feat(commands): /compact alias + preview flags for /compress (#3243 salvage)

cli.py

@@ -9314,9 +9314,11 @@ def _manual_compress(self, cmd_original: str = ""):
             return
 
         from hermes_cli.partial_compress import (
+            extract_compress_flags,
             parse_partial_compress_args,
             rejoin_compressed_head_and_tail,
             split_history_for_partial_compress,
+            summarize_compress_preview,
         )
 
         # Args after the command word (e.g. "/compress here 3" -> "here 3").
@@ -9326,9 +9328,42 @@ def _manual_compress(self, cmd_original: str = ""):
             if len(_parts) > 1:
                 raw_args = _parts[1].strip()
 
+        # Strip --preview/--dry-run/--aggressive before positional parsing
+        # so the flags coexist with 'here [N]' / focus-topic forms.
+        raw_args, preview, aggressive = extract_compress_flags(raw_args)
         partial, keep_last, focus_topic = parse_partial_compress_args(raw_args)
         focus_topic = focus_topic or ""
 
+        if aggressive:
+            # LLM-free hard truncation is not supported: it would need its
+            # own transcript-persistence path outside the guarded
+            # _compress_context rotation machinery. Surface that instead of
+            # silently mis-parsing the flag as a focus topic.
+            print("(._.) --aggressive is not supported; use '/compress here [N]' "
+                  "to keep only recent exchanges, or /undo to drop turns.")
+            if not preview:
+                return
+
+        if preview:
+            from agent.model_metadata import estimate_request_tokens_rough
+            _sys_prompt = getattr(self.agent, "_cached_system_prompt", "") or ""
+            _tools = getattr(self.agent, "tools", None) or None
+            approx_tokens = estimate_request_tokens_rough(
+                self.conversation_history,
+                system_prompt=_sys_prompt,
+                tools=_tools,
+            )
+            report = summarize_compress_preview(
+                self.conversation_history,
+                partial,
+                keep_last,
+                focus_topic or None,
+                approx_tokens,
+            )
+            for line in report["lines"]:
+                print(f"🗜️  {line}")
+            return
+
         original_count = len(self.conversation_history)
         with self._busy_command("Compressing context..."):
             try:

gateway/slash_commands.py

@@ -2899,13 +2899,12 @@ async def _handle_verbose_command(self, event: MessageEvent) -> str:
             return t("gateway.verbose.not_enabled")
 
         # --- cycle mode (per-platform) ----------------------------------------
-        cycle = ["off", "new", "all", "verbose", "log"]
+        cycle = ["off", "new", "all", "verbose"]
         descriptions = {
             "off": t("gateway.verbose.mode_off"),
             "new": t("gateway.verbose.mode_new"),
             "all": t("gateway.verbose.mode_all"),
             "verbose": t("gateway.verbose.mode_verbose"),
-            "log": t("gateway.verbose.mode_log"),
         }
 
         # Read current effective mode for this platform via the resolver
@@ -3044,13 +3043,44 @@ async def _handle_compress_command(self, event: MessageEvent) -> str:
         # Parse args: either a focus topic (full compress) or the
         # boundary-aware "here [N]" form (partial compress).
         from hermes_cli.partial_compress import (
+            extract_compress_flags,
             parse_partial_compress_args,
             rejoin_compressed_head_and_tail,
             split_history_for_partial_compress,
+            summarize_compress_preview,
         )
         _raw_args = (event.get_command_args() or "").strip()
+        # Strip --preview/--dry-run/--aggressive before positional parsing
+        # so the flags coexist with 'here [N]' / focus-topic forms.
+        _raw_args, _preview, _aggressive = extract_compress_flags(_raw_args)
         partial, keep_last, focus_topic = parse_partial_compress_args(_raw_args)
 
+        _agg_note = ""
+        if _aggressive:
+            # LLM-free hard truncation is not supported on this surface —
+            # it would need its own transcript-persistence branch outside
+            # the guarded _compress_context rotation machinery (#44794).
+            _agg_note = t("gateway.compress.aggressive_unsupported")
+            if not _preview:
+                return _agg_note
+
+        if _preview:
+            # Report what WOULD be compressed — no agent, no writes.
+            from agent.model_metadata import estimate_request_tokens_rough
+            _pv_msgs = [
+                {"role": m.get("role"), "content": m.get("content")}
+                for m in history
+                if m.get("role") in {"user", "assistant"} and m.get("content")
+            ]
+            approx_tokens = estimate_request_tokens_rough(_pv_msgs)
+            report = summarize_compress_preview(
+                _pv_msgs, partial, keep_last, focus_topic, approx_tokens
+            )
+            lines = [f"🗜️ {line}" for line in report["lines"]]
+            if _aggressive:
+                lines.append(_agg_note)
+            return "\n".join(lines)
+
         try:
             from run_agent import AIAgent
             from agent.manual_compression_feedback import summarize_manual_compression

hermes_cli/commands.py

@@ -88,8 +88,8 @@ class CommandDef:
                args_hint="<platform>", cli_only=True),
     CommandDef("branch", "Branch the current session (explore a different path)", "Session",
                aliases=("fork",), args_hint="[name]"),
-    CommandDef("compress", "Compress conversation context (add 'here [N]' to keep recent N turns)", "Session",
-               args_hint="[here [N] | focus topic]"),
+    CommandDef("compress", "Compress conversation context (add 'here [N]' to keep recent N turns; --preview shows what would happen)", "Session",
+               aliases=("compact",), args_hint="[here [N] | focus topic | --preview|--dry-run]"),
     CommandDef("rollback", "List or restore filesystem checkpoints", "Session",
                args_hint="[number]"),
     CommandDef("snapshot", "Create or restore state snapshots of Hermes config/state", "Session",
@@ -144,7 +144,7 @@ class CommandDef:
     CommandDef("timestamps", "Toggle [HH:MM] timestamps on messages and /history", "Configuration",
                cli_only=True, args_hint="[on|off|status]",
                subcommands=("on", "off", "status"), aliases=("ts",)),
-    CommandDef("verbose", "Cycle tool progress display: off -> new -> all -> verbose -> log",
+    CommandDef("verbose", "Cycle tool progress display: off -> new -> all -> verbose",
                "Configuration", cli_only=True,
                gateway_config_gate="display.tool_progress_command"),
     CommandDef("footer", "Toggle gateway runtime-metadata footer on final replies",

hermes_cli/partial_compress.py

@@ -108,6 +108,95 @@ def parse_partial_compress_args(
     return False, DEFAULT_KEEP_LAST, text or None
 
 
+def extract_compress_flags(raw_args: str) -> Tuple[str, bool, bool]:
+    """Strip ``--preview``/``--dry-run``/``--aggressive`` flags from the
+    argument string after ``/compress`` (or its ``/compact`` alias).
+
+    Flags may appear anywhere and coexist with the positional forms
+    (``here [N]``, ``--keep N``, or a focus topic); the returned
+    remainder is what :func:`parse_partial_compress_args` should see.
+
+    Returns ``(remaining_args, preview, aggressive_requested)``:
+
+    * ``preview`` — True when ``--preview`` or ``--dry-run`` was given.
+      The caller must report what WOULD be compressed (message counts,
+      token estimate, boundary) and make **no changes**.
+    * ``aggressive_requested`` — True when ``--aggressive`` was given.
+      The current surfaces do not implement an LLM-free hard-truncate
+      path (it would need its own transcript-persistence branch outside
+      the guarded ``_compress_context`` rotation machinery), so callers
+      surface a "not supported" note instead of silently treating the
+      flag as a focus topic.
+    """
+    preview = False
+    aggressive = False
+    kept: List[str] = []
+    for tok in (raw_args or "").split():
+        low = tok.lower()
+        if low in ("--preview", "--dry-run", "--dryrun"):
+            preview = True
+        elif low == "--aggressive":
+            aggressive = True
+        else:
+            kept.append(tok)
+    return " ".join(kept), preview, aggressive
+
+
+def summarize_compress_preview(
+    history: List[Dict[str, Any]],
+    partial: bool,
+    keep_last: int,
+    focus_topic: Optional[str],
+    approx_tokens: int,
+) -> Dict[str, Any]:
+    """Build the ``/compress --preview`` report — pure, no side effects.
+
+    Shared by the CLI (``cli.py::_manual_compress``) and the gateway
+    (``gateway/slash_commands.py::_handle_compress_command``) so both
+    surfaces report the same numbers the real run would use.
+
+    Returns a dict with ``head_count``/``tail_count``/``lines`` where
+    ``lines`` is a ready-to-print list of report strings.
+    """
+    total = len(history)
+    head = list(history)
+    tail: List[Dict[str, Any]] = []
+    effective_partial = partial
+    if partial:
+        head, tail = split_history_for_partial_compress(history, keep_last)
+        if not tail:
+            # Same degenerate-split fallback the real run applies.
+            effective_partial = False
+            head, tail = list(history), []
+
+    lines = [
+        "Preview — no changes made.",
+        f"Would compress {len(head)} of {total} message(s) "
+        f"(~{approx_tokens:,} tokens currently in context).",
+    ]
+    if effective_partial:
+        lines.append(
+            f"Boundary: keeping the last {keep_last} exchange(s) "
+            f"({len(tail)} message(s)) verbatim."
+        )
+    elif partial:
+        lines.append(
+            "Boundary: 'here' split would keep everything — "
+            "falling back to full compression."
+        )
+    if focus_topic:
+        lines.append(f'Focus topic: "{focus_topic}"')
+    lines.append("Run the command again without --preview to apply.")
+
+    return {
+        "head_count": len(head),
+        "tail_count": len(tail),
+        "total": total,
+        "partial": effective_partial,
+        "lines": lines,
+    }
+
+
 def _coerce_keep(value: str) -> int:
     """Parse a keep-count token, clamping to [1, MAX_KEEP_LAST]."""
     try:

locales/af.yaml

@@ -88,6 +88,7 @@ gateway:
     not_enough:            "Nie genoeg gesprek om saam te pers nie (ten minste 4 boodskappe nodig)."
     no_provider:           "Geen verskaffer opgestel nie -- kan nie saampers nie."
     nothing_to_do:         "Niks om saam te pers nie (die transkripsie is steeds heeltemal beskermde konteks)."
+    aggressive_unsupported: "--aggressive word nie ondersteun nie; gebruik '/compress here [N]' om net onlangse uitruilings te behou, of /undo om beurte te verwyder."
     focus_line:            "Fokus: \"{topic}\""
     summary_failed:        "⚠️ Opsomming kon nie gegenereer word nie ({error}). {count} historiese boodskap(pe) is verwyder en met 'n plekhouer vervang; vroeëre konteks kan nie meer herstel word nie. Oorweeg om jou auxiliary.compression-modelopstelling na te gaan."
     aborted:               "⚠️ Kompressie gestaak ({error}). Geen boodskappe is laat val nie — die gesprek is onveranderd. Voer /compress uit om weer te probeer, /reset vir 'n skoon sessie, of kyk na jou auxiliary.compression-modelkonfigurasie."
@@ -367,7 +368,6 @@ Future messages in this room will use that transcript until `/reset` or another
     mode_new:              "⚙️ Gereedskap-vordering: **NUUT** — vertoon wanneer gereedskap verander (voorskoulengte: `display.tool_preview_length`, verstek 40)."
     mode_all:              "⚙️ Gereedskap-vordering: **ALMAL** — elke gereedskaps-oproep vertoon (voorskoulengte: `display.tool_preview_length`, verstek 40)."
     mode_verbose:          "⚙️ Gereedskap-vordering: **OMSLAGTIG** — elke gereedskaps-oproep met volle argumente."
-    mode_log:              "⚙️ Gereedskap-vordering: **LOG** — stil in die klets; gereedskaps-oproepe word na ~/.hermes/logs/tool_calls.log geskryf."
     saved_suffix:          "_(gestoor vir **{platform}** — neem effek by die volgende boodskap)_"
     save_failed:           "_(kon nie in konfigurasie stoor nie: {error})_"
 

locales/de.yaml

@@ -88,6 +88,7 @@ gateway:
     not_enough:            "Nicht genug Konversation zum Komprimieren (mindestens 4 Nachrichten erforderlich)."
     no_provider:           "Kein Anbieter konfiguriert — Komprimierung nicht möglich."
     nothing_to_do:         "Noch nichts zu komprimieren (das Transkript ist weiterhin vollständig geschützter Kontext)."
+    aggressive_unsupported: "--aggressive wird nicht unterstützt; verwende '/compress here [N]', um nur die letzten Austausche zu behalten, oder /undo, um Beiträge zu entfernen."
     focus_line:            "Fokus: \"{topic}\""
     summary_failed:        "⚠️ Zusammenfassungsgenerierung fehlgeschlagen ({error}). {count} historische Nachricht(en) wurden entfernt und durch einen Platzhalter ersetzt; früherer Kontext ist nicht mehr wiederherstellbar. Überprüfen Sie die Konfiguration des auxiliary.compression-Modells."
     aborted:               "⚠️ Komprimierung abgebrochen ({error}). Keine Nachrichten wurden entfernt — die Konversation ist unverändert. Führe /compress aus, um es erneut zu versuchen, /reset für eine neue Sitzung, oder prüfe deine auxiliary.compression-Modellkonfiguration."
@@ -367,7 +368,6 @@ Future messages in this room will use that transcript until `/reset` or another
     mode_new:              "⚙️ Tool-Fortschritt: **NEW** — angezeigt bei Tool-Wechsel (Vorschaulänge: `display.tool_preview_length`, Standard 40)."
     mode_all:              "⚙️ Tool-Fortschritt: **ALL** — jeder Tool-Aufruf wird angezeigt (Vorschaulänge: `display.tool_preview_length`, Standard 40)."
     mode_verbose:          "⚙️ Tool-Fortschritt: **VERBOSE** — jeder Tool-Aufruf mit vollständigen Argumenten."
-    mode_log:              "⚙️ Tool-Fortschritt: **LOG** — still im Chat; Tool-Aufrufe werden in ~/.hermes/logs/tool_calls.log geschrieben."
     saved_suffix:          "_(für **{platform}** gespeichert — wird ab nächster Nachricht wirksam)_"
     save_failed:           "_(konnte nicht in der Konfiguration gespeichert werden: {error})_"
 

locales/en.yaml

@@ -103,6 +103,7 @@ gateway:
     not_enough:            "Not enough conversation to compress (need at least 4 messages)."
     no_provider:           "No provider configured -- cannot compress."
     nothing_to_do:         "Nothing to compress yet (the transcript is still all protected context)."
+    aggressive_unsupported: "--aggressive is not supported; use '/compress here [N]' to keep only recent exchanges, or /undo to drop turns."
     focus_line:            "Focus: \"{topic}\""
     summary_failed:        "⚠️ Summary generation failed ({error}). {count} historical message(s) were removed and replaced with a placeholder; earlier context is no longer recoverable. Consider checking your auxiliary.compression model configuration."
     aborted:               "⚠️ Compression aborted ({error}). No messages were dropped — conversation is unchanged. Run /compress to retry, /reset for a clean session, or check your auxiliary.compression model configuration."
@@ -379,7 +380,6 @@ gateway:
     mode_new:              "⚙️ Tool progress: **NEW** — shown when tool changes (preview length: `display.tool_preview_length`, default 40)."
     mode_all:              "⚙️ Tool progress: **ALL** — every tool call shown (preview length: `display.tool_preview_length`, default 40)."
     mode_verbose:          "⚙️ Tool progress: **VERBOSE** — every tool call with full arguments."
-    mode_log:              "⚙️ Tool progress: **LOG** — silent in chat; tool calls written to ~/.hermes/logs/tool_calls.log."
     saved_suffix:          "_(saved for **{platform}** — takes effect on next message)_"
     save_failed:           "_(could not save to config: {error})_"
 

locales/es.yaml

@@ -88,6 +88,7 @@ gateway:
     not_enough:            "No hay suficiente conversación para comprimir (se necesitan al menos 4 mensajes)."
     no_provider:           "No hay proveedor configurado — no se puede comprimir."
     nothing_to_do:         "Aún no hay nada que comprimir (la transcripción sigue siendo todo contexto protegido)."
+    aggressive_unsupported: "--aggressive no es compatible; usa '/compress here [N]' para conservar solo los intercambios recientes, o /undo para eliminar turnos."
     focus_line:            "Enfoque: \"{topic}\""
     summary_failed:        "⚠️ Falló la generación del resumen ({error}). Se eliminaron {count} mensaje(s) históricos y se reemplazaron por un marcador; el contexto anterior ya no se puede recuperar. Considera revisar la configuración del modelo auxiliary.compression."
     aborted:               "⚠️ Compresión abortada ({error}). No se eliminó ningún mensaje — la conversación está intacta. Ejecuta /compress para reintentar, /reset para una sesión limpia, o revisa la configuración de tu modelo auxiliary.compression."
@@ -364,7 +365,6 @@ gateway:
     mode_new:              "⚙️ Progreso de herramientas: **NEW** — se muestra al cambiar de herramienta (longitud de vista previa: `display.tool_preview_length`, por defecto 40)."
     mode_all:              "⚙️ Progreso de herramientas: **ALL** — se muestra cada llamada a herramienta (longitud de vista previa: `display.tool_preview_length`, por defecto 40)."
     mode_verbose:          "⚙️ Progreso de herramientas: **VERBOSE** — cada llamada a herramienta con sus argumentos completos."
-    mode_log:              "⚙️ Progreso de herramientas: **LOG** — silencioso en el chat; las llamadas a herramientas se escriben en ~/.hermes/logs/tool_calls.log."
     saved_suffix:          "_(guardado para **{platform}** — se aplica en el próximo mensaje)_"
     save_failed:           "_(no se pudo guardar en la configuración: {error})_"
 

locales/fr.yaml

@@ -88,6 +88,7 @@ gateway:
     not_enough:            "Conversation insuffisante pour la compression (au moins 4 messages nécessaires)."
     no_provider:           "Aucun fournisseur configuré — compression impossible."
     nothing_to_do:         "Rien à compresser pour l'instant (la transcription est encore entièrement du contexte protégé)."
+    aggressive_unsupported: "--aggressive n'est pas pris en charge ; utilisez '/compress here [N]' pour ne conserver que les échanges récents, ou /undo pour supprimer des tours."
     focus_line:            "Focus : \"{topic}\""
     summary_failed:        "⚠️ Échec de la génération du résumé ({error}). {count} message(s) historique(s) ont été supprimés et remplacés par un espace réservé ; le contexte antérieur n'est plus récupérable. Vérifiez la configuration du modèle auxiliary.compression."
     aborted:               "⚠️ Compression interrompue ({error}). Aucun message n'a été supprimé — la conversation est inchangée. Lancez /compress pour réessayer, /reset pour une nouvelle session, ou vérifiez la configuration de votre modèle auxiliary.compression."
@@ -367,7 +368,6 @@ Future messages in this room will use that transcript until `/reset` or another
     mode_new:              "⚙️ Progression des outils : **NEW** — affichée lors d'un changement d'outil (longueur d'aperçu : `display.tool_preview_length`, par défaut 40)."
     mode_all:              "⚙️ Progression des outils : **ALL** — chaque appel d'outil est affiché (longueur d'aperçu : `display.tool_preview_length`, par défaut 40)."
     mode_verbose:          "⚙️ Progression des outils : **VERBOSE** — chaque appel d'outil avec ses arguments complets."
-    mode_log:              "⚙️ Progression des outils : **LOG** — silencieux dans le chat ; les appels d'outils sont écrits dans ~/.hermes/logs/tool_calls.log."
     saved_suffix:          "_(enregistré pour **{platform}** — prend effet au prochain message)_"
     save_failed:           "_(impossible d'enregistrer dans la configuration : {error})_"