@seanyao/roll 2026.529.5 → 2026.601.2

package/lib/i18n/agent.sh

@@ -19,3 +19,57 @@ _i18n_set en agent.not_found_in_path_setting_anyway "%s not found in PATH — se
 _i18n_set zh agent.not_found_in_path_setting_anyway "未找到 %s，仍写入配置"
 _i18n_set en agent.available_agents "Available agents"
 _i18n_set zh agent.available_agents "可用 agent"
+
+# US-AGENT-025: `roll agent` four-slot view.
+_i18n_set en agent.view_header "Complexity routing (.roll/agents.yaml)"
+_i18n_set zh agent.view_header "复杂度路由（.roll/agents.yaml）"
+_i18n_set en agent.view_col_slot "SLOT"
+_i18n_set zh agent.view_col_slot "槽位"
+_i18n_set en agent.view_col_agent "AGENT"
+_i18n_set zh agent.view_col_agent "AGENT"
+_i18n_set en agent.view_col_status "STATUS"
+_i18n_set zh agent.view_col_status "状态"
+_i18n_set en agent.view_col_note "NOTE"
+_i18n_set zh agent.view_col_note "备注"
+_i18n_set en agent.view_slot_unset "(unset)"
+_i18n_set zh agent.view_slot_unset "(未设)"
+_i18n_set en agent.view_fallback_idle "idle"
+_i18n_set zh agent.view_fallback_idle "待命"
+_i18n_set en agent.view_fallback_active "active (in use)"
+_i18n_set zh agent.view_fallback_active "已启用"
+_i18n_set en agent.view_recent_downgrades "Recent downgrades"
+_i18n_set zh agent.view_recent_downgrades "近期降级"
+_i18n_set en agent.view_downgrade_line "%s slot %s → ran %s"
+_i18n_set zh agent.view_downgrade_line "%s 档本该 %s → 实际 %s"
+_i18n_set en agent.view_no_config "No .roll/agents.yaml yet — routing falls back to the first installed agent."
+_i18n_set zh agent.view_no_config "尚无 .roll/agents.yaml — 路由回退到首个已装 agent。"
+_i18n_set en agent.view_no_config_hint "Set up routing: roll agent set <slot> <agent>  (or migrate from a legacy config)"
+_i18n_set zh agent.view_no_config_hint "配置路由：roll agent set <slot> <agent>（或从旧配置迁移）"
+
+# US-AGENT-026: `roll agent set` cascade picker.
+_i18n_set en agent.set_pick_slot "Pick a complexity slot to configure:"
+_i18n_set zh agent.set_pick_slot "选择要配置的复杂度档："
+_i18n_set en agent.set_enter_number "Enter number"
+_i18n_set zh agent.set_enter_number "输入编号"
+_i18n_set en agent.set_no_input "No input received — aborting."
+_i18n_set zh agent.set_no_input "未收到输入 — 已取消。"
+_i18n_set en agent.set_invalid_choice "Invalid choice: %s"
+_i18n_set zh agent.set_invalid_choice "无效选择：%s"
+_i18n_set en agent.set_unknown_slot "Unknown slot '%s' (expected easy|default|hard|fallback)"
+_i18n_set zh agent.set_unknown_slot "未知槽位 '%s'（应为 easy|default|hard|fallback）"
+_i18n_set en agent.set_unknown_agent "Unknown agent '%s' — run: roll agent list"
+_i18n_set zh agent.set_unknown_agent "未知 agent '%s' — 运行：roll agent list"
+_i18n_set en agent.set_no_online_agents "No installed agent is online right now — nothing to pick from."
+_i18n_set zh agent.set_no_online_agents "当前没有在线的已装 agent — 无可选项。"
+_i18n_set en agent.set_write_failed "Could not write slot '%s' to agents.yaml"
+_i18n_set zh agent.set_write_failed "无法将槽位 '%s' 写入 agents.yaml"
+_i18n_set en agent.set_saved "%s → %s  saved"
+_i18n_set zh agent.set_saved "%s → %s  已保存"
+
+# US-AGENT-027: `roll agent use <name>` — lock easy/default/hard to one agent.
+_i18n_set en agent.use_usage "Usage: roll agent use <name>   (locks easy/default/hard to one agent)"
+_i18n_set zh agent.use_usage "用法：roll agent use <name>（把 easy/default/hard 三档一次锁成同一 agent）"
+_i18n_set en agent.use_unknown_agent "Unknown or uninstalled agent '%s' — run: roll agent list"
+_i18n_set zh agent.use_unknown_agent "未知或未安装的 agent '%s' — 运行：roll agent list"
+_i18n_set en agent.use_locked "easy/default/hard all locked to %s  (fallback unchanged)"
+_i18n_set zh agent.use_locked "已将 easy/default/hard 三档锁为 %s（fallback 不变）"

package/lib/i18n/init.sh

@@ -53,6 +53,28 @@ _i18n_set zh init.syncing_conventions_to_ai_tools "正在同步约定到 AI 工
 _i18n_set en init.onboard_apply_complete_onboard "Onboard apply complete.  Onboard"
 _i18n_set zh init.onboard_apply_complete_onboard "应用完成。"
 
+# US-ONBOARD-017: render the three analysis sections + gated BACKLOG seeding.
+_i18n_set en init.onboard_rendered "Rendered: %s"
+_i18n_set zh init.onboard_rendered "已生成：%s"
+_i18n_set en init.onboard_render_failed "Could not render analysis markdown (skipping); onboard continues."
+_i18n_set zh init.onboard_render_failed "分析 markdown 生成失败（已跳过），onboard 继续。"
+_i18n_set en init.onboard_seed_preview_story "About to seed %s candidate stories to BACKLOG:"
+_i18n_set zh init.onboard_seed_preview_story "准备向 BACKLOG 播种 %s 条候选 story："
+_i18n_set en init.onboard_seed_preview_fix "About to seed %s HIGH-severity risks as FIX entries:"
+_i18n_set zh init.onboard_seed_preview_fix "准备将 %s 条高严重度风险播种为 FIX 条目："
+_i18n_set en init.onboard_seed_prompt "Seed these to BACKLOG?"
+_i18n_set zh init.onboard_seed_prompt "把这些播种到 BACKLOG 吗？"
+_i18n_set en init.onboard_seed_cancelled "Seeding cancelled. The analysis markdown was still generated."
+_i18n_set zh init.onboard_seed_cancelled "已取消播种。分析 markdown 仍已生成。"
+_i18n_set en init.onboard_seed_noninteractive "Non-interactive stdin — skipping BACKLOG seeding (markdown still generated)."
+_i18n_set zh init.onboard_seed_noninteractive "stdin 非交互 —— 跳过 BACKLOG 播种（markdown 仍已生成）。"
+_i18n_set en init.onboard_seeded_stories "Seeded %s candidate stories to BACKLOG."
+_i18n_set zh init.onboard_seeded_stories "已向 BACKLOG 播种 %s 条候选 story。"
+_i18n_set en init.onboard_seeded_fixes "Seeded %s candidate FIX entries to BACKLOG."
+_i18n_set zh init.onboard_seeded_fixes "已向 BACKLOG 播种 %s 条候选 FIX。"
+_i18n_set en init.onboard_seed_no_backlog "No .roll/backlog.md (backlog scope not approved) — skipping seeding."
+_i18n_set zh init.onboard_seed_no_backlog "无 .roll/backlog.md（未批准 backlog scope）—— 跳过播种。"
+
 _i18n_set en init.no_ai_agent_detected_install_one "No AI agent detected. Install one (e.g., claude, codex, kimi) and try again."
 _i18n_set zh init.no_ai_agent_detected_install_one "未检测到 AI agent。请先安装 (如 claude / codex / kimi) 后重试。"
 _i18n_set en init.the_process_will_use_your_agent "The process will use your agent to call models. Token cost is on your own account."

package/lib/i18n/peer.sh

@@ -32,3 +32,10 @@ _i18n_set en peer.en_peer_review "[EN:启动 peer review: %s → %s (第 %s 轮,
 _i18n_set zh peer.en_peer_review "启动 peer review: %s → %s (第 %s 轮, tag: %s)"
 _i18n_set en peer.en_enter_n "[EN:按 Enter 执行或输入 n 取消。%s 秒后自动执行......]"
 _i18n_set zh peer.en_enter_n "按 Enter 执行或输入 n 取消。%s 秒后自动执行..."
+
+_i18n_set en peer.no_peer_runs_yet "No peer review runs yet."
+_i18n_set zh peer.no_peer_runs_yet "还没有 peer review 记录。"
+_i18n_set en peer.no_peer_logs_found "No peer logs found."
+_i18n_set zh peer.no_peer_logs_found "还没有 peer 日志。"
+_i18n_set en peer.jq_required_for_roll_peer_runs "jq is required for 'roll peer runs'."
+_i18n_set zh peer.jq_required_for_roll_peer_runs "'roll peer runs' 需要安装 jq。"

package/lib/i18n/peer_help.sh

@@ -15,6 +15,10 @@ _i18n_set en peer_help.yes_yolo_skip_opt_out_prompt "  --yes, --yolo       Skip
 _i18n_set zh peer_help.yes_yolo_skip_opt_out_prompt "跳过确认提示"
 _i18n_set en peer_help.status_show_peer_review_state "  status              Show peer review state"
 _i18n_set zh peer_help.status_show_peer_review_state "显示状态"
+_i18n_set en peer_help.log_show_latest_peer_transcript "  log                 Show latest peer review transcript"
+_i18n_set zh peer_help.log_show_latest_peer_transcript "查看最新 peer 日志"
+_i18n_set en peer_help.runs_show_recent_peer_review_runs "  runs [N]            Show recent peer review runs"
+_i18n_set zh peer_help.runs_show_recent_peer_review_runs "查看最近 peer review 记录"
 _i18n_set en peer_help.reset_pair_all_reset_peer_state "  reset <pair|--all>  Reset peer state"
 _i18n_set zh peer_help.reset_pair_all_reset_peer_state "重置状态"
 _i18n_set en peer_help.help_show_this_help "  help                Show this help"

package/lib/i18n/skills_catalog.sh

@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+# Roll i18n catalog — `roll skills` command (US-SKILL-016).
+# Generated skills catalog: scan skills/*/SKILL.md frontmatter → guide/skills.md.
+
+_i18n_set en skills.generated "Generated skill catalog: %s"
+_i18n_set zh skills.generated "已生成技能清单：%s"
+
+_i18n_set en skills.check_ok "Skill catalog is up to date."
+_i18n_set zh skills.check_ok "技能清单已是最新。"
+
+_i18n_set en skills.check_drift "Skill catalog drift: %s differs from a fresh scan. Run 'roll skills generate'."
+_i18n_set zh skills.check_drift "技能清单漂移：%s 与最新扫描不一致。请运行 'roll skills generate'。"
+
+_i18n_set en skills.check_missing "Skill catalog not found at %s. Run 'roll skills generate'."
+_i18n_set zh skills.check_missing "未找到技能清单 %s。请运行 'roll skills generate'。"
+
+_i18n_set en skills.unknown_sub "Unknown 'roll skills' subcommand: %s"
+_i18n_set zh skills.unknown_sub "未知的 'roll skills' 子命令：%s"
+
+_i18n_set en skills.usage "Usage: roll skills <generate|check>"
+_i18n_set zh skills.usage "用法：roll skills <generate|check>"
+
+_i18n_set en skills.doctor_heading "Skill catalog"
+_i18n_set zh skills.doctor_heading "技能清单"
+
+_i18n_set en skills.doctor_ok "✅ guide/skills.md matches skills/*/SKILL.md"
+_i18n_set zh skills.doctor_ok "✅ guide/skills.md 与 skills/*/SKILL.md 一致"
+
+_i18n_set en skills.doctor_drift "⚠️  guide/skills.md is stale — run 'roll skills generate'"
+_i18n_set zh skills.doctor_drift "⚠️  guide/skills.md 已过期 — 请运行 'roll skills generate'"

package/lib/loop-exit-summary.py

@@ -0,0 +1,393 @@
+#!/usr/bin/env python3
+"""Render a one-cycle exit summary block for the loop's .command window (US-LOOP-040).
+
+When a `roll loop` cycle ends, the macOS `.command` Terminal window that was
+attached to the tmux session is left showing only a `press enter to close`
+prompt. The full Cycle Phase Breakdown / runs.jsonl data already exists on disk
+but the user has to scroll back through tmux scrollback or open
+``~/.shared/roll/loop/cron-<slug>.log`` to see what the cycle did.
+
+This helper renders a compact ``─── Cycle <CYCLE_ID> Summary ───`` block to
+stdout, consumed by the ``.command`` shell *before* the `press enter` prompt.
+It is a pure read-side view: it never writes new files and never mutates loop
+state.
+
+Five signals (per the US-LOOP-040 issue):
+
+  1. result   — runs.jsonl latest row's ``status`` + ``built[]`` + ``tcr_count``
+                (idle cycle → ``idle: no story picked``)
+  2. ci       — newest ``ci`` event outcome from events.ndjson tail
+                (``ok``→green, ``red``→red, ``heal-attempting`` passthrough,
+                no event → ``ci: n/a``)
+  3. todo     — count of ``📋 Todo`` lines in .roll/backlog.md
+  4. phases   — runs.jsonl ``phases`` map, top 5 by duration desc
+  5. alerts   — raw failure / alert text placeholder
+
+US-LOOP-041 layers failure / alert *highlighting* on top of the US-LOOP-040
+renderer. The relevant signal lines are flagged with a severity prefix and
+ANSI colour:
+
+  * RED + ``✗`` — runs.jsonl ``status`` is ``failed`` / ``aborted``; the latest
+    ``ci`` outcome is ``red``; or the events tail has a ``cycle_end`` whose
+    outcome is not ``ok`` / ``idle``.
+  * YELLOW + ``⚠`` — latest ``ci`` outcome is ``heal-attempting``; an
+    ``ALERT-<slug>.md`` exists and is non-empty; or ``tcr_count == 0`` while
+    ``built[]`` is non-empty (suspected zero-diff).
+  * default colour, no prefix — a fully green cycle (built/idle + ci green +
+    no alert).
+
+ANSI escapes are only emitted when stdout is a TTY (``sys.stdout.isatty()``)
+and ``NO_COLOR`` is unset (see https://no-color.org). Pipes, redirects and
+test captures get plain text — no escape codes are written.
+
+Data-source priority: runs.jsonl latest matching row > events.ndjson tail >
+fall back to the cron log's last 30 lines. Any missing source degrades
+silently — the renderer never errors and never blocks `press enter`.
+
+When no usable data exists at all (idle/aborted early-exit, runs.jsonl not yet
+flushed) it prints a single
+``(summary unavailable — see log: <cron-log>)`` line instead.
+
+Invocation::
+
+    python3 loop-exit-summary.py \
+        --runs   <runs.jsonl> \
+        --events <events.ndjson> \
+        --backlog <.roll/backlog.md> \
+        --cron-log <cron-<slug>.log> \
+        --alert  <ALERT-<slug>.md> \
+        [--cycle-id <id>] [--color {auto,always,never}]
+
+All paths are optional; a missing / unreadable file is treated as absent.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+from typing import Any, Dict, List, Optional
+
+
+def _read_last_json_line(path: Optional[str], cycle_id: str = "") -> Optional[Dict[str, Any]]:
+    """Return the last well-formed JSON object from a .jsonl file.
+
+    When ``cycle_id`` is given, prefer the last row whose ``cycle_id`` matches;
+    otherwise fall back to the last parseable row. Returns None when the file
+    is absent, empty, or has no parseable rows.
+    """
+    if not path or not os.path.isfile(path):
+        return None
+    last: Optional[Dict[str, Any]] = None
+    matched: Optional[Dict[str, Any]] = None
+    try:
+        with open(path, "r", encoding="utf-8", errors="replace") as fh:
+            for line in fh:
+                line = line.strip()
+                if not line:
+                    continue
+                try:
+                    obj = json.loads(line)
+                except (ValueError, TypeError):
+                    continue
+                if not isinstance(obj, dict):
+                    continue
+                last = obj
+                if cycle_id and obj.get("cycle_id") == cycle_id:
+                    matched = obj
+    except OSError:
+        return None
+    return matched if matched is not None else last
+
+
+def _read_json_lines(path: Optional[str]) -> List[Dict[str, Any]]:
+    """Return all well-formed JSON objects from a .ndjson file (in order)."""
+    out: List[Dict[str, Any]] = []
+    if not path or not os.path.isfile(path):
+        return out
+    try:
+        with open(path, "r", encoding="utf-8", errors="replace") as fh:
+            for line in fh:
+                line = line.strip()
+                if not line:
+                    continue
+                try:
+                    obj = json.loads(line)
+                except (ValueError, TypeError):
+                    continue
+                if isinstance(obj, dict):
+                    out.append(obj)
+    except OSError:
+        return out
+    return out
+
+
+def _latest_ci_outcome(events: List[Dict[str, Any]]) -> Optional[str]:
+    """Newest ``ci`` event outcome from an events stream, or None."""
+    for ev in reversed(events):
+        if ev.get("stage") == "ci":
+            outcome = ev.get("outcome")
+            if outcome:
+                return str(outcome)
+    return None
+
+
+def _latest_cycle_end_outcome(events: List[Dict[str, Any]]) -> Optional[str]:
+    """Newest ``cycle_end`` event outcome from an events stream, or None."""
+    for ev in reversed(events):
+        if ev.get("stage") == "cycle_end":
+            outcome = ev.get("outcome")
+            if outcome:
+                return str(outcome)
+    return None
+
+
+def _alert_active(path: Optional[str]) -> bool:
+    """True when an ALERT-<slug>.md file exists and has non-whitespace content."""
+    if not path or not os.path.isfile(path):
+        return False
+    try:
+        with open(path, "r", encoding="utf-8", errors="replace") as fh:
+            return bool(fh.read().strip())
+    except OSError:
+        return False
+
+
+# ── ANSI colouring (US-LOOP-041) ─────────────────────────────────────────────
+# Severity ranks: 0 = none/green, 1 = warn (yellow), 2 = fail (red).
+_SEV_NONE = 0
+_SEV_WARN = 1
+_SEV_FAIL = 2
+
+_ANSI = {_SEV_WARN: "\033[33m", _SEV_FAIL: "\033[31m"}
+_ANSI_RESET = "\033[0m"
+_PREFIX = {_SEV_NONE: "", _SEV_WARN: "⚠ ", _SEV_FAIL: "✗ "}
+
+
+def _color_enabled(mode: str) -> bool:
+    """Decide whether ANSI escapes should be emitted.
+
+    ``always`` forces colour, ``never`` forces plain text, ``auto`` (default)
+    honours NO_COLOR (https://no-color.org) and only colours a real TTY.
+    """
+    if mode == "always":
+        return True
+    if mode == "never":
+        return False
+    if os.environ.get("NO_COLOR") is not None:
+        return False
+    try:
+        return bool(sys.stdout.isatty())
+    except (ValueError, AttributeError):
+        return False
+
+
+def _decorate(text: str, sev: int, color: bool) -> str:
+    """Apply severity prefix + (optional) ANSI colour to a single line.
+
+    The leading indentation is preserved; the prefix and colour wrap only the
+    non-indented payload so columns still line up.
+    """
+    if sev == _SEV_NONE:
+        return text
+    stripped = text.lstrip(" ")
+    indent = text[: len(text) - len(stripped)]
+    payload = _PREFIX[sev] + stripped
+    if color:
+        payload = _ANSI[sev] + payload + _ANSI_RESET
+    return indent + payload
+
+
+def _count_todo(path: Optional[str]) -> Optional[int]:
+    """Count lines bearing the 📋 Todo marker in backlog.md. None if absent."""
+    if not path or not os.path.isfile(path):
+        return None
+    count = 0
+    try:
+        with open(path, "r", encoding="utf-8", errors="replace") as fh:
+            for line in fh:
+                if "📋" in line and "Todo" in line:
+                    count += 1
+    except OSError:
+        return None
+    return count
+
+
+def _tail_lines(path: Optional[str], n: int) -> List[str]:
+    """Last ``n`` non-empty lines of a text file, or [] when absent."""
+    if not path or not os.path.isfile(path):
+        return []
+    try:
+        with open(path, "r", encoding="utf-8", errors="replace") as fh:
+            lines = [ln.rstrip("\n") for ln in fh]
+    except OSError:
+        return []
+    return lines[-n:]
+
+
+def _fmt_ci(outcome: Optional[str]) -> str:
+    if outcome is None:
+        return "ci: n/a"
+    mapping = {"ok": "green", "green": "green", "red": "red",
+               "heal-attempting": "heal-attempting"}
+    return "ci: " + mapping.get(outcome, outcome)
+
+
+def _fmt_result(row: Dict[str, Any]) -> str:
+    status = row.get("status", "")
+    built = row.get("built") or []
+    tcr = row.get("tcr_count", 0)
+    if status == "idle" or (not built and status in ("", "idle")):
+        return "idle: no story picked"
+    if built:
+        built_str = " ".join(str(b) for b in built)
+        return "built: {0} · tcr commits: {1}".format(built_str, tcr)
+    # non-idle terminal with no built[] (failed/aborted/blocked/orphan)
+    return "{0} · tcr commits: {1}".format(status or "unknown", tcr)
+
+
+def _fmt_phases(phases: Dict[str, Any], limit: int = 5) -> List[str]:
+    rows: List[tuple] = []
+    for name, dur in phases.items():
+        try:
+            rows.append((int(dur), str(name)))
+        except (ValueError, TypeError):
+            continue
+    rows.sort(key=lambda r: (-r[0], r[1]))
+    out = []
+    for dur, name in rows[:limit]:
+        out.append("  {0:<22} {1:>5}s".format(name, dur))
+    return out
+
+
+def _result_severity(row: Dict[str, Any]) -> int:
+    """Severity for the result line (US-LOOP-041)."""
+    status = str(row.get("status", ""))
+    if status in ("failed", "aborted"):
+        return _SEV_FAIL
+    built = row.get("built") or []
+    tcr = row.get("tcr_count", 0)
+    if built and tcr == 0:  # suspected zero-diff: built something but no commit
+        return _SEV_WARN
+    return _SEV_NONE
+
+
+def _ci_severity(outcome: Optional[str]) -> int:
+    """Severity for the ci line (US-LOOP-041)."""
+    if outcome == "red":
+        return _SEV_FAIL
+    if outcome == "heal-attempting":
+        return _SEV_WARN
+    return _SEV_NONE
+
+
+def render(runs: Optional[str], events: Optional[str], backlog: Optional[str],
+           cron_log: Optional[str], cycle_id: str = "",
+           alert: Optional[str] = None, color: bool = False) -> str:
+    """Build the summary block as a string.
+
+    ``color`` toggles ANSI escapes; severity prefixes (``✗`` / ``⚠``) are
+    always applied to flagged lines regardless of ``color`` (US-LOOP-041).
+    """
+    row = _read_last_json_line(runs, cycle_id)
+    ev_list = _read_json_lines(events)
+    ci_outcome = _latest_ci_outcome(ev_list)
+    cycle_end_outcome = _latest_cycle_end_outcome(ev_list)
+    alert_on = _alert_active(alert)
+    todo = _count_todo(backlog)
+
+    # Source priority: a usable runs.jsonl row is the primary feed. With no
+    # row AND no events, fall back to the cron log's tail; if even that is
+    # empty, emit the single "unavailable" placeholder line.
+    have_primary = row is not None
+    have_events = bool(ev_list)
+
+    cid = cycle_id or (row.get("cycle_id") if row else "") or "unknown"
+    lines: List[str] = []
+    title = "─── Cycle {0} Summary ───".format(cid)
+
+    if not have_primary and not have_events:
+        tail = _tail_lines(cron_log, 30)
+        if not tail:
+            log_hint = cron_log or "~/.shared/roll/loop/cron-<slug>.log"
+            return "(summary unavailable — see log: {0})".format(log_hint)
+        # Degraded view: header + raw cron tail so the user still sees output.
+        lines.append(title)
+        lines.append("  (runs.jsonl + events unavailable — showing cron log tail)")
+        for ln in tail:
+            lines.append("  " + ln)
+        return "\n".join(lines)
+
+    lines.append(title)
+
+    # cycle_end fail severity applies to the result line when it's worse than
+    # what the runs.jsonl status alone implies.
+    cycle_end_sev = _SEV_NONE
+    if cycle_end_outcome is not None and cycle_end_outcome not in ("ok", "idle"):
+        cycle_end_sev = _SEV_FAIL
+
+    # 1. result
+    if row is not None:
+        result_sev = max(_result_severity(row), cycle_end_sev)
+        lines.append(_decorate("  " + _fmt_result(row), result_sev, color))
+    else:
+        lines.append(_decorate("  result: n/a", cycle_end_sev, color))
+
+    # 2. ci
+    lines.append(_decorate("  " + _fmt_ci(ci_outcome),
+                           _ci_severity(ci_outcome), color))
+
+    # 3. todo
+    if todo is not None:
+        lines.append("  todo remaining: {0}".format(todo))
+    else:
+        lines.append("  todo remaining: n/a")
+
+    # 4. phases (top 5 by duration desc)
+    phases = (row.get("phases") if row else None) or {}
+    if isinstance(phases, dict) and phases:
+        phase_rows = _fmt_phases(phases)
+        if phase_rows:
+            lines.append("  phases (top 5 by time):")
+            lines.extend(phase_rows)
+
+    # 5. alerts / failure highlight (US-LOOP-041)
+    alerts = (row.get("alerts") if row else None) or []
+    if isinstance(alerts, list) and alerts:
+        lines.append(_decorate("  alerts:", _SEV_FAIL, color))
+        for a in alerts:
+            lines.append(_decorate("    " + str(a), _SEV_FAIL, color))
+    if alert_on:
+        lines.append(_decorate("  alert: ALERT file active — see log",
+                               _SEV_WARN, color))
+
+    return "\n".join(lines)
+
+
+def main(argv: Optional[List[str]] = None) -> int:
+    parser = argparse.ArgumentParser(
+        description="Render a loop cycle exit summary block.")
+    parser.add_argument("--runs", default=None, help="path to runs.jsonl")
+    parser.add_argument("--events", default=None, help="path to events.ndjson")
+    parser.add_argument("--backlog", default=None, help="path to .roll/backlog.md")
+    parser.add_argument("--cron-log", default=None, help="path to cron-<slug>.log")
+    parser.add_argument("--cycle-id", default="", help="cycle id to prefer")
+    parser.add_argument("--alert", default=None, help="path to ALERT-<slug>.md")
+    parser.add_argument("--color", choices=("auto", "always", "never"),
+                        default="auto", help="ANSI colour mode (default: auto)")
+    args = parser.parse_args(argv)
+
+    try:
+        color = _color_enabled(args.color)
+        out = render(args.runs, args.events, args.backlog,
+                     args.cron_log, args.cycle_id,
+                     alert=args.alert, color=color)
+    except Exception:  # noqa: BLE001 — silent fallback per AC: never error
+        return 0
+    sys.stdout.write(out + "\n")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())