Hermes-agent

hermes_cli/goals.py

@@ -0,0 +1,912 @@
+"""Persistent session goals — the Ralph loop for Hermes.
+
+A goal is a free-form user objective that stays active across turns. After
+each turn completes, a small judge call asks an auxiliary model "is this
+goal satisfied by the assistant's last response?". If not, Hermes feeds a
+continuation prompt back into the same session and keeps working until the
+goal is done, turn budget is exhausted, the user pauses/clears it, or the
+user sends a new message (which takes priority and pauses the goal loop).
+
+State is persisted in SessionDB's ``state_meta`` table keyed by
+``goal:<session_id>`` so ``/resume`` picks it up.
+
+Design notes / invariants:
+
+- The continuation prompt is just a normal user message appended to the
+  session via ``run_conversation``. No system-prompt mutation, no toolset
+  swap — prompt caching stays intact.
+- Judge failures are fail-OPEN: ``continue``. A broken judge must not wedge
+  progress; the turn budget is the backstop.
+- When a real user message arrives mid-loop it preempts the continuation
+  prompt and also pauses the goal loop for that turn (we still re-judge
+  after, so if the user's message happens to complete the goal the judge
+  will say ``done``).
+- This module has zero hard dependency on ``cli.HermesCLI`` or the gateway
+  runner — both wire the same ``GoalManager`` in.
+
+Nothing in this module touches the agent's system prompt or toolset.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import re
+import time
+from dataclasses import dataclass, field, asdict
+from datetime import datetime, timezone
+from typing import Any, Dict, List, Optional, Tuple
+
+logger = logging.getLogger(__name__)
+
+
+# ──────────────────────────────────────────────────────────────────────
+# Constants & defaults
+# ──────────────────────────────────────────────────────────────────────
+
+DEFAULT_MAX_TURNS = 20
+DEFAULT_JUDGE_TIMEOUT = 30.0
+# Judge output budget. The freeform judge returns a one-line JSON verdict, but
+# reasoning models (deepseek-v4, qwq, etc.) burn tokens on hidden reasoning
+# before emitting the visible JSON — and the first /goal turn's prompt is
+# larger than later turns, which pushes total reply length past tight caps.
+# 200 tokens (the original default) reliably truncated the JSON on reasoning
+# models, leaving '{"done": true, "reason": "The agent successfully' and
+# triggering the auto-pause. 4096 covers reasoning + verdict on every model
+# we've live-tested; override via auxiliary.goal_judge.max_tokens for
+# specifically constrained setups.
+DEFAULT_JUDGE_MAX_TOKENS = 4096
+# Cap how much of the last response + recent messages we send to the judge.
+_JUDGE_RESPONSE_SNIPPET_CHARS = 4000
+# After this many consecutive judge *parse* failures (empty output / non-JSON),
+# the loop auto-pauses and points the user at the goal_judge config. API /
+# transport errors do NOT count toward this — those are transient. This guards
+# against small models (e.g. deepseek-v4-flash) that cannot follow the strict
+# JSON reply contract; without it the loop runs until the turn budget is
+# exhausted with every reply shaped like `judge returned empty response` or
+# `judge reply was not JSON`.
+DEFAULT_MAX_CONSECUTIVE_PARSE_FAILURES = 3
+
+
+CONTINUATION_PROMPT_TEMPLATE = (
+    "[Continuing toward your standing goal]\n"
+    "Goal: {goal}\n\n"
+    "Continue working toward this goal. Take the next concrete step. "
+    "If you believe the goal is complete, state so explicitly and stop. "
+    "If you are blocked and need input from the user, say so clearly and stop."
+)
+
+# Used when the user has added one or more /subgoal criteria. Surfaced
+# to the agent verbatim so it sees what to target on the next turn,
+# and surfaced to the judge so the verdict considers them too.
+CONTINUATION_PROMPT_WITH_SUBGOALS_TEMPLATE = (
+    "[Continuing toward your standing goal]\n"
+    "Goal: {goal}\n\n"
+    "Additional criteria the user added mid-loop:\n"
+    "{subgoals_block}\n\n"
+    "Continue working toward the goal AND all additional criteria. Take "
+    "the next concrete step. If you believe the goal and every "
+    "additional criterion are complete, state so explicitly and stop. "
+    "If you are blocked and need input from the user, say so clearly "
+    "and stop."
+)
+
+
+JUDGE_SYSTEM_PROMPT = (
+    "You are a strict judge evaluating whether an autonomous agent has "
+    "achieved a user's stated goal. You receive the goal text and the "
+    "agent's most recent response. Your only job is to decide whether "
+    "the goal is fully satisfied based on that response.\n\n"
+    "A goal is DONE only when:\n"
+    "- The response explicitly confirms the goal was completed, OR\n"
+    "- The response clearly shows the final deliverable was produced, OR\n"
+    "- The response explains the goal is unachievable / blocked / needs "
+    "user input (treat this as DONE with reason describing the block).\n\n"
+    "Otherwise the goal is NOT done — CONTINUE.\n\n"
+    "Reply ONLY with a single JSON object on one line:\n"
+    '{\"done\": <true|false>, \"reason\": \"<one-sentence rationale>\"}'
+)
+
+
+JUDGE_USER_PROMPT_TEMPLATE = (
+    "Goal:\n{goal}\n\n"
+    "Agent's most recent response:\n{response}\n\n"
+    "Current time: {current_time}\n\n"
+    "Is the goal satisfied?"
+)
+
+# Used when the user has added /subgoal criteria. The judge must
+# evaluate ALL of them being met, not just the original goal.
+JUDGE_USER_PROMPT_WITH_SUBGOALS_TEMPLATE = (
+    "Goal:\n{goal}\n\n"
+    "Additional criteria the user added mid-loop (all must also be "
+    "satisfied for the goal to be DONE):\n{subgoals_block}\n\n"
+    "Agent's most recent response:\n{response}\n\n"
+    "Current time: {current_time}\n\n"
+    "Decision: For each numbered criterion above, find concrete "
+    "evidence in the agent's response that the criterion is "
+    "satisfied. Do not accept generic phrases like 'all requirements "
+    "met' or 'implying it was done' — require specific evidence (a "
+    "file contents excerpt, an output line, a command result). If "
+    "ANY criterion lacks specific evidence in the response, the goal "
+    "is NOT done — return CONTINUE.\n\n"
+    "Is the goal AND every additional criterion satisfied?"
+)
+
+
+# ──────────────────────────────────────────────────────────────────────
+# Dataclass
+# ──────────────────────────────────────────────────────────────────────
+
+
+@dataclass
+class GoalState:
+    """Serializable goal state stored per session."""
+
+    goal: str
+    status: str = "active"          # active | paused | done | cleared
+    turns_used: int = 0
+    max_turns: int = DEFAULT_MAX_TURNS
+    created_at: float = 0.0
+    last_turn_at: float = 0.0
+    last_verdict: Optional[str] = None        # "done" | "continue" | "skipped"
+    last_reason: Optional[str] = None
+    paused_reason: Optional[str] = None       # why we auto-paused (budget, etc.)
+    consecutive_parse_failures: int = 0       # judge-output parse failures in a row
+    # User-added criteria appended mid-loop via the /subgoal command.
+    # When non-empty the judge prompt and continuation prompt both
+    # include them so the agent works toward them and the judge factors
+    # them into the verdict. Backwards-compatible: defaults to empty so
+    # old state_meta rows load unchanged.
+    subgoals: List[str] = field(default_factory=list)
+
+    def to_json(self) -> str:
+        return json.dumps(asdict(self), ensure_ascii=False)
+
+    @classmethod
+    def from_json(cls, raw: str) -> "GoalState":
+        data = json.loads(raw)
+        raw_subgoals = data.get("subgoals") or []
+        subgoals: List[str] = []
+        if isinstance(raw_subgoals, list):
+            subgoals = [str(s).strip() for s in raw_subgoals if str(s).strip()]
+        return cls(
+            goal=data.get("goal", ""),
+            status=data.get("status", "active"),
+            turns_used=int(data.get("turns_used", 0) or 0),
+            max_turns=int(data.get("max_turns", DEFAULT_MAX_TURNS) or DEFAULT_MAX_TURNS),
+            created_at=float(data.get("created_at", 0.0) or 0.0),
+            last_turn_at=float(data.get("last_turn_at", 0.0) or 0.0),
+            last_verdict=data.get("last_verdict"),
+            last_reason=data.get("last_reason"),
+            paused_reason=data.get("paused_reason"),
+            consecutive_parse_failures=int(data.get("consecutive_parse_failures", 0) or 0),
+            subgoals=subgoals,
+        )
+
+    # --- subgoals helpers -------------------------------------------------
+
+    def render_subgoals_block(self) -> str:
+        """Render the subgoals as a numbered ``- N. text`` block. Empty
+        when no subgoals exist."""
+        if not self.subgoals:
+            return ""
+        return "\n".join(f"- {i}. {text}" for i, text in enumerate(self.subgoals, start=1))
+
+
+# ──────────────────────────────────────────────────────────────────────
+# Persistence (SessionDB state_meta)
+# ──────────────────────────────────────────────────────────────────────
+
+
+def _meta_key(session_id: str) -> str:
+    return f"goal:{session_id}"
+
+
+_DB_CACHE: Dict[str, Any] = {}
+
+
+def _get_session_db() -> Optional[Any]:
+    """Return a SessionDB instance for the current HERMES_HOME.
+
+    SessionDB has no built-in singleton, but opening a new connection per
+    /goal call would thrash the file. We cache one instance per
+    ``hermes_home`` path so profile switches still pick up the right DB.
+    Defensive against import/instantiation failures so tests and
+    non-standard launchers can still use the GoalManager.
+    """
+    try:
+        from hermes_constants import get_hermes_home
+        from hermes_state import SessionDB
+
+        home = str(get_hermes_home())
+    except Exception as exc:  # pragma: no cover
+        logger.debug("GoalManager: SessionDB bootstrap failed (%s)", exc)
+        return None
+
+    cached = _DB_CACHE.get(home)
+    if cached is not None:
+        return cached
+    try:
+        db = SessionDB()
+    except Exception as exc:  # pragma: no cover
+        logger.debug("GoalManager: SessionDB() raised (%s)", exc)
+        return None
+    _DB_CACHE[home] = db
+    return db
+
+
+def load_goal(session_id: str) -> Optional[GoalState]:
+    """Load the goal for a session, or None if none exists."""
+    if not session_id:
+        return None
+    db = _get_session_db()
+    if db is None:
+        return None
+    try:
+        raw = db.get_meta(_meta_key(session_id))
+    except Exception as exc:
+        logger.debug("GoalManager: get_meta failed: %s", exc)
+        return None
+    if not raw:
+        return None
+    try:
+        return GoalState.from_json(raw)
+    except Exception as exc:
+        logger.warning("GoalManager: could not parse stored goal for %s: %s", session_id, exc)
+        return None
+
+
+def save_goal(session_id: str, state: GoalState) -> None:
+    """Persist a goal to SessionDB. No-op if DB unavailable."""
+    if not session_id:
+        return
+    db = _get_session_db()
+    if db is None:
+        return
+    try:
+        db.set_meta(_meta_key(session_id), state.to_json())
+    except Exception as exc:
+        logger.debug("GoalManager: set_meta failed: %s", exc)
+
+
+def clear_goal(session_id: str) -> None:
+    """Mark a goal cleared in the DB (preserved for audit, status=cleared)."""
+    state = load_goal(session_id)
+    if state is None:
+        return
+    state.status = "cleared"
+    save_goal(session_id, state)
+
+
+# ──────────────────────────────────────────────────────────────────────
+# Judge
+# ──────────────────────────────────────────────────────────────────────
+
+
+def _truncate(text: str, limit: int) -> str:
+    if not text:
+        return ""
+    if len(text) <= limit:
+        return text
+    return text[:limit] + "… [truncated]"
+
+
+_JSON_OBJECT_RE = re.compile(r"\{.*?\}", re.DOTALL)
+
+
+def _goal_judge_max_tokens() -> int:
+    """Resolve auxiliary.goal_judge.max_tokens, falling back to the default.
+
+    ``load_config()`` is cached on the config file's (mtime, size), so calling
+    this once per judge turn is cheap. A non-positive or non-int value falls
+    back to the default rather than crashing the goal loop.
+    """
+    try:
+        from hermes_cli.config import load_config
+
+        cfg = load_config()
+        value = (
+            (cfg.get("auxiliary") or {})
+            .get("goal_judge", {})
+            .get("max_tokens", DEFAULT_JUDGE_MAX_TOKENS)
+        )
+        value = int(value)
+        if value > 0:
+            return value
+    except Exception:
+        pass
+    return DEFAULT_JUDGE_MAX_TOKENS
+
+
+def _parse_judge_response(raw: str) -> Tuple[bool, str, bool]:
+    """Parse the judge's reply. Fail-open to ``(False, "<reason>", parse_failed)``.
+
+    Returns ``(done, reason, parse_failed)``. ``parse_failed`` is True when the
+    judge returned output that couldn't be interpreted as the expected JSON
+    verdict (empty body, prose, malformed JSON). Callers use that flag to
+    auto-pause after N consecutive parse failures so a weak judge model
+    doesn't silently burn the turn budget.
+    """
+    if not raw:
+        return False, "judge returned empty response", True
+
+    text = raw.strip()
+
+    # Strip markdown code fences the model may wrap JSON in.
+    if text.startswith("```"):
+        text = text.strip("`")
+        # Peel off leading json/JSON/etc tag
+        nl = text.find("\n")
+        if nl != -1:
+            text = text[nl + 1:]
+
+    # First try: parse the whole blob.
+    data: Optional[Dict[str, Any]] = None
+    try:
+        data = json.loads(text)
+    except Exception:
+        # Second try: pull the first JSON object out.
+        match = _JSON_OBJECT_RE.search(text)
+        if match:
+            try:
+                data = json.loads(match.group(0))
+            except Exception:
+                data = None
+
+    if not isinstance(data, dict):
+        return False, f"judge reply was not JSON: {_truncate(raw, 200)!r}", True
+
+    done_val = data.get("done")
+    if isinstance(done_val, str):
+        done = done_val.strip().lower() in {"true", "yes", "1", "done"}
+    else:
+        done = bool(done_val)
+    reason = str(data.get("reason") or "").strip()
+    if not reason:
+        reason = "no reason provided"
+    return done, reason, False
+
+
+def judge_goal(
+    goal: str,
+    last_response: str,
+    *,
+    timeout: float = DEFAULT_JUDGE_TIMEOUT,
+    subgoals: Optional[List[str]] = None,
+) -> Tuple[str, str, bool]:
+    """Ask the auxiliary model whether the goal is satisfied.
+
+    Returns ``(verdict, reason, parse_failed)`` where verdict is ``"done"``,
+    ``"continue"``, or ``"skipped"`` (when the judge couldn't be reached).
+
+    ``parse_failed`` is True only when the judge call succeeded but its output
+    was unusable (empty or non-JSON). API/transport errors return False — they
+    are transient and should fail-open silently. Callers use this flag to
+    auto-pause after N consecutive parse failures (see
+    ``DEFAULT_MAX_CONSECUTIVE_PARSE_FAILURES``).
+
+    ``subgoals`` is an optional list of user-added criteria (from
+    ``/subgoal``) that the judge must also factor into its DONE/CONTINUE
+    decision. When non-empty the prompt switches to the with-subgoals
+    template; otherwise behavior is identical to the original judge.
+
+    This is deliberately fail-open: any error returns ``("continue", "...", False)``
+    so a broken judge doesn't wedge progress — the turn budget and the
+    consecutive-parse-failures auto-pause are the backstops.
+    """
+    if not goal.strip():
+        return "skipped", "empty goal", False
+    if not last_response.strip():
+        # No substantive reply this turn — almost certainly not done yet.
+        return "continue", "empty response (nothing to evaluate)", False
+
+    try:
+        from agent.auxiliary_client import get_auxiliary_extra_body, get_text_auxiliary_client
+    except Exception as exc:
+        logger.debug("goal judge: auxiliary client import failed: %s", exc)
+        return "continue", "auxiliary client unavailable", False
+
+    try:
+        client, model = get_text_auxiliary_client("goal_judge")
+    except Exception as exc:
+        logger.debug("goal judge: get_text_auxiliary_client failed: %s", exc)
+        return "continue", "auxiliary client unavailable", False
+
+    if client is None or not model:
+        return "continue", "no auxiliary client configured", False
+
+    # Build the prompt — pick the with-subgoals variant when applicable.
+    clean_subgoals = [s.strip() for s in (subgoals or []) if s and s.strip()]
+    current_time = datetime.now(tz=timezone.utc).astimezone().strftime("%Y-%m-%d %H:%M:%S %Z")
+    if clean_subgoals:
+        subgoals_block = "\n".join(
+            f"- {i}. {text}" for i, text in enumerate(clean_subgoals, start=1)
+        )
+        prompt = JUDGE_USER_PROMPT_WITH_SUBGOALS_TEMPLATE.format(
+            goal=_truncate(goal, 2000),
+            subgoals_block=_truncate(subgoals_block, 2000),
+            response=_truncate(last_response, _JUDGE_RESPONSE_SNIPPET_CHARS),
+            current_time=current_time,
+        )
+    else:
+        prompt = JUDGE_USER_PROMPT_TEMPLATE.format(
+            goal=_truncate(goal, 2000),
+            response=_truncate(last_response, _JUDGE_RESPONSE_SNIPPET_CHARS),
+            current_time=current_time,
+        )
+
+    try:
+        resp = client.chat.completions.create(
+            model=model,
+            messages=[
+                {"role": "system", "content": JUDGE_SYSTEM_PROMPT},
+                {"role": "user", "content": prompt},
+            ],
+            temperature=0,
+            max_tokens=_goal_judge_max_tokens(),
+            timeout=timeout,
+            extra_body=get_auxiliary_extra_body() or None,
+        )
+    except Exception as exc:
+        logger.info("goal judge: API call failed (%s) — falling through to continue", exc)
+        return "continue", f"judge error: {type(exc).__name__}", False
+
+    try:
+        raw = resp.choices[0].message.content or ""
+    except Exception:
+        raw = ""
+
+    done, reason, parse_failed = _parse_judge_response(raw)
+    verdict = "done" if done else "continue"
+    logger.info("goal judge: verdict=%s reason=%s", verdict, _truncate(reason, 120))
+    return verdict, reason, parse_failed
+
+
+# ──────────────────────────────────────────────────────────────────────
+# GoalManager — the orchestration surface CLI + gateway talk to
+# ──────────────────────────────────────────────────────────────────────
+
+
+class GoalManager:
+    """Per-session goal state + continuation decisions.
+
+    The CLI and gateway each hold one ``GoalManager`` per live session.
+
+    Methods:
+
+    - ``set(goal)`` — start a new standing goal.
+    - ``clear()`` — remove the active goal.
+    - ``pause()`` / ``resume()`` — explicit user controls.
+    - ``status()`` — printable one-liner.
+    - ``evaluate_after_turn(last_response)`` — call the judge, update state,
+      and return a decision dict the caller uses to drive the next turn.
+    - ``next_continuation_prompt()`` — the canonical user-role message to
+      feed back into ``run_conversation``.
+    """
+
+    def __init__(self, session_id: str, *, default_max_turns: int = DEFAULT_MAX_TURNS):
+        self.session_id = session_id
+        self.default_max_turns = int(default_max_turns or DEFAULT_MAX_TURNS)
+        self._state: Optional[GoalState] = load_goal(session_id)
+
+    # --- introspection ------------------------------------------------
+
+    @property
+    def state(self) -> Optional[GoalState]:
+        return self._state
+
+    def is_active(self) -> bool:
+        return self._state is not None and self._state.status == "active"
+
+    def has_goal(self) -> bool:
+        return self._state is not None and self._state.status in {"active", "paused"}
+
+    def status_line(self) -> str:
+        s = self._state
+        if s is None or s.status in {"cleared",}:
+            return "No active goal. Set one with /goal <text>."
+        turns = f"{s.turns_used}/{s.max_turns} turns"
+        sub = f", {len(s.subgoals)} subgoal{'s' if len(s.subgoals) != 1 else ''}" if s.subgoals else ""
+        if s.status == "active":
+            return f"⊙ Goal (active, {turns}{sub}): {s.goal}"
+        if s.status == "paused":
+            extra = f" — {s.paused_reason}" if s.paused_reason else ""
+            return f"⏸ Goal (paused, {turns}{sub}{extra}): {s.goal}"
+        if s.status == "done":
+            return f"✓ Goal done ({turns}{sub}): {s.goal}"
+        return f"Goal ({s.status}, {turns}{sub}): {s.goal}"
+
+    # --- mutation -----------------------------------------------------
+
+    def set(self, goal: str, *, max_turns: Optional[int] = None) -> GoalState:
+        goal = (goal or "").strip()
+        if not goal:
+            raise ValueError("goal text is empty")
+        state = GoalState(
+            goal=goal,
+            status="active",
+            turns_used=0,
+            max_turns=int(max_turns) if max_turns else self.default_max_turns,
+            created_at=time.time(),
+            last_turn_at=0.0,
+        )
+        self._state = state
+        save_goal(self.session_id, state)
+        return state
+
+    def pause(self, reason: str = "user-paused") -> Optional[GoalState]:
+        if not self._state:
+            return None
+        self._state.status = "paused"
+        self._state.paused_reason = reason
+        save_goal(self.session_id, self._state)
+        return self._state
+
+    def resume(self, *, reset_budget: bool = True) -> Optional[GoalState]:
+        if not self._state:
+            return None
+        self._state.status = "active"
+        self._state.paused_reason = None
+        if reset_budget:
+            self._state.turns_used = 0
+        save_goal(self.session_id, self._state)
+        return self._state
+
+    def clear(self) -> None:
+        if self._state is None:
+            return
+        self._state.status = "cleared"
+        save_goal(self.session_id, self._state)
+        self._state = None
+
+    def mark_done(self, reason: str) -> None:
+        if not self._state:
+            return
+        self._state.status = "done"
+        self._state.last_verdict = "done"
+        self._state.last_reason = reason
+        save_goal(self.session_id, self._state)
+
+    # --- /subgoal user controls ---------------------------------------
+
+    def add_subgoal(self, text: str) -> str:
+        """Append a user-added criterion to the active goal. Requires
+        ``has_goal()``; raises ``RuntimeError`` otherwise.
+
+        Returns the cleaned text so the caller can show it back to the user.
+        """
+        if self._state is None or not self.has_goal():
+            raise RuntimeError("no active goal")
+        text = (text or "").strip()
+        if not text:
+            raise ValueError("subgoal text is empty")
+        self._state.subgoals.append(text)
+        save_goal(self.session_id, self._state)
+        return text
+
+    def remove_subgoal(self, index_1based: int) -> str:
+        """Remove a subgoal by 1-based index. Returns the removed text."""
+        if self._state is None or not self.has_goal():
+            raise RuntimeError("no active goal")
+        idx = int(index_1based) - 1
+        if idx < 0 or idx >= len(self._state.subgoals):
+            raise IndexError(
+                f"index out of range (1..{len(self._state.subgoals)})"
+            )
+        removed = self._state.subgoals.pop(idx)
+        save_goal(self.session_id, self._state)
+        return removed
+
+    def clear_subgoals(self) -> int:
+        """Wipe all subgoals. Returns the previous count."""
+        if self._state is None or not self.has_goal():
+            raise RuntimeError("no active goal")
+        prev = len(self._state.subgoals)
+        self._state.subgoals = []
+        save_goal(self.session_id, self._state)
+        return prev
+
+    def render_subgoals(self) -> str:
+        """Public helper for the /subgoal slash command."""
+        if self._state is None:
+            return "(no active goal)"
+        if not self._state.subgoals:
+            return "(no subgoals — use /subgoal <text> to add criteria)"
+        return self._state.render_subgoals_block()
+
+    # --- the main entry point called after every turn -----------------
+
+    def evaluate_after_turn(
+        self,
+        last_response: str,
+        *,
+        user_initiated: bool = True,
+    ) -> Dict[str, Any]:
+        """Run the judge and update state. Return a decision dict.
+
+        ``user_initiated`` distinguishes a real user prompt (True) from a
+        continuation prompt we fed ourselves (False). Both increment
+        ``turns_used`` because both consume model budget.
+
+        Decision keys:
+          - ``status``: current goal status after update
+          - ``should_continue``: bool — caller should fire another turn
+          - ``continuation_prompt``: str or None
+          - ``verdict``: "done" | "continue" | "skipped" | "inactive"
+          - ``reason``: str
+          - ``message``: user-visible one-liner to print/send
+        """
+        state = self._state
+        if state is None or state.status != "active":
+            return {
+                "status": state.status if state else None,
+                "should_continue": False,
+                "continuation_prompt": None,
+                "verdict": "inactive",
+                "reason": "no active goal",
+                "message": "",
+            }
+
+        # Count the turn that just finished.
+        state.turns_used += 1
+        state.last_turn_at = time.time()
+
+        verdict, reason, parse_failed = judge_goal(
+            state.goal, last_response, subgoals=state.subgoals or None
+        )
+        state.last_verdict = verdict
+        state.last_reason = reason
+
+        # Track consecutive judge parse failures. Reset on any usable reply,
+        # including API / transport errors (parse_failed=False) so a flaky
+        # network doesn't trip the auto-pause meant for bad judge models.
+        if parse_failed:
+            state.consecutive_parse_failures += 1
+        else:
+            state.consecutive_parse_failures = 0
+
+        if verdict == "done":
+            state.status = "done"
+            save_goal(self.session_id, state)
+            return {
+                "status": "done",
+                "should_continue": False,
+                "continuation_prompt": None,
+                "verdict": "done",
+                "reason": reason,
+                "message": f"✓ Goal achieved: {reason}",
+            }
+
+        # Auto-pause when the judge model can't produce the expected JSON
+        # verdict N turns in a row. Points the user at the goal_judge config
+        # so they can route this side task to a model that follows the
+        # contract (e.g. google/gemini-3-flash-preview). Without this guard,
+        # weak judge models burn the entire turn budget returning prose or
+        # empty strings.
+        if state.consecutive_parse_failures >= DEFAULT_MAX_CONSECUTIVE_PARSE_FAILURES:
+            state.status = "paused"
+            state.paused_reason = (
+                f"judge model returned unparseable output {state.consecutive_parse_failures} turns in a row"
+            )
+            save_goal(self.session_id, state)
+            return {
+                "status": "paused",
+                "should_continue": False,
+                "continuation_prompt": None,
+                "verdict": "continue",
+                "reason": reason,
+                "message": (
+                    f"⏸ Goal paused — the judge model ({state.consecutive_parse_failures} turns) "
+                    "isn't returning the required JSON verdict. Route the judge to a stricter "
+                    "model in ~/.hermes/config.yaml:\n"
+                    "  auxiliary:\n"
+                    "    goal_judge:\n"
+                    "      provider: openrouter\n"
+                    "      model: google/gemini-3-flash-preview\n"
+                    "Then /goal resume to continue."
+                ),
+            }
+
+        if state.turns_used >= state.max_turns:
+            state.status = "paused"
+            state.paused_reason = f"turn budget exhausted ({state.turns_used}/{state.max_turns})"
+            save_goal(self.session_id, state)
+            return {
+                "status": "paused",
+                "should_continue": False,
+                "continuation_prompt": None,
+                "verdict": "continue",
+                "reason": reason,
+                "message": (
+                    f"⏸ Goal paused — {state.turns_used}/{state.max_turns} turns used. "
+                    "Use /goal resume to keep going, or /goal clear to stop."
+                ),
+            }
+
+        save_goal(self.session_id, state)
+        return {
+            "status": "active",
+            "should_continue": True,
+            "continuation_prompt": self.next_continuation_prompt(),
+            "verdict": "continue",
+            "reason": reason,
+            "message": (
+                f"↻ Continuing toward goal ({state.turns_used}/{state.max_turns}): {reason}"
+            ),
+        }
+
+    def next_continuation_prompt(self) -> Optional[str]:
+        if not self._state or self._state.status != "active":
+            return None
+        if self._state.subgoals:
+            return CONTINUATION_PROMPT_WITH_SUBGOALS_TEMPLATE.format(
+                goal=self._state.goal,
+                subgoals_block=self._state.render_subgoals_block(),
+            )
+        return CONTINUATION_PROMPT_TEMPLATE.format(goal=self._state.goal)
+
+
+# ──────────────────────────────────────────────────────────────────────
+# Kanban worker goal loop
+# ──────────────────────────────────────────────────────────────────────
+
+# Continuation prompt fed back to a kanban goal-mode worker that has not
+# yet completed/blocked its task. The card's own acceptance criteria are
+# the goal — the worker already has the full task body in its first turn,
+# so we keep this short and point it back at the lifecycle contract.
+KANBAN_GOAL_CONTINUATION_TEMPLATE = (
+    "[Continuing toward this kanban task — judge says it is not done yet]\n"
+    "Reason: {reason}\n\n"
+    "Take the next concrete step toward completing the task. When the work "
+    "is genuinely finished, call kanban_complete with a summary. If you are "
+    "blocked and need human input, call kanban_block with a reason. Do not "
+    "stop without calling one of them."
+)
+
+# Fed when the judge believes the work is done but the worker never called
+# kanban_complete / kanban_block. One explicit nudge to terminate the task
+# the right way before the loop gives up.
+KANBAN_GOAL_FINALIZE_TEMPLATE = (
+    "[The work looks complete, but the task is still open]\n"
+    "Reason: {reason}\n\n"
+    "If the task is genuinely done, call kanban_complete now with a short "
+    "summary of what you did. If something still blocks completion, call "
+    "kanban_block with the reason instead."
+)
+
+
+def run_kanban_goal_loop(
+    *,
+    task_id: str,
+    goal_text: str,
+    run_turn,
+    task_status_fn,
+    block_fn,
+    max_turns: int = DEFAULT_MAX_TURNS,
+    first_response: str = "",
+    log=None,
+) -> Dict[str, Any]:
+    """Drive a kanban worker through a Ralph-style goal loop.
+
+    The dispatcher spawns a goal-mode worker exactly like a normal worker
+    (``hermes -p <profile> chat -q "work kanban task <id>"``). The worker's
+    first turn has already run by the time this is called; ``first_response``
+    is that turn's reply. From here we:
+
+    1. Check whether the worker already terminated the task (called
+       ``kanban_complete`` / ``kanban_block``). If so, stop — nothing to do.
+    2. Otherwise judge the latest response against ``goal_text`` (the card's
+       title + body). ``continue`` → feed a continuation prompt and run
+       another turn IN THE SAME SESSION via ``run_turn``. ``done`` but the
+       task is still open → one explicit "call kanban_complete" nudge.
+    3. When the turn budget is exhausted and the worker still hasn't
+       terminated the task, ``block_fn`` is invoked so the card lands in a
+       sticky ``blocked`` state for human review (NOT a silent exit).
+
+    This function performs NO SessionDB persistence — a worker process is
+    ephemeral, so the turn budget lives in a local counter. It is fully
+    decoupled from the CLI for testability: callers inject ``run_turn``
+    (str -> str), ``task_status_fn`` (() -> str|None), and ``block_fn``
+    (reason: str -> None).
+
+    Returns a decision dict: ``{"outcome", "turns_used", "reason"}`` where
+    outcome is one of ``"completed_by_worker"``, ``"blocked_budget"``,
+    ``"blocked_by_worker"``, or ``"stopped"``.
+    """
+
+    def _log(msg: str) -> None:
+        if log is not None:
+            try:
+                log(msg)
+            except Exception:
+                pass
+
+    max_turns = int(max_turns or DEFAULT_MAX_TURNS)
+    if max_turns < 1:
+        max_turns = DEFAULT_MAX_TURNS
+
+    last_response = first_response or ""
+    # The first turn already consumed one unit of budget.
+    turns_used = 1
+    nudged_to_finalize = False
+
+    while True:
+        # Did the worker terminate the task itself this turn?
+        try:
+            status = task_status_fn()
+        except Exception as exc:
+            _log(f"kanban goal loop: status check failed ({exc}); stopping")
+            return {"outcome": "stopped", "turns_used": turns_used, "reason": "status check failed"}
+
+        if status == "done":
+            _log(f"kanban goal loop: task {task_id} completed by worker after {turns_used} turn(s)")
+            return {"outcome": "completed_by_worker", "turns_used": turns_used, "reason": "worker completed the task"}
+        if status == "blocked":
+            _log(f"kanban goal loop: task {task_id} blocked by worker after {turns_used} turn(s)")
+            return {"outcome": "blocked_by_worker", "turns_used": turns_used, "reason": "worker blocked the task"}
+        if status not in ("running", "ready"):
+            # Reclaimed / archived / unexpected — let the dispatcher own it.
+            _log(f"kanban goal loop: task {task_id} status={status!r}; stopping")
+            return {"outcome": "stopped", "turns_used": turns_used, "reason": f"status={status}"}
+
+        # Still open — judge whether the latest response satisfies the card.
+        verdict, reason, _parse_failed = judge_goal(goal_text, last_response)
+        _log(f"kanban goal loop: turn {turns_used}/{max_turns} verdict={verdict} reason={_truncate(reason, 120)}")
+
+        if verdict == "done":
+            if nudged_to_finalize:
+                # Already asked once to call kanban_complete and it still
+                # didn't — block for review rather than spin.
+                _log(f"kanban goal loop: task {task_id} judged done but worker won't finalize; blocking")
+                try:
+                    block_fn(
+                        f"Goal-mode worker's output looked complete but it never "
+                        f"called kanban_complete after a finalize nudge ({reason})."
+                    )
+                except Exception as exc:
+                    _log(f"kanban goal loop: block_fn failed ({exc})")
+                return {"outcome": "blocked_budget", "turns_used": turns_used, "reason": "judged done, never finalized"}
+            prompt = KANBAN_GOAL_FINALIZE_TEMPLATE.format(reason=_truncate(reason, 400))
+            nudged_to_finalize = True
+        else:
+            prompt = KANBAN_GOAL_CONTINUATION_TEMPLATE.format(reason=_truncate(reason, 400))
+
+        # Budget check BEFORE spending another turn.
+        if turns_used >= max_turns:
+            _log(f"kanban goal loop: task {task_id} exhausted {turns_used}/{max_turns} turns; blocking")
+            try:
+                block_fn(
+                    f"Goal-mode worker exhausted its turn budget "
+                    f"({turns_used}/{max_turns}) without completing the task. "
+                    f"Last judge verdict: {_truncate(reason, 300)}"
+                )
+            except Exception as exc:
+                _log(f"kanban goal loop: block_fn failed ({exc})")
+            return {"outcome": "blocked_budget", "turns_used": turns_used, "reason": "turn budget exhausted"}
+
+        # Run another turn in the same session.
+        try:
+            last_response = run_turn(prompt) or ""
+        except Exception as exc:
+            _log(f"kanban goal loop: run_turn failed ({exc}); stopping")
+            return {"outcome": "stopped", "turns_used": turns_used, "reason": f"run_turn error: {type(exc).__name__}"}
+        turns_used += 1
+
+
+__all__ = [
+    "GoalState",
+    "GoalManager",
+    "CONTINUATION_PROMPT_TEMPLATE",
+    "CONTINUATION_PROMPT_WITH_SUBGOALS_TEMPLATE",
+    "JUDGE_USER_PROMPT_TEMPLATE",
+    "JUDGE_USER_PROMPT_WITH_SUBGOALS_TEMPLATE",
+    "KANBAN_GOAL_CONTINUATION_TEMPLATE",
+    "KANBAN_GOAL_FINALIZE_TEMPLATE",
+    "DEFAULT_MAX_TURNS",
+    "load_goal",
+    "save_goal",
+    "clear_goal",
+    "judge_goal",
+    "run_kanban_goal_loop",
+]