elif [ -n "$SUBMIT" ]; then
+  BODY='{"submit":true}'
+elif [ -n "$ROUTING" ]; then
+  BODY=$(jq -n --arg r "$ROUTING" '{routing: $r}')
+else
+  require_param "member" "$MEMBER"
+  require_param "tier" "$TIER"
+  require_param "reason" "$REASON"
+  case "$TIER" in
+    strong|mid|weak) ;;
+    *) error_exit "Invalid tier '${TIER}': use strong, mid or weak" ;;
+  esac
+  BODY=$(jq -n --arg m "$MEMBER" --arg t "$TIER" --arg r "$REASON" '{member: $m, tier: $t, reason: $r}')
+fi
+
+api_call POST "/teams/model-tiers/proposals" "$BODY"

package/config/skills/team-leader/propose-tier-change/execute.test.sh

@@ -0,0 +1,75 @@
+#!/bin/bash
+# Tests for TL propose-tier-change execute.sh
+# A fake `curl` on PATH records the request body, so the real script (and the
+# real lib.sh) run unmodified.
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+EXECUTE="$SCRIPT_DIR/execute.sh"
+PASS=0
+FAIL=0
+
+TMP=$(mktemp -d)
+trap 'rm -rf "$TMP"' EXIT
+mkdir -p "$TMP/bin"
+cat > "$TMP/bin/curl" << 'CURL_EOF'
+#!/bin/bash
+method="GET"; body=""; url=""; session=""
+while [ $# -gt 0 ]; do
+  case "$1" in
+    -X) method="$2"; shift 2 ;;
+    -d) body="$2"; shift 2 ;;
+    -H) case "$2" in "X-Agent-Session: "*) session="${2#X-Agent-Session: }" ;; esac; shift 2 ;;
+    -w) shift 2 ;;
+    -s) shift ;;
+    *) url="$1"; shift ;;
+  esac
+done
+case "$method $url" in
+  "POST "*/api/teams/model-tiers/proposals)
+    printf '%s' "$body" > "$FAKE_CURL_LOG"
+    printf '%s' "$session" > "$FAKE_CURL_LOG.session"
+    printf '%s\n200' '{"success":true,"data":{"ok":true}}'
+    ;;
+  *)
+    printf '%s\n404' '{"error":"unknown"}'
+    ;;
+esac
+CURL_EOF
+chmod +x "$TMP/bin/curl"
+
+run() {
+  : > "$TMP/body.json"
+  env PATH="$TMP/bin:$PATH" CREWLY_HOME="$TMP/home" CREWLY_SESSION_NAME="tl-session" \
+    FAKE_CURL_LOG="$TMP/body.json" bash "$EXECUTE" "$@"
+}
+
+check() {
+  local desc="$1" ok="$2"
+  if [ "$ok" = "yes" ]; then echo "  PASS: $desc"; PASS=$((PASS + 1)); else echo "  FAIL: $desc"; FAIL=$((FAIL + 1)); fi
+}
+
+echo "=== TL propose-tier-change tests ==="
+
+OUT=$(run --member "Ella" --tier Weak --reason "polls only" 2>&1) && RC=0 || RC=$?
+check "member change: succeeds" "$([ "$RC" = 0 ] && echo yes || echo no)"
+check "member change: posts member/tier/reason" "$([ "$(jq -c . "$TMP/body.json")" = '{"member":"Ella","tier":"weak","reason":"polls only"}' ] && echo yes || echo no)"
+check "member change: sent as the lead" "$([ "$(cat "$TMP/body.json.session")" = "tl-session" ] && echo yes || echo no)"
+
+OUT=$(run --routing "polling -> Ella" 2>&1) && RC=0 || RC=$?
+check "routing: posts the rule" "$([ "$RC" = 0 ] && [ "$(jq -r .routing "$TMP/body.json")" = "polling -> Ella" ] && echo yes || echo no)"
+
+OUT=$(run --submit 2>&1) && RC=0 || RC=$?
+check "submit: posts submit:true" "$([ "$RC" = 0 ] && [ "$(jq -r .submit "$TMP/body.json")" = "true" ] && echo yes || echo no)"
+
+OUT=$(run '{"clear":true}' 2>&1) && RC=0 || RC=$?
+check "json clear: posts clear:true" "$([ "$RC" = 0 ] && [ "$(jq -r .clear "$TMP/body.json")" = "true" ] && echo yes || echo no)"
+
+OUT=$(run --member "Ella" --tier super --reason "x" 2>&1) && RC=0 || RC=$?
+check "invalid tier: fails without a request" "$([ "$RC" != 0 ] && [ ! -s "$TMP/body.json" ] && echo "$OUT" | grep -q "Invalid tier" && echo yes || echo no)"
+
+OUT=$(run --member "Ella" --tier weak 2>&1) && RC=0 || RC=$?
+check "missing reason: fails" "$([ "$RC" != 0 ] && echo "$OUT" | grep -q "Missing required parameter: reason" && echo yes || echo no)"
+
+echo ""
+echo "Results: $PASS passed, $FAIL failed"
+[ "$FAIL" -eq 0 ]

package/dist/backend/backend/src/constants.d.ts

@@ -3296,6 +3296,52 @@ export declare const MEMBER_MODEL_DEFAULT_CONSTANTS: {
     /** Rejections of the same task after which the reviewer is told about the upgrade option */
     readonly UPGRADE_HINT_AFTER_REJECTIONS: 2;
 };
+/**
+ * Model tiers (crewly#1173, specs/2026-10-08-model-tiers.md): strong / mid /
+ * weak, mapped to a concrete model per runtime. A team may override the map
+ * (`Team.tierModels`); an explicit `TeamMember.modelId` always wins.
+ *
+ * A runtime with no model for `weak` uses its `mid` model (still cheaper than
+ * the strong default). Any other unmapped tier gets no model flag: the
+ * runtime's own default. `opencode-cli` and `crewly-agent` are unmapped (their
+ * model ids are provider/model pairs the owner picks).
+ */
+export declare const MODEL_TIER_CONSTANTS: {
+    /** Tiers, strongest first */
+    readonly TIERS: readonly ["strong", "mid", "weak"];
+    /** Global tier → model map per runtime (names each harness accepts at launch) */
+    readonly DEFAULT_TIER_MODELS: Record<string, Partial<Record<"strong" | "mid" | "weak", string>>>;
+    /** Store of review drafts / applied changes, under CREWLY_HOME */
+    readonly STORE_FILE: "model-tier-reviews.json";
+    /** Decision kind of the owner card */
+    readonly DECISION_KIND: "model_tier_change";
+    /** Days between automatic reviews while "Optimize usage" is on */
+    readonly REVIEW_INTERVAL_DAYS: 7;
+    /** Usage window the report covers (days) */
+    readonly REPORT_WINDOW_DAYS: 7;
+    /** How often the service checks for due reviews / guards */
+    readonly TICK_MS: number;
+    /** A draft nobody submitted is sent to the owner after this long */
+    readonly DRAFT_AUTO_SUBMIT_MS: number;
+    /** How long a tier card waits for the owner (default = keep as is) */
+    readonly DECISION_DEADLINE_MS: number;
+    /** Work titles listed per member in the report */
+    readonly REPORT_MAX_TITLES: 5;
+    /** Max characters of a reason / routing rule */
+    readonly MAX_REASON_CHARS: 300;
+    /** Max routing rules per proposal */
+    readonly MAX_ROUTING_RULES: 10;
+    /** Quality guard: settled work items needed after a lowered tier before judging */
+    readonly GUARD_MIN_ITEMS: 5;
+    /** Quality guard: settled items before the change used as the baseline */
+    readonly GUARD_BASELINE_ITEMS: 10;
+    /** Quality guard: send-back rate rise (0..1) over the baseline that counts as worse */
+    readonly GUARD_WORSE_MARGIN: 0.2;
+    /** Quality guard: at least this many sent-back items after the change */
+    readonly GUARD_MIN_BAD: 2;
+    /** Quality guard: stop watching a change after this many days */
+    readonly GUARD_MAX_DAYS: 30;
+};
 export declare const CHAT_CONTEXT_CONSTANTS: {
     /** Whether preceding messages are included at all */
     readonly ENABLED: true;
@@ -6560,7 +6606,23 @@ export declare const DECISION_CONSTANTS: {
         readonly TICKET_DONE: "ticket done";
         readonly TICKET_CANCELLED: "ticket cancelled";
         readonly HANDLED_IN_THREAD: "already handled in this thread";
+        /** `answered in D-7`: the same question, answered on another card */
+        readonly ANSWERED_IN_PREFIX: "answered in ";
+        /** `answered by your reply to D-7`: the owner's words on another card moved the thread on */
+        readonly ANSWERED_BY_REPLY_PREFIX: "answered by your reply to ";
+        /** The open item the card tracked expired (nothing happened for 7 days) */
+        readonly STALE: "no longer needed";
     };
+    /**
+     * Two questions with at least this character-bigram overlap, from the same
+     * asker in the same thread / ticket / request, are one question: a new ask
+     * reuses the open card, and an answer closes the other open ones.
+     */
+    readonly SAME_OPEN_QUESTION_SIMILARITY: 0.6;
+    /** A settled card Slack refused to update (buttons still live) is retried this often (ms) */
+    readonly CARD_SYNC_RETRY_MS: number;
+    /** Open cards are checked this often for what they track having closed (then withdrawn silently) (ms) */
+    readonly MOOT_CHECK_MS: number;
 };
 /**
  * Open items: commitments and questions agents put in their replies to the
@@ -6633,12 +6695,6 @@ export declare const OPEN_ITEMS_CONSTANTS: {
     readonly CONTEXT_BLOCK_MAX_CHARS: 1200;
     /** Max characters of the "About:" line (the owner's original ask) */
     readonly CONTEXT_ABOUT_MAX_CHARS: 200;
-    /**
-     * A ticket thread with no activity for this long (ms) is not revived by a
-     * question the agent asked somewhere else: the card goes where the agent's
-     * message is, with a link to the old thread.
-     */
-    readonly STALE_THREAD_MS: number;
     /**
      * Words that make a question point back at earlier text ("这样安排行不行？", "按这个草稿回？",
      * "Does this plan work?"): its card prefers the text right before it and