ouyangbo/hermes-agent
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

refactor(hermes_cli): AST-neutral closer hugging across r3-17 slice files

Browse Source
This commit is contained in:
Teknium
2026-09-02 23:58:31 -07:00
parent 1ceca3c326
commit 3494f7cf23
15 changed files with 182 additions and 369 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hermes_cli/build_info.py
+1 -2
View File
@@ -103,8 +103,7 @@ def get_code_identity(refresh: bool = False) -> dict:
"sha": sha,
"short_sha": sha[:8] if sha else None,
"version": version,
"source": source,
}
"source": source}
return dict(_code_identity_cache)
hermes_cli/bundles.py
+9 -19
View File
@@ -31,8 +31,7 @@ def _cmd_list(args) -> None:
c.print(
f"[dim]No bundles installed yet. Create one with:\n"
f" hermes bundles create <name> --skill skill1 --skill skill2[/]\n"
f"Bundles directory: [bold]{_bundles_dir()}[/]"
)
f"Bundles directory: [bold]{_bundles_dir()}[/]")
return
table = Table(title=f"Skill Bundles ({len(bundles)})", show_lines=False)
@@ -46,8 +45,7 @@ def _cmd_list(args) -> None:
f"/{info['slug']}",
info["name"],
str(len(info.get("skills", []))),
info.get("description") or "",
)
info.get("description") or "")
c.print(table)
c.print(f"\n[dim]Bundles directory: {_bundles_dir()}[/]")
@@ -76,8 +74,7 @@ def _cmd_create(args) -> None:
# Interactive prompt for skills if none were passed on the CLI.
c.print(
"[dim]No skills passed via --skill. Enter one skill name per line.\n"
"Submit an empty line to finish.[/]"
)
"Submit an empty line to finish.[/]")
try:
while True:
line = line_input("skill> ").strip()
@@ -92,8 +89,7 @@ def _cmd_create(args) -> None:
try:
path = save_bundle(
name, skills, description=args.description or "", instruction=args.instruction or "",
overwrite=bool(args.force),
)
overwrite=bool(args.force))
except FileExistsError as exc:
_fail(c, f"[bold red]{exc}[/]\n[dim]Pass --force to overwrite.[/]")
except ValueError as exc:
@@ -104,8 +100,7 @@ def _cmd_create(args) -> None:
if info:
c.print(
f" Invoke with: [bold cyan]/{info['slug']}[/] "
f"(loads {len(info['skills'])} skills)"
)
f"(loads {len(info['skills'])} skills)")
def _cmd_delete(args) -> None:
@@ -151,22 +146,17 @@ def register_cli(subparser) -> None:
help="Create a new skill bundle",
description=(
"Create a new bundle. Skills can be passed via --skill (repeat for "
"multiple) or entered interactively when omitted."
),
)
"multiple) or entered interactively when omitted."))
p_create.add_argument("name", help="Bundle name (becomes the /slash command)")
p_create.add_argument(
"--skill", "-s", action="append", default=[],
help="Skill name to include (repeat for multiple)",
)
help="Skill name to include (repeat for multiple)")
p_create.add_argument(
"--description", "-d", default="",
help="Human-readable description shown in /help and `hermes bundles list`",
)
help="Human-readable description shown in /help and `hermes bundles list`")
p_create.add_argument(
"--instruction", "-i", default="",
help="Extra guidance prepended to the loaded skill content",
)
help="Extra guidance prepended to the loaded skill content")
p_create.add_argument(
"--force", "-f", action="store_true", help="Overwrite an existing bundle with the same name"
)
hermes_cli/checkpoints.py
+1 -2
View File
@@ -109,8 +109,7 @@ def cmd_prune(args: argparse.Namespace) -> int:
retention_days=args.retention_days,
delete_orphans=delete_orphans,
max_total_size_mb=args.max_size_mb,
orphan_allowlist=orphan_allowlist,
)
orphan_allowlist=orphan_allowlist)
print(f"Scanned: {result['scanned']}")
print(f"Deleted orphan: {result['deleted_orphan']}")
print(f"Deleted stale: {result['deleted_stale']}")
hermes_cli/claw.py
+3 -6
View File
@@ -32,16 +32,14 @@ _OPENCLAW_DIR_NAMES = (".openclaw", ".clawdbot", ".moltbot")
_MIGRATE_ARG_DEFAULTS = (
("source", None), ("dry_run", False), ("preset", "full"), ("overwrite", False),
("migrate_secrets", False), ("workspace_target", None), ("skill_conflict", "skip"),
("no_backup", False), ("yes", False),
)
("no_backup", False), ("yes", False))
# (status, heading, color, default reason) — printed in this order after migrated items.
_REPORT_REASON_GROUPS = (
("conflict", " ⚠ Conflicts (skipped — use --overwrite to force):", Colors.YELLOW,
"already exists"),
("skipped", " ─ Skipped:", Colors.DIM, ""),
("error", " ✗ Errors:", Colors.RED, "unknown error"),
)
("error", " ✗ Errors:", Colors.RED, "unknown error"))
# Summary-line counters after the migrated count: (summary key, label).
_SUMMARY_COUNT_LABELS = (("conflict", "conflict(s)"), ("skipped", "skipped"), ("error", "error(s)"))
@@ -49,8 +47,7 @@ _SUMMARY_COUNT_LABELS = (("conflict", "conflict(s)"), ("skipped", "skipped"), ("
_WORKSPACE_MARKERS = ("todo.json", "SOUL.md", "MEMORY.md", "USER.md")
_WORKSPACE_ITEM_LABELS = (
("todo.json", "todo.json", Path.exists), ("sessions", "sessions/", Path.is_dir),
("SOUL.md", "SOUL.md", Path.exists), ("MEMORY.md", "MEMORY.md", Path.exists),
)
("SOUL.md", "SOUL.md", Path.exists), ("MEMORY.md", "MEMORY.md", Path.exists))
def _print_banner(title: str) -> None:
hermes_cli/cli_agent_setup_mixin.py
+2 -4
View File
@@ -93,8 +93,7 @@ def _tool_calls_summary(tool_calls) -> str:
_RESUME_EVENT_TEXT = {
"model_switch": "model changed",
"async_delegation_complete": "background delegation completed",
"auto_continue": "resumed interrupted turn",
}
"auto_continue": "resumed interrupted turn"}
def _collect_resume_entries(display_history, disp: dict, clean_assistant):
"""Displayable ``(role, text)`` recap entries from stored history, truncated per the
@@ -153,8 +152,7 @@ def _collect_resume_entries(display_history, disp: dict, clean_assistant):
# (skin key, fallback) for recap panel colors: body text, session label, border, assistant label.
_RESUME_SKIN_COLORS = (
("banner_text", "#FFF8DC"), ("session_label", "#DAA520"), ("session_border", "#8B8682"),
("ui_ok", "#8FBC8F"),
)
("ui_ok", "#8FBC8F"))
def _resume_panel_colors() -> tuple:
hermes_cli/cli_info_mixin.py
+1 -2
View File
@@ -863,8 +863,7 @@ class CLIInfoMixin:
diff = {
"Added": connected_servers - old_servers,
"Removed": old_servers - connected_servers,
"Reconnected": connected_servers & old_servers,
}
"Reconnected": connected_servers & old_servers}
for label, icon in (("Reconnected", "♻️ "), ("Added", "➕"), ("Removed", "➖")):
if diff[label]:
print(f" {icon} {label}: {', '.join(sorted(diff[label]))}")
hermes_cli/cli_model_switch_mixin.py
+42 -87
View File
@@ -21,8 +21,7 @@ from utils import base_url_host_matches
# override and restored wholesale on rollback.
_RUNTIME_FIELDS = (
"model", "provider", "requested_provider", "_explicit_api_key", "_explicit_base_url",
"api_key", "base_url", "api_mode",
)
"api_key", "base_url", "api_mode")
def _runtime_fields(cli) -> dict:
@@ -57,8 +56,7 @@ def _merge_preflight_warning(cli, result, custom_providers) -> None:
messages=list(cli.conversation_history or []),
custom_providers=custom_providers if custom_providers is not None
else getattr(cli.agent, "_custom_providers", None),
config_context_length=getattr(cli.agent, "_config_context_length", None),
)
config_context_length=getattr(cli.agent, "_config_context_length", None))
except Exception as exc:
logger.debug("preflight-compression switch warning failed: %s", exc)
@@ -78,8 +76,7 @@ def _print_switch_summary(cli, result, old_model, *, one_turn: bool, strict_cont
f"[Note: model was just switched from {_display_old} to {_display_new} "
f"via {result.provider_label or result.target_provider}. "
f"{'This override applies to the next turn only. ' if one_turn else ''}"
f"Adjust your self-identification accordingly.]"
)
f"Adjust your self-identification accordingly.]")
_cprint(f" ✓ Model switched: {_display_new}")
_cprint(f" Provider: {result.provider_label or result.target_provider}")
@@ -93,8 +90,7 @@ def _print_switch_summary(cli, result, old_model, *, one_turn: bool, strict_cont
base_url=result.base_url or cli.base_url or "",
api_key=result.api_key or cli.api_key or "", model_info=mi,
config_context_length=getattr(agent, "_config_context_length", None) if agent else None,
custom_providers=getattr(agent, "_custom_providers", None) if agent else None,
)
custom_providers=getattr(agent, "_custom_providers", None) if agent else None)
except Exception:
if strict_context:
raise
@@ -107,8 +103,7 @@ def _print_switch_summary(cli, result, old_model, *, one_turn: bool, strict_cont
_cprint(f" Capabilities: {mi.format_capabilities()}")
cache_enabled = (
(base_url_host_matches(result.base_url or "", "openrouter.ai") and "claude" in result.new_model.lower())
or result.api_mode == "anthropic_messages"
)
or result.api_mode == "anthropic_messages")
if cache_enabled:
_cprint(" Prompt caching: enabled")
if result.warning_message:
@@ -116,16 +111,14 @@ def _print_switch_summary(cli, result, old_model, *, one_turn: bool, strict_cont
def _switch_model_from(
cli, raw_input, *, is_global, explicit_provider, user_providers, custom_providers
):
cli, raw_input, *, is_global, explicit_provider, user_providers, custom_providers):
"""``switch_model`` seeded with this CLI's live route."""
from hermes_cli.model_switch import switch_model
return switch_model(
raw_input=raw_input, current_provider=cli.provider or "", current_model=cli.model or "",
current_base_url=cli.base_url or "", current_api_key=cli.api_key or "", is_global=is_global,
explicit_provider=explicit_provider, user_providers=user_providers,
custom_providers=custom_providers,
)
custom_providers=custom_providers)
def _run_confirm_and_apply(cli, target, *args) -> None:
@@ -142,8 +135,7 @@ def _run_confirm_and_apply(cli, target, *args) -> None:
def _commit_model_switch(
cli, result, *, persist_global: bool, one_turn: bool = False, picker: bool = False
) -> None:
cli, result, *, persist_global: bool, one_turn: bool = False, picker: bool = False) -> None:
"""Stage + swap, print the summary, persist (session row unless --once; config on --global).
``picker``: tolerate context-resolution errors and label the config write "(--global)"; the
typed path additionally records the one-turn restore snapshot."""
@@ -207,8 +199,7 @@ def _show_model_picker(cli, ctx, force_refresh: bool) -> None:
cli._open_model_picker(
providers, cli.model or "unknown", get_label(cli.provider) if cli.provider else "unknown",
user_provs=ctx.user_providers if ctx is not None else None,
custom_provs=ctx.custom_providers if ctx is not None else None,
)
custom_provs=ctx.custom_providers if ctx is not None else None)
class CLIModelSwitchMixin:
@@ -247,15 +238,12 @@ class CLIModelSwitchMixin:
try:
from hermes_cli.model_normalize import (
_AGGREGATOR_PROVIDERS, normalize_model_for_provider
)
_AGGREGATOR_PROVIDERS, normalize_model_for_provider)
if resolved_provider not in _AGGREGATOR_PROVIDERS:
_adopt(
normalize_model_for_provider(current_model, resolved_provider),
lambda new: (
f"Normalized model '{current_model}' to '{new}' for {resolved_provider}."
),
)
f"Normalized model '{current_model}' to '{new}' for {resolved_provider}."))
except Exception:
pass
@@ -264,8 +252,7 @@ class CLIModelSwitchMixin:
return _adopt_with_mode(
lambda m: normalize_copilot_model_id(m, api_key=self.api_key),
lambda m: copilot_model_api_mode(m, api_key=self.api_key),
lambda new: f"Normalized Copilot model '{current_model}' to '{new}'.",
)
lambda new: f"Normalized Copilot model '{current_model}' to '{new}'.")
from hermes_cli.models import opencode_provider_family
if opencode_provider_family(resolved_provider) is not None:
@@ -275,9 +262,7 @@ class CLIModelSwitchMixin:
lambda m: opencode_model_api_mode(resolved_provider, m),
lambda new: (
f"Stripped provider prefix from '{current_model}'; "
f"using '{new}' for {resolved_provider}."
),
)
f"using '{new}' for {resolved_provider}."))
if resolved_provider != "openai-codex":
return changed
@@ -288,8 +273,7 @@ class CLIModelSwitchMixin:
if not self._model_is_default:
self._console_print(
f"[yellow]⚠️ Stripped provider prefix from '{current_model}'; "
f"using '{slug}' for OpenAI Codex.[/]"
)
f"using '{slug}' for OpenAI Codex.[/]")
self.model = slug
current_model = slug
changed = True
@@ -327,8 +311,7 @@ class CLIModelSwitchMixin:
result.target_provider, base_url=result.base_url, model=result.new_model,
) or None,
"base_url": result.base_url or None,
"api_mode": result.api_mode or None,
}
"api_mode": result.api_mode or None}
try:
db.update_session_model(sid, result.new_model)
db.patch_session_model_config(sid, {"gateway_runtime": route, **route})
@@ -355,8 +338,7 @@ class CLIModelSwitchMixin:
# Stricter than the TUI gateway's recovery (which keeps bare "custom" when a
# base_url exists) — the CLI's resolve path would hard-fail on it.
stored_provider = _heal_bare_custom_provider(
_stored_runtime.get("provider") or None, base_url=stored_base_url, model=stored_model,
)
_stored_runtime.get("provider") or None, base_url=stored_base_url, model=stored_model)
model_changed = stored_model != self.model
provider_changed = bool(stored_provider) and stored_provider != self.provider
if not model_changed and not provider_changed:
@@ -389,16 +371,14 @@ class CLIModelSwitchMixin:
logger.debug(
"Credential re-resolution for resumed session provider "
"%s failed; keeping ambient credentials",
stored_provider, exc_info=True,
)
stored_provider, exc_info=True)
# Mid-chat /resume swaps the live agent; on startup --resume _init_agent picks up
# self.model / self.provider.
if self.agent is not None:
try:
self.agent.switch_model(
new_model=self.model, new_provider=self.provider, api_key=self.api_key or "",
base_url=self.base_url or "", api_mode=self.api_mode or "",
)
base_url=self.base_url or "", api_mode=self.api_mode or "")
except Exception:
logger.debug("In-place agent model swap on resume failed", exc_info=True)
msg = f"Model restored from session: {stored_model}"
@@ -420,8 +400,7 @@ class CLIModelSwitchMixin:
"current_provider": current_provider,
"user_provs": user_provs,
"custom_provs": custom_provs,
"filter": "",
}
"filter": ""}
self._invalidate(min_interval=0.0)
def _confirm_expensive_model_switch(self, result) -> bool:
@@ -433,32 +412,27 @@ class CLIModelSwitchMixin:
warning = combined_selection_warning(
result.new_model, provider=result.target_provider,
base_url=result.base_url or self.base_url or "",
api_key=result.api_key or self.api_key or "", model_info=result.model_info,
)
api_key=result.api_key or self.api_key or "", model_info=result.model_info)
except Exception:
warning = None
if warning is None:
return True
choices = [
("once", "Switch anyway", "Use this model for the current Hermes session."),
("cancel", "Cancel", "Keep the current model."),
]
("cancel", "Cancel", "Keep the current model.")]
raw = self._prompt_text_input_modal(
title=f"!!! {warning.title} !!!", detail=warning.message, choices=choices, timeout=120,
)
title=f"!!! {warning.title} !!!", detail=warning.message, choices=choices, timeout=120)
return self._normalize_slash_confirm_choice(raw, choices) == "once"
def _confirm_and_apply_model_switch_result(
self, result, persist_global: bool, custom_providers=None
) -> None:
self, result, persist_global: bool, custom_providers=None) -> None:
from cli import _cprint
try:
if result.success and not self._confirm_expensive_model_switch(result):
_cprint(" Model switch cancelled.")
return
self._apply_model_switch_result(
result, persist_global, custom_providers=custom_providers
)
result, persist_global, custom_providers=custom_providers)
except Exception as exc:
_cprint(f" ✗ Model selection failed: {exc}")
@@ -474,8 +448,7 @@ class CLIModelSwitchMixin:
**_runtime_fields(self),
"agent_primary_runtime": copy.deepcopy(
getattr(agent, "_primary_runtime", None)
) if agent is not None else None,
}
) if agent is not None else None}
def _restore_model_runtime_snapshot(self, snapshot: dict | None) -> None:
"""Restore a model runtime captured before a one-turn override."""
@@ -505,8 +478,7 @@ class CLIModelSwitchMixin:
new_model=snapshot.get("model", ""), new_provider=snapshot.get("provider", ""),
api_key=snapshot.get("api_key", ""), base_url=snapshot.get("base_url", ""),
api_mode=snapshot.get("api_mode", ""),
capabilities=snapshot.get("capabilities"),
)
capabilities=snapshot.get("capabilities"))
except Exception as exc:
logger.warning("CLI one-turn model restore failed: %s", exc)
@@ -529,8 +501,7 @@ class CLIModelSwitchMixin:
@staticmethod
def _compute_model_picker_viewport(
selected: int, scroll_offset: int, n: int, term_rows: int, reserved_below: int = 6,
panel_chrome: int = 6, min_visible: int = 3,
) -> tuple[int, int]:
panel_chrome: int = 6, min_visible: int = 3) -> tuple[int, int]:
"""Resolve (scroll_offset, visible) for the /model picker viewport. ``reserved_below``
matches the approval/clarify panels (input, status bar, separators); ``panel_chrome`` is
borders + blanks + hint row. The offset slides to keep ``selected`` on screen."""
@@ -557,8 +528,7 @@ class CLIModelSwitchMixin:
if should_clear_context_pin(
model_cfg.get("default") or model_cfg.get("model"), result.new_model,
model_cfg.get("base_url"), result.base_url,
model_cfg.get("provider"), result.target_provider,
):
model_cfg.get("provider"), result.target_provider):
save_config_value("model.context_length", None)
except Exception:
save_config_value("model.context_length", None)
@@ -591,21 +561,18 @@ class CLIModelSwitchMixin:
self.agent.switch_model(
new_model=result.new_model, new_provider=result.target_provider,
api_key=result.api_key, base_url=result.base_url, api_mode=result.api_mode,
capabilities=getattr(result, "runtime_capabilities", None),
)
capabilities=getattr(result, "runtime_capabilities", None))
except Exception as exc:
for _k, _v in _cli_snapshot.items():
setattr(self, _k, _v)
_cprint(
f" ⚠ Model switch to {result.new_model} failed ({exc}); "
f"staying on {old_model}."
)
f"staying on {old_model}.")
return False
return True
def _apply_model_switch_result(
self, result, persist_global: bool, custom_providers=None
) -> None:
self, result, persist_global: bool, custom_providers=None) -> None:
"""Picker-path commit (see _commit_model_switch)."""
from cli import _cprint
if not result.success:
@@ -637,8 +604,7 @@ class CLIModelSwitchMixin:
pass
state.update(
stage="model", provider_data=provider_data, model_list=model_list,
selected=0, filter="", _filtered_pairs=None,
)
selected=0, filter="", _filtered_pairs=None)
self._invalidate(min_interval=0.0)
return
if stage == "model":
@@ -654,8 +620,7 @@ class CLIModelSwitchMixin:
state.update(
stage="provider", filter="", _filtered_pairs=None,
selected=next((i for i, p in enumerate(state.get("providers") or [])
if p.get("slug") == provider_data.get("slug")), 0),
)
if p.get("slug") == provider_data.get("slug")), 0))
self._invalidate(min_interval=0.0)
return
if selected > back_idx: # cancel row (and anything past it)
@@ -666,15 +631,13 @@ class CLIModelSwitchMixin:
self, visible_labels[selected], is_global=persist_global,
explicit_provider=provider_data.get("slug"),
user_providers=state.get("user_provs"),
custom_providers=state.get("custom_provs"),
)
custom_providers=state.get("custom_provs"))
# Capture before close — picker state is cleared on close.
_picker_custom_provs = state.get("custom_provs")
self._close_model_picker()
_run_confirm_and_apply(
self, self._confirm_and_apply_model_switch_result,
result, persist_global, _picker_custom_provs,
)
result, persist_global, _picker_custom_provs)
return
self._close_model_picker()
@@ -704,8 +667,7 @@ class CLIModelSwitchMixin:
one_turn = request.is_once
persist_global = resolve_persist_behavior(
request.is_global, request.is_session, is_once=one_turn,
explicit_provider=request.explicit_provider,
)
explicit_provider=request.explicit_provider)
# --refresh: wipe the picker cache so every authed provider's /v1/models is re-fetched.
if request.force_refresh:
@@ -721,8 +683,7 @@ class CLIModelSwitchMixin:
try:
ctx = load_picker_context().with_overrides(
current_provider=self.provider or "", current_model=self.model or "",
current_base_url=self.base_url or "",
)
current_base_url=self.base_url or "")
except Exception:
ctx = None
# switch_model() + _open_model_picker still need the raw provider dicts.
@@ -735,20 +696,17 @@ class CLIModelSwitchMixin:
result = _switch_model_from(
self, request.target, is_global=persist_global,
explicit_provider=request.explicit_provider,
user_providers=user_provs, custom_providers=custom_provs,
)
user_providers=user_provs, custom_providers=custom_provs)
if not result.success:
_cprint(f" ✗ {result.error_message}")
return
_merge_preflight_warning(self, result, custom_provs)
_run_confirm_and_apply(
self, self._confirm_and_apply_cli_model_switch,
result, persist_global, one_turn, custom_provs,
)
result, persist_global, one_turn, custom_provs)
def _confirm_and_apply_cli_model_switch(
self, result, persist_global: bool, one_turn: bool, custom_provs=None
) -> None:
self, result, persist_global: bool, one_turn: bool, custom_provs=None) -> None:
"""Confirm an expensive model switch and apply it (typed /model path). Runs on a worker
thread when the TUI is active (see _run_confirm_and_apply) so the modal can render."""
from cli import _cprint
@@ -782,8 +740,7 @@ class CLIModelSwitchMixin:
return
result = crs.apply(
load_config(), new_value,
persist_callback=(save_config if new_value is not None else None),
)
persist_callback=(save_config if new_value is not None else None))
prefix = "✓" if result.success else "✗"
for line in result.message.splitlines():
_cprint(f" {prefix} {line}" if line.startswith("openai_runtime") else f" {line}")
@@ -817,9 +774,7 @@ class CLIModelSwitchMixin:
self._pending_moa_restore_model = {
key: getattr(self, key, None)
for key in (
"requested_provider", "provider", "model", "api_key", "base_url", "api_mode",
)
}
"requested_provider", "provider", "model", "api_key", "base_url", "api_mode")}
self.requested_provider = "moa"
self.provider = "moa"
self.model = preset
hermes_cli/cli_session_mixin.py
+47 -94
View File
@@ -28,8 +28,7 @@ def _user_turn_indices(history: list) -> list[int]:
return [
i for i, m in enumerate(history)
if not _is_ephemeral_scaffolding(m) and user_originated_turn_view(m) is not None
]
if not _is_ephemeral_scaffolding(m) and user_originated_turn_view(m) is not None]
def _timestamp_or(value, default):
@@ -84,16 +83,14 @@ def _reset_model_to_config_default(cli, silent: bool) -> None:
current_base_url=cli.base_url or "",
current_api_key=cli.api_key or "",
is_global=False,
explicit_provider=_config_provider or "",
)
explicit_provider=_config_provider or "")
if not r.success:
return
if cli.agent:
cli.agent.switch_model(
new_model=r.new_model, new_provider=r.target_provider, api_key=r.api_key,
base_url=r.base_url, api_mode=r.api_mode,
capabilities=getattr(r, "runtime_capabilities", None),
)
capabilities=getattr(r, "runtime_capabilities", None))
cli.model = r.new_model
cli.provider = r.target_provider
cli.requested_provider = r.target_provider
@@ -175,8 +172,7 @@ class CLISessionMixin:
try:
from hermes_state import SessionDB
from tools.approval import (
_YOLO_MODE_FROZEN, enable_session_yolo, is_session_yolo_enabled,
)
_YOLO_MODE_FROZEN, enable_session_yolo, is_session_yolo_enabled)
except Exception:
return
if _YOLO_MODE_FROZEN or not SessionDB.session_yolo_enabled(session_meta):
@@ -187,8 +183,7 @@ class CLISessionMixin:
enable_session_yolo(session_key)
_dim_notice(self,
"⚡ YOLO mode restored from session — all commands auto-approved. /yolo to turn off.",
quiet,
)
quiet)
def _render_resume_history_panel_lines(self, panel) -> list[str]:
"""Render the resume panel at the current terminal width for resize replay."""
@@ -198,8 +193,7 @@ class CLISessionMixin:
buf = StringIO()
console = Console(
file=buf, force_terminal=True, color_system="truecolor", highlight=False,
width=shutil.get_terminal_size((80, 24)).columns,
)
width=shutil.get_terminal_size((80, 24)).columns)
with _suspend_output_history():
console.print(panel)
return buf.getvalue().rstrip("\n").splitlines()
@@ -246,8 +240,7 @@ class CLISessionMixin:
provider_info += f"{sep}[dim]auth: {self._provider_source}[/]"
self._console_print(
f" {api_indicator} [{accent_color}]{model_short}[/]{sep}"
f"[bold {label_color}]{tool_status}[/]{toolsets_info}{provider_info}"
)
f"[bold {label_color}]{tool_status}[/]{toolsets_info}{provider_info}")
def _show_session_status(self):
"""Show gateway-style status for the current CLI session."""
@@ -320,8 +313,7 @@ class CLISessionMixin:
f"Created: {created_at.strftime('%Y-%m-%d %H:%M')}",
f"Last Activity: {updated_at.strftime('%Y-%m-%d %H:%M')}",
f"Tokens: {total_tokens:,}",
f"Agent Running: {'Yes' if is_running else 'No'}",
])
f"Agent Running: {'Yes' if is_running else 'No'}"])
self._console_print("\n".join(lines), highlight=False, markup=False)
def _list_recent_sessions(self, limit: int = 10) -> list[dict[str, Any]]:
@@ -334,8 +326,7 @@ class CLISessionMixin:
return query_session_listing(
self._session_db, source="cli", current_session_id=self.session_id,
include_all_sources=False, include_unnamed=True, limit=limit,
exclude_sources=["kanban", "tool"],
)
exclude_sources=["kanban", "tool"])
except Exception:
return []
@@ -448,8 +439,7 @@ class CLISessionMixin:
context = {
"session_id": self.agent.session_id if self.agent else None,
"platform": getattr(self, "platform", None) or "cli",
"reason": "new_session" if event_type == "on_session_reset" else "session_boundary",
}
"reason": "new_session" if event_type == "on_session_reset" else "session_boundary"}
if event_type == "on_session_finalize":
finalize_session(**context)
else:
@@ -469,15 +459,13 @@ class CLISessionMixin:
try:
from hermes_constants import get_hermes_home as _ghh
return self._session_db.delete_session_if_empty(
session_id, sessions_dir=_ghh() / "sessions"
)
session_id, sessions_dir=_ghh() / "sessions")
except Exception:
logger.debug("Could not prune empty session %s", session_id, exc_info=True)
return False
def _launch_session_boundary_memory_flush(
self, history_snapshot: list, *, session_id: Optional[str] = None,
) -> Optional[list]:
self, history_snapshot: list, *, session_id: Optional[str] = None) -> Optional[list]:
"""Stage old-session memory extraction so /new stays responsive.
The context-engine ``on_session_end`` is delivered synchronously here: cheap (no LLM)
@@ -508,8 +496,7 @@ class CLISessionMixin:
"""Start a fresh session with a new session ID and cleared agent state."""
from cli import (
CLI_CONFIG, _parse_reasoning_config, _parse_service_tier_config,
_sync_process_session_id, datetime,
)
_sync_process_session_id, datetime)
old_session_id = self.session_id
_boundary_snapshot = None
if self.agent:
@@ -517,8 +504,7 @@ class CLISessionMixin:
# Context-engine boundary now; provider extraction is queued below (after
# rotation) so /new never blocks on the LLM-bound call.
_boundary_snapshot = self._launch_session_boundary_memory_flush(
list(self.conversation_history), session_id=old_session_id,
)
list(self.conversation_history), session_id=old_session_id)
self._notify_session_boundary("on_session_finalize")
if self._session_db and old_session_id:
@@ -527,8 +513,7 @@ class CLISessionMixin:
if self.agent:
with contextlib.suppress(Exception):
self.agent._flush_messages_to_session_db(
self.conversation_history, conversation_history=self.conversation_history,
)
self.conversation_history, conversation_history=self.conversation_history)
with contextlib.suppress(Exception):
self._session_db.end_session(old_session_id, "new_session")
self._discard_session_if_empty(old_session_id)
@@ -543,8 +528,7 @@ class CLISessionMixin:
# An explicit -m/--model was for the previous session only.
self._explicit_model_override = False
self.reasoning_config = _parse_reasoning_config(
CLI_CONFIG["agent"].get("reasoning_effort", "")
)
CLI_CONFIG["agent"].get("reasoning_effort", ""))
# Session-scoped overrides (/model --session, /fast, one-turn restores) don't carry over.
self._pending_one_turn_model_restore = None
self.service_tier = _parse_service_tier_config(CLI_CONFIG["agent"].get("service_tier", ""))
@@ -574,8 +558,7 @@ class CLISessionMixin:
model=self.model,
model_config={
"max_iterations": self.max_turns, "reasoning_config": self.reasoning_config,
},
)
})
self.agent._session_db_created = True
if title:
title = _apply_new_session_title(self, title)
@@ -588,13 +571,11 @@ class CLISessionMixin:
if _mm is not None and _boundary_snapshot:
_mm.commit_session_boundary_async(
_boundary_snapshot, new_session_id=self.session_id,
parent_session_id=old_session_id or "", reason="new_session",
)
parent_session_id=old_session_id or "", reason="new_session")
elif _mm is not None:
_mm.on_session_switch(
self.session_id, parent_session_id=old_session_id or "",
reset=True, reason="new_session",
)
reset=True, reason="new_session")
self._notify_session_boundary("on_session_reset")
if not silent:
@@ -638,8 +619,7 @@ class CLISessionMixin:
"""
from cli import datetime
from hermes_cli.session_export import (
SAVE_USAGE, normalize_save_format, render_session_for_save,
)
SAVE_USAGE, normalize_save_format, render_session_for_save)
parts = cmd.split()[1:]
redact = bool(parts) and parts[-1].lower() in ("redact", "--redact")
@@ -672,8 +652,7 @@ class CLISessionMixin:
return
session_data = {
"id": self.session_id, "model": self.model,
"started_at": self.session_start.timestamp(), "messages": self.conversation_history,
}
"started_at": self.session_start.timestamp(), "messages": self.conversation_history}
if redact:
from hermes_cli.session_export_md import redact_session_data
@@ -718,8 +697,7 @@ class CLISessionMixin:
from agent.context_compressor import (
history_before_user_originated_turn,
split_user_originated_turn,
user_originated_turn_view,
)
user_originated_turn_view)
from agent.memory_manager import sanitize_context
from agent.tool_dispatch_helpers import _is_multimodal_tool_result, _multimodal_text_summary
from run_agent import _is_ephemeral_scaffolding
@@ -734,8 +712,7 @@ class CLISessionMixin:
if isinstance(part, dict) and part.get("type") == "text":
text_parts.append(str(part.get("text", "")))
elif isinstance(part, dict) and part.get("type") in {
"image", "image_url", "input_image",
}:
"image", "image_url", "input_image"}:
text_parts.append("[screenshot]")
return "\n".join(text_parts) if text_parts else None
return content
@@ -752,8 +729,7 @@ class CLISessionMixin:
changed = RuntimeError("session history changed before the rewind could be persisted")
expected_active_ids = self._session_db.get_active_message_ids(self.session_id)
durable = self._session_db.get_messages_as_conversation(
self.session_id, include_row_ids=True
)
self.session_id, include_row_ids=True)
warm_persistence_history = [m for m in warm_history if not _is_ephemeral_scaffolding(m)]
warm_user_indices = _user_indices(warm_persistence_history)
durable_user_indices = _user_indices(durable)
@@ -763,13 +739,11 @@ class CLISessionMixin:
raise RuntimeError("persisted rewind target is no longer available")
warm_prefix, _ = history_before_user_originated_turn(
warm_persistence_history, warm_user_indices[user_ordinal]
)
warm_persistence_history, warm_user_indices[user_ordinal])
durable_target_index = durable_user_indices[user_ordinal]
durable_target = durable[durable_target_index]
durable_prefix, durable_live_view = history_before_user_originated_turn(
durable, durable_target_index
)
durable, durable_target_index)
if _comparison_content(durable_live_view) != _comparison_content(warm_live_view):
raise changed
target_row_id = durable_target.get("_row_id")
@@ -780,8 +754,7 @@ class CLISessionMixin:
self.session_id, target_row_id,
preserve_compaction_handoff=scaffold is not None,
expected_active_ids=expected_active_ids,
expected_target_content=durable_live_view.get("content"),
)
expected_target_content=durable_live_view.get("content"))
if scaffold is not None:
replacement_id = result.get("replacement_message_id")
if not isinstance(replacement_id, int) or not durable_prefix:
@@ -817,8 +790,7 @@ class CLISessionMixin:
return None
from agent.context_compressor import (
history_before_user_originated_turn, retryable_user_text,
)
history_before_user_originated_turn, retryable_user_text)
from agent.memory_manager import sanitize_context
warm_history = list(self.conversation_history)
@@ -833,8 +805,7 @@ class CLISessionMixin:
# by /retry, so fail closed before archiving anything.
try:
truncated, live_view = history_before_user_originated_turn(
warm_history, user_indices[-1]
)
warm_history, user_indices[-1])
live_content = live_view.get("content")
if isinstance(live_content, str):
live_content = sanitize_context(live_content).strip()
@@ -850,8 +821,7 @@ class CLISessionMixin:
truncated, _, _ = self._rewind_persisted_user_turn(
warm_history=warm_history,
user_ordinal=len(user_indices) - 1,
warm_live_view=live_view,
)
warm_live_view=live_view)
except Exception as exc:
print(f"(x_x) Retry rewind failed; history was not changed: {exc}")
return None
@@ -896,8 +866,7 @@ class CLISessionMixin:
truncated, durable_live_view, result = self._rewind_persisted_user_turn(
warm_history=warm_history,
user_ordinal=target_ordinal,
warm_live_view=live_view,
)
warm_live_view=live_view)
# Canonical editable prefill: the raw carrier holds the reference-summary wrapper.
durable_text = self._undo_content_to_text(durable_live_view.get("content"))
if durable_text:
@@ -919,8 +888,7 @@ class CLISessionMixin:
turn_word = "turn" if turns_undone == 1 else "turns"
print(
f"(^_^)b Undid {turns_undone} {turn_word} ({rewound_rows or removed_count} message(s)). "
f"Backed up to: \"{removed_text[:60]}{'...' if len(removed_text) > 60 else ''}\""
)
f"Backed up to: \"{removed_text[:60]}{'...' if len(removed_text) > 60 else ''}\"")
print(f" {len(self.conversation_history)} message(s) remaining in history.")
# Editable, not auto-sent (Claude-Code-style).
if prefill and removed_text:
@@ -955,8 +923,7 @@ class CLISessionMixin:
return
try:
from tools.approval import (
disable_session_yolo, enable_session_yolo, is_session_yolo_enabled,
)
disable_session_yolo, enable_session_yolo, is_session_yolo_enabled)
except Exception:
return
if is_session_yolo_enabled(old_session_id):
@@ -993,8 +960,7 @@ class CLISessionMixin:
from cli import _cprint
from hermes_cli.colors import Colors as _Colors
from tools.approval import (
_YOLO_MODE_FROZEN, disable_session_yolo, enable_session_yolo, is_session_yolo_enabled,
)
_YOLO_MODE_FROZEN, disable_session_yolo, enable_session_yolo, is_session_yolo_enabled)
# A frozen process-level bypass short-circuits the approval gate ahead of the session
# check — toggling "OFF" would be a false safety claim. Say so instead.
@@ -1003,8 +969,7 @@ class CLISessionMixin:
f" ⚡ YOLO is {_Colors.BOLD}{_Colors.RED}locked ON{_Colors.RESET}"
" for this process (started with --yolo / HERMES_YOLO_MODE)."
" /yolo cannot disable it — restart without the flag to"
" re-enable approvals."
)
" re-enable approvals.")
return
session_key = self.session_id or "default"
@@ -1016,16 +981,14 @@ class CLISessionMixin:
_persist(session_key, False)
_cprint(
f" ⚠ YOLO mode {_Colors.BOLD}{_Colors.RED}OFF{_Colors.RESET}"
" — dangerous commands will require approval."
)
" — dangerous commands will require approval.")
else:
enable_session_yolo(session_key)
if _persist:
_persist(session_key, True)
_cprint(
f" ⚡ YOLO mode {_Colors.BOLD}{_Colors.GREEN}ON{_Colors.RESET}"
" — all commands auto-approved. Use with caution."
)
" — all commands auto-approved. Use with caution.")
def _persist_session_yolo(self, session_key: str, enabled: bool) -> None:
"""Persist the YOLO flag to the session row so --resume restores it. Best-effort; the
@@ -1056,8 +1019,7 @@ class CLISessionMixin:
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,
)
split_history_for_partial_compress, summarize_compress_preview)
from agent.conversation_compression import finalize_context_engine_compression_notification
from agent.model_metadata import estimate_request_tokens_rough
@@ -1080,13 +1042,11 @@ class CLISessionMixin:
# understates real request pressure and can even appear to grow after compression.
_estimate_kw = {
"system_prompt": getattr(self.agent, "_cached_system_prompt", "") or "",
"tools": getattr(self.agent, "tools", None) or None,
}
"tools": getattr(self.agent, "tools", None) or None}
if preview:
approx_tokens = estimate_request_tokens_rough(self.conversation_history, **_estimate_kw)
report = summarize_compress_preview(
self.conversation_history, partial, keep_last, focus_topic or None, approx_tokens,
)
self.conversation_history, partial, keep_last, focus_topic or None, approx_tokens)
for line in report["lines"]:
print(f"🗜️ {line}")
return
@@ -1122,8 +1082,7 @@ class CLISessionMixin:
# passing _cached_system_prompt duplicated the identity block.
compressed, _ = self.agent._compress_context(
head, None, approx_tokens=approx_tokens, focus_topic=focus_topic or None,
force=True, defer_context_engine_notification=True,
)
force=True, defer_context_engine_notification=True)
# Unchanged because a concurrent compression lock is held: say so instead of
# the misleading "No changes" no-op text. Type-pinned check (is True / str) —
@@ -1154,17 +1113,14 @@ class CLISessionMixin:
self.agent._flush_messages_to_session_db(self.conversation_history, None)
finalize_context_engine_compression_notification(self.agent, committed=True)
new_tokens = estimate_request_tokens_rough(
self.conversation_history, **_estimate_kw
)
self.conversation_history, **_estimate_kw)
summary = summarize_manual_compression(
original_history, self.conversation_history, approx_tokens, new_tokens,
compression_state=getattr(self.agent, "context_compressor", None),
)
compression_state=getattr(self.agent, "context_compressor", None))
if (
summary.get("aborted")
or summary.get("fallback_used")
or summary.get("refused_would_grow")
):
or summary.get("refused_would_grow")):
icon = "⚠️"
else:
icon = "🗜️" if summary["noop"] else "✅"
@@ -1226,8 +1182,7 @@ class CLISessionMixin:
if not isinstance(messages, list):
return
if isinstance(pending_cli_message, dict) and not any(
m is pending_cli_message for m in messages
):
m is pending_cli_message for m in messages):
# The UI accepted a new input but the worker still exposes its prior snapshot.
messages = [*messages, pending_cli_message]
if not messages:
@@ -1241,8 +1196,7 @@ class CLISessionMixin:
if (
isinstance(conversation_history, list)
and conversation_history
and conversation_history[-1] is pending_cli_message
):
and conversation_history[-1] is pending_cli_message):
# Accepted but not yet durable: exclude it from the resumed-history baseline.
conversation_history = conversation_history[:-1]
elif not isinstance(conversation_history, list) or conversation_history is messages:
@@ -1295,8 +1249,7 @@ class CLISessionMixin:
user_msgs = len([m for m in self.conversation_history if m.get("role") == "user"])
tool_calls = len([
m for m in self.conversation_history if m.get("role") == "tool" or m.get("tool_calls")
])
m for m in self.conversation_history if m.get("role") == "tool" or m.get("tool_calls")])
elapsed = datetime.now() - self.session_start
hours, remainder = divmod(int(elapsed.total_seconds()), 3600)
minutes, seconds = divmod(remainder, 60)
hermes_cli/cli_status_bar_mixin.py
+16 -32
View File
@@ -22,8 +22,7 @@ _STRONG = "class:status-bar-strong"
_AGENT_COUNTERS = (
"session_input_tokens", "session_output_tokens", "session_cache_read_tokens",
"session_cache_write_tokens", "session_prompt_tokens", "session_completion_tokens",
"session_total_tokens", "session_api_calls",
)
"session_total_tokens", "session_api_calls")
def _threshold_style(value, ladder, fallback: str) -> str:
@@ -135,8 +134,7 @@ class CLIStatusBarMixin:
@staticmethod
def _format_prompt_elapsed(
prompt_start_time: Optional[float], prompt_duration: float, live: bool = False
) -> str:
prompt_start_time: Optional[float], prompt_duration: float, live: bool = False) -> str:
"""Per-prompt elapsed time. Always a string (``⏲ 0s`` on fresh start); seconds stay
visible at every scale so it increments smoothly (``1m 59s → 2m → 2m 1s``). ⏱ while
live, ⏲ frozen — width-1 glyphs (no variation selector) keep the bar aligned."""
@@ -196,11 +194,9 @@ class CLIStatusBarMixin:
"duration": format_duration_compact(elapsed_seconds),
"session_title": self._get_status_bar_session_title(),
"prompt_elapsed": self._format_prompt_elapsed(
prompt_start, getattr(self, "_prompt_duration", 0.0), live=turn_live,
),
prompt_start, getattr(self, "_prompt_duration", 0.0), live=turn_live),
"idle_since": self._format_idle_since(
getattr(self, "_last_turn_finished_at", None), turn_live=turn_live,
),
getattr(self, "_last_turn_finished_at", None), turn_live=turn_live),
"context_tokens": 0,
"context_length": None,
"context_percent": None,
@@ -214,15 +210,13 @@ class CLIStatusBarMixin:
"focus_label": "", # /focus badge: the reduced-output mode is never invisible.
"goal_active": False,
"goal_turns_used": 0,
"goal_max_turns": 0,
}
"goal_max_turns": 0}
try:
from hermes_cli.focus_view import focus_statusbar_segment
snapshot["focus_label"] = focus_statusbar_segment(
bool(getattr(self, "_focus_view_enabled", False))
)
bool(getattr(self, "_focus_view_enabled", False)))
except Exception:
pass
@@ -291,8 +285,7 @@ class CLIStatusBarMixin:
_anchored = anchored_context_tokens(
_msgs if isinstance(_msgs, list) else [],
getattr(agent, "_turn_base_usage_anchor", None),
charge_stale_thinking=False,
)
charge_stale_thinking=False)
if _anchored is not None and _anchored > 0:
context_tokens = _anchored
except Exception:
@@ -542,8 +535,7 @@ class CLIStatusBarMixin:
from agent.turn_summary import format_token_flow
produced = (getattr(agent, "session_output_tokens", 0) or 0) - (
getattr(self, "_turn_token_baseline", 0) or 0
)
getattr(self, "_turn_token_baseline", 0) or 0)
return format_token_flow(produced)
except Exception:
return ""
@@ -651,11 +643,9 @@ class CLIStatusBarMixin:
or self._pet_slug != pet.slug
or self._pet_cols != cols
or self._pet_scale != scale
or self._pet_renderer.mode != renderer_mode
):
or self._pet_renderer.mode != renderer_mode):
self._pet_renderer = pet_render.PetRenderer(
str(pet.spritesheet), mode=renderer_mode, scale=scale, unicode_cols=cols
)
str(pet.spritesheet), mode=renderer_mode, scale=scale, unicode_cols=cols)
self._pet_slug = pet.slug
self._pet_cols = cols
self._pet_scale = scale
@@ -711,8 +701,7 @@ class CLIStatusBarMixin:
or self._clarify_state
or self._sudo_state
or self._secret_state
or getattr(self, "_slash_confirm_state", None)
)
or getattr(self, "_slash_confirm_state", None))
return derive_pet_state(
awaiting_input=awaiting_input,
busy=getattr(self, "_agent_running", False),
@@ -968,8 +957,7 @@ class CLIStatusBarMixin:
return result
def _status_bar_segments(
self, snapshot, width: int, field_set, yolo_active: bool, *, styled: bool
) -> list:
self, snapshot, width: int, field_set, yolo_active: bool, *, styled: bool) -> list:
"""Ordered status-bar segments for one width tier (<52 / <76 / wide), each a list of
``(style, text)`` fragments. Shared by the plain-text and prompt_toolkit renderers so
the two can never drift; ``styled`` selects the graphical context bar."""
@@ -1026,8 +1014,7 @@ class CLIStatusBarMixin:
add("cache_hit", self._cache_hit_rate_style(cache[0]), cache[1])
if wide:
for name, key, glyph in (
("latency", "avg_latency_label", "◷"), ("tps", "avg_velocity_label", "↑"),
):
("latency", "avg_latency_label", "◷"), ("tps", "avg_velocity_label", "↑")):
label = snapshot.get(key) or ""
if label:
add(name, _DIM, f"{glyph} {label}")
@@ -1067,8 +1054,7 @@ class CLIStatusBarMixin:
show_title = field_set is None or "title" in field_set
session_title = (snapshot.get("session_title") or "") if show_title else ""
segs = self._status_bar_segments(
snapshot, width, field_set, self._is_session_yolo_active(), styled=False
)
snapshot, width, field_set, self._is_session_yolo_active(), styled=False)
parts = ["".join(t for _, t in seg) for seg in segs] or [f"⚕ {model_short}"]
# Narrow bars always join the battery with │; wider tiers use the tier separator.
if battery_label:
@@ -1085,8 +1071,7 @@ class CLIStatusBarMixin:
if (
not self._status_bar_visible
or getattr(self, "_model_picker_state", None)
or getattr(self, "_command_palette_state", None)
):
or getattr(self, "_command_palette_state", None)):
return []
try:
snapshot = self._get_status_bar_snapshot()
@@ -1100,8 +1085,7 @@ class CLIStatusBarMixin:
session_title = (snapshot.get("session_title") or "") if _ok("title") else ""
segs = self._status_bar_segments(
snapshot, width, field_set, self._is_session_yolo_active(), styled=True
)
snapshot, width, field_set, self._is_session_yolo_active(), styled=True)
sep = " · " if width < 76 else " │ "
frags: list = []
for seg in segs or [[(_SB, " ⚕ "), (_STRONG, snapshot["model_short"])]]:
hermes_cli/cli_stream_mixin.py
+24 -48
View File
@@ -20,8 +20,7 @@ from rich.markup import escape as _escape
# Model-generated reasoning tags: suppressed during streaming (they'd display as raw XML;
# the agent strips them from final_response too) unless show_reasoning routes them to the box.
_OPEN_TAGS = (
"<REASONING_SCRATCHPAD>", "<think>", "<reasoning>", "<THINKING>", "<thinking>", "<thought>",
)
"<REASONING_SCRATCHPAD>", "<think>", "<reasoning>", "<THINKING>", "<thinking>", "<thought>")
_CLOSE_TAGS = tuple("</" + t[1:] for t in _OPEN_TAGS)
_MAX_CLOSE_TAG_LEN = max(len(t) for t in _CLOSE_TAGS)
@@ -29,13 +28,11 @@ _MAX_CLOSE_TAG_LEN = max(len(t) for t in _CLOSE_TAGS)
_SLOW_COMMAND_STATUS = (
("/skills search", "Searching skills..."), ("/skills browse", "Loading skills..."),
("/skills inspect", "Inspecting skill..."), ("/skills install", "Installing skill..."),
("/skills", "Processing skills command..."), ("/browser", "Configuring browser..."),
)
("/skills", "Processing skills command..."), ("/browser", "Configuring browser..."))
_SLOW_COMMAND_STATUS_EXACT = {
"/reload-mcp": "Reloading MCP servers...",
"/reload-skills": "Reloading skills...",
"/reload_skills": "Reloading skills...",
}
"/reload_skills": "Reloading skills..."}
def _terminal_columns(default: int = 80) -> int:
@@ -145,18 +142,15 @@ class CLIStreamMixin:
min_newline_flush = max(16, target_width // 3)
if line_break != -1 and (
line_break >= min_newline_flush
or buf.endswith(("\n\n", ".\n", "!\n", "?\n", ":\n"))
):
or buf.endswith(("\n\n", ".\n", "!\n", "?\n", ":\n"))):
flush_text, buf = buf[: line_break + 1], buf[line_break + 1 :]
elif len(buf) >= target_width:
search_start = max(20, target_width // 2)
search_end = min(
len(buf), max(target_width + (target_width // 3), target_width + 8)
)
len(buf), max(target_width + (target_width // 3), target_width + 8))
cut = max(
buf.rfind(b, search_start, search_end)
for b in (" ", "\t", ".", "!", "?", ",", ";", ":")
)
for b in (" ", "\t", ".", "!", "?", ",", ";", ":"))
if cut != -1:
flush_text, buf = buf[: cut + 1], buf[cut + 1 :]
@@ -169,8 +163,7 @@ class CLIStreamMixin:
from cli import _accent_hex, datetime
ts_suffix = (
f" [dim]{datetime.now().strftime(getattr(self, 'timestamp_format', '%H:%M'))}[/]"
if getattr(self, "show_timestamps", False) else ""
)
if getattr(self, "show_timestamps", False) else "")
lines = user_input.split("\n")
if len(lines) <= 1:
return f"[bold {_accent_hex()}]●[/] [bold]{_escape(user_input)}[/]{ts_suffix}"
@@ -305,8 +298,7 @@ class CLIStreamMixin:
# Boundary: only whitespace since the last newline — or, with no newline
# buffered yet, since the last emit (which must have ended a line).
is_block_boundary = preceding[preceding.rfind("\n") + 1:].strip() == "" and (
"\n" in preceding or getattr(self, "_stream_last_was_newline", True)
)
"\n" in preceding or getattr(self, "_stream_last_was_newline", True))
if is_block_boundary:
if preceding:
self._emit_stream_text(preceding)
@@ -363,15 +355,13 @@ class CLIStreamMixin:
from cli import _RST, _STREAM_PAD, _cprint
_tc = getattr(self, "_stream_text_ansi", "")
_cprint(
f"{_STREAM_PAD}{_tc}{printed_line}{_RST}" if _tc else f"{_STREAM_PAD}{printed_line}"
)
f"{_STREAM_PAD}{_tc}{printed_line}{_RST}" if _tc else f"{_STREAM_PAD}{printed_line}")
def _flush_stream_table_buf(self) -> None:
"""Emit the held table block re-aligned as a whole. Cell-level markdown is stripped FIRST
so the realigner pads to the final visible width, not the marker-decorated width."""
from cli import (
_strip_markdown_syntax, _terminal_width_for_streaming, realign_markdown_tables
)
_strip_markdown_syntax, _terminal_width_for_streaming, realign_markdown_tables)
buf = self._stream_table_buf
self._stream_table_buf = []
self._in_stream_table = False
@@ -388,8 +378,7 @@ class CLIStreamMixin:
"""Emit filtered text to the streaming display."""
from cli import (
HermesCLI, _ACCENT, _RST, _STREAM_PARTIAL_PREVIEW_LEN, _cprint, _strip_markdown_syntax,
datetime, is_table_divider, looks_like_table_row,
)
datetime, is_table_divider, looks_like_table_row)
if not text:
return
# Defer content while the reasoning box renders so reasoning always lands BEFORE it.
@@ -448,8 +437,7 @@ class CLIStreamMixin:
self._stream_buf
and not self._in_stream_table
and not self._stream_buf.lstrip().startswith("|")
and len(self._stream_buf) >= 80
):
and len(self._stream_buf) >= 80):
preview = self._stream_buf[-int(_STREAM_PARTIAL_PREVIEW_LEN):]
cut = preview.find(" ")
if 0 < cut < len(preview) - 1:
@@ -463,8 +451,7 @@ class CLIStreamMixin:
def _flush_stream(self) -> None:
"""Emit any remaining partial line from the stream buffer and close the box."""
from cli import (
_ACCENT, _RST, _cprint, _strip_markdown_syntax, is_table_divider, looks_like_table_row
)
_ACCENT, _RST, _cprint, _strip_markdown_syntax, is_table_divider, looks_like_table_row)
# Still inside a "reasoning block" at end-of-stream = false positive (the model
# mentioned a tag in prose and never closed it): recover the buffer as regular text.
if getattr(self, "_in_reasoning_block", False) and getattr(self, "_stream_prefilt", ""):
@@ -477,8 +464,7 @@ class CLIStreamMixin:
if (
self._stream_buf
and getattr(self, "_in_stream_table", False)
and (looks_like_table_row(self._stream_buf) or is_table_divider(self._stream_buf))
):
and (looks_like_table_row(self._stream_buf) or is_table_divider(self._stream_buf))):
self._stream_table_buf.append(self._stream_buf)
self._stream_buf = ""
if getattr(self, "_stream_table_buf", None):
@@ -554,8 +540,7 @@ class CLIStreamMixin:
analysis_prompt = (
"Describe everything visible in this image in thorough detail. "
"Include any text, code, data, objects, people, layout, colors, "
"and any other notable visual information."
)
"and any other notable visual information.")
enriched_parts = []
for img_path in images:
if not img_path.exists():
@@ -565,32 +550,28 @@ class CLIStreamMixin:
_cprint(f" {_DIM}👁️ analyzing {img_path.name} ({size_kb}KB)...{_RST}")
try:
result_json = _asyncio.run(
vision_analyze_tool(image_url=str(img_path), user_prompt=analysis_prompt)
)
vision_analyze_tool(image_url=str(img_path), user_prompt=analysis_prompt))
result = json.loads(result_json)
if result.get("success"):
description = result.get("analysis", "")
enriched_parts.append(
f"[The user attached an image. Here's what it contains:\n{description}]\n"
f"[If you need a closer look, use vision_analyze with "
f"image_url: {img_path}]"
)
f"image_url: {img_path}]")
if announce:
_cprint(f" {_DIM}✓ image analyzed{_RST}")
else:
enriched_parts.append(
f"[The user attached an image but it couldn't be analyzed. "
f"You can try examining it with vision_analyze using "
f"image_url: {img_path}]"
)
f"image_url: {img_path}]")
if announce:
_cprint(f" {_DIM}⚠ vision analysis failed — path included for retry{_RST}")
except Exception as e:
enriched_parts.append(
f"[The user attached an image but analysis failed ({e}). "
f"You can try examining it with vision_analyze using "
f"image_url: {img_path}]"
)
f"image_url: {img_path}]")
if announce:
_cprint(f" {_DIM}⚠ vision analysis error — path included for retry{_RST}")
@@ -663,8 +644,7 @@ class CLIStreamMixin:
if event_type == "tool.completed":
self._tool_start_time = 0.0
self._turn_summary_record(
function_name, kwargs.get("result"), kwargs.get("is_error", False)
)
function_name, kwargs.get("result"), kwargs.get("is_error", False))
# Focus view: count the hidden scrollback line for the post-turn recovery report.
if getattr(self, "_focus_view_enabled", False):
try:
@@ -697,11 +677,9 @@ class CLIStreamMixin:
if (
not getattr(self, "_long_tool_hint_fired", False)
and self.tool_progress_mode == "all"
and duration >= 30.0
):
and duration >= 30.0):
from agent.onboarding import (
TOOL_PROGRESS_FLAG, is_seen, mark_seen, tool_progress_hint_cli
)
TOOL_PROGRESS_FLAG, is_seen, mark_seen, tool_progress_hint_cli)
if not is_seen(CLI_CONFIG, TOOL_PROGRESS_FLAG):
self._long_tool_hint_fired = True
_cprint(f" {_DIM}{tool_progress_hint_cli()}{_RST}")
@@ -723,8 +701,7 @@ class CLIStreamMixin:
self._tool_start_time = time.monotonic()
# Store args for stacked scrollback line on completion
self._pending_tool_info.setdefault(function_name, []).append(
function_args if function_args is not None else {}
)
function_args if function_args is not None else {})
self._invalidate()
def _on_tool_start(self, tool_call_id: str, function_name: str, function_args: dict):
@@ -760,7 +737,6 @@ class CLIStreamMixin:
from agent.display import render_edit_diff_with_delta
render_edit_diff_with_delta(
function_name, function_result, function_args=function_args, snapshot=snapshot,
print_fn=_cprint,
)
print_fn=_cprint)
except Exception:
logger.debug("Edit diff preview failed for %s", function_name, exc_info=True)
hermes_cli/cli_tui_mixin.py
+16 -32
View File
@@ -393,8 +393,7 @@ class CLITuiMixin:
input_area,
input_rule_bot,
voice_status_bar,
completions_menu,
]
completions_menu]
return [item for item in ordered if item is not None]
def _tui_spinner_loop(self):
@@ -630,8 +629,7 @@ class CLITuiMixin:
choices.append("Cancel")
hint = (
f"Current: {state.get('current_model', 'unknown')} "
f"on {state.get('current_provider', 'unknown')}"
)
f"on {state.get('current_provider', 'unknown')}")
else:
provider_data = state.get("provider_data") or {}
model_list = state.get("model_list") or []
@@ -652,8 +650,7 @@ class CLITuiMixin:
else:
hint = "No models listed for this provider. Use Back or Cancel."
return self._render_scroll_list_panel(
state, title, hint, choices, min_width=46, max_width=84, indent=' ',
)
state, title, hint, choices, min_width=46, max_width=84, indent=' ')
def _get_command_palette_display_fragments(self):
state = self._command_palette_state
@@ -774,8 +771,7 @@ class CLITuiMixin:
try:
_stash = self._prompt_stash
return self._render_stash_panel(
_stash.panel_rows(), _stash.panel_cursor, self._get_tui_terminal_width(),
)
_stash.panel_rows(), _stash.panel_cursor, self._get_tui_terminal_width())
except Exception:
return []
@@ -1314,8 +1310,7 @@ class CLITuiMixin:
from cli import (
_apply_backslash_line_continuation,
_is_backslash_line_continuation,
_looks_like_slash_command,
)
_looks_like_slash_command)
if self._tui_enter_overlay(event):
return
buf = event.app.current_buffer
@@ -1431,8 +1426,7 @@ class CLITuiMixin:
with open(_hermes_home / "interrupt_debug.log", "a", encoding="utf-8") as _f:
_f.write(
f"{time.strftime('%H:%M:%S')} ENTER: queued interrupt msg={str(payload)[:60]!r}, "
f"agent_running={self._agent_running}\n"
)
f"agent_running={self._agent_running}\n")
except Exception:
pass
# First-touch onboarding: one-line tip about the /busy knob on the first busy-while-
@@ -1816,8 +1810,7 @@ class CLITuiMixin:
CLI_CONFIG,
_bind_prompt_submit_keys,
_cli_multiline_shortcuts_enabled,
_preserve_ctrl_enter_newline,
)
_preserve_ctrl_enter_newline)
from prompt_toolkit.keys import Keys
kb = KeyBindings()
_multiline_shortcuts_enabled = _cli_multiline_shortcuts_enabled(self.config or CLI_CONFIG)
@@ -1825,8 +1818,7 @@ class CLITuiMixin:
kb.add(Keys.Ignore, eager=True)(self._tui_handle_ignored_terminal_sequence)
_bind_prompt_submit_keys(
kb, self._tui_handle_enter, multiline_shortcuts_enabled=_multiline_shortcuts_enabled,
)
kb, self._tui_handle_enter, multiline_shortcuts_enabled=_multiline_shortcuts_enabled)
kb.add('escape', 'enter')(self._tui_insert_newline)
# Ctrl+J inserts a newline (Claude Code / Codex / OpenCode). Windows Terminal delivers
# Ctrl+Enter as the same c-j code. display.cli_multiline_shortcuts: false restores legacy
@@ -1852,8 +1844,7 @@ class CLITuiMixin:
kb.add('c-q')(self._tui_handle_ctrl_q)
kb.add('c-d')(self._tui_handle_ctrl_d)
_modal_prompt_active = Condition(
lambda: bool(self._secret_state or self._sudo_state or self._slash_confirm_state)
)
lambda: bool(self._secret_state or self._sudo_state or self._slash_confirm_state))
kb.add('escape', filter=_modal_prompt_active, eager=True)(self._tui_handle_escape_modal)
kb.add('escape', 'escape', filter=~_modal_prompt_active)(self._tui_handle_double_escape)
kb.add('c-z')(self._tui_handle_ctrl_z)
@@ -1869,11 +1860,9 @@ class CLITuiMixin:
# unbound there and arrives as ('escape', 'g') — register it as a fallback.
_editor_filter = Condition(
lambda: not self._clarify_state and not self._approval_state
and not self._sudo_state and not self._secret_state
)
and not self._sudo_state and not self._secret_state)
kb.add('c-g', filter=_editor_filter)(
kb.add('escape', 'g', filter=_editor_filter)(self._tui_handle_open_in_editor)
)
kb.add('escape', 'g', filter=_editor_filter)(self._tui_handle_open_in_editor))
# Ctrl+S prompt stash: park a draft, send something else, bring it back. Suppressed while
# a modal prompt owns the composer so Ctrl+S can't stash a password.
_stash_filter = Condition(
@@ -1896,8 +1885,7 @@ class CLITuiMixin:
_clarify_nav = Condition(lambda: bool(self._clarify_state) and not self._clarify_freetext)
_clarify_batch = Condition(
lambda: bool(self._clarify_state) and bool(self._clarify_state.get("questions"))
and not self._clarify_freetext
)
and not self._clarify_freetext)
kb.add('up', filter=_clarify_nav)(self._tui_clarify_up)
kb.add('down', filter=_clarify_nav)(self._tui_clarify_down)
# Multi-select: Space toggles the checkbox under the cursor.
@@ -2005,16 +1993,14 @@ class CLITuiMixin:
spinner_widget = Window(
content=FormattedTextControl(self._tui_spinner_text),
height=self._tui_spinner_height,
wrap_lines=True,
)
wrap_lines=True)
# Petdex mascot — right-aligned Kitty placeholder or half-block sprite above the prompt;
# height 0 when no pet is enabled. The animation thread queues virtual Kitty frames;
# after_render writes them out-of-band while prompt_toolkit owns the placeholder grid.
self._pet_widget = Window(
content=FormattedTextControl(self._pet_fragments),
height=self._pet_widget_height,
align=WindowAlign.RIGHT,
)
align=WindowAlign.RIGHT)
# Hint line above the input: only for interactive prompts that need extra instructions
# (sudo countdown, approval navigation, clarify); the agent-running hint is the placeholder.
spacer = Window(content=FormattedTextControl(self._tui_hint_text), height=self._tui_hint_height)
@@ -2023,11 +2009,9 @@ class CLITuiMixin:
secret_widget = self._tui_overlay_widget(self._get_secret_display_fragments, "_secret_state")
approval_widget = self._tui_overlay_widget(self._get_approval_display_fragments, "_approval_state")
slash_confirm_widget = self._tui_overlay_widget(
self._get_slash_confirm_display_fragments, "_slash_confirm_state",
)
self._get_slash_confirm_display_fragments, "_slash_confirm_state")
model_picker_widget = self._tui_overlay_widget(
self._get_model_picker_display_fragments, "_model_picker_state",
)
self._get_model_picker_display_fragments, "_model_picker_state")
command_palette_widget = self._tui_overlay_widget(
self._get_command_palette_display_fragments, "_command_palette_state")
# Rules above/below the input; narrow terminals hide the bottom one to recover a row.
hermes_cli/cli_voice_mixin.py
+11 -22
View File
@@ -82,16 +82,14 @@ class CLIVoiceMixin:
)
raise RuntimeError(
"Voice mode requires sounddevice and numpy.\n"
f"Install with: {sys.executable} -m pip install sounddevice numpy"
)
f"Install with: {sys.executable} -m pip install sounddevice numpy")
if not reqs.get("stt_available", reqs.get("stt_key_set")):
raise RuntimeError(
"Voice mode requires an STT provider for transcription.\n"
"Option 1: uv pip install faster-whisper "
"(free, local; `pip install faster-whisper` also works if pip is on PATH)\n"
"Option 2: Set GROQ_API_KEY (free tier)\n"
"Option 3: Set VOICE_TOOLS_OPENAI_KEY (paid)"
)
"Option 3: Set VOICE_TOOLS_OPENAI_KEY (paid)")
# Prevent double-start from concurrent threads (atomic check-and-set)
with self._voice_lock:
@@ -223,8 +221,7 @@ class CLIVoiceMixin:
if self._voice_stt_provider() == "local":
_cprint(
f"{_DIM}Preparing local STT model '{stt_model}' "
f"(first use may download it from Hugging Face)...{_RST}"
)
f"(first use may download it from Hugging Face)...{_RST}")
else:
_cprint(f"{_DIM}Transcribing...{_RST}")
from tools.voice_mode import is_voice_stop_phrase, transcribe_recording
@@ -270,8 +267,7 @@ class CLIVoiceMixin:
_tts_done = getattr(self, "_voice_tts_done", None)
_activity_hold = bool(
getattr(self, "_agent_running", False)
or (_tts_done is not None and not _tts_done.is_set())
)
or (_tts_done is not None and not _tts_done.is_set()))
if submitted:
self._no_speech_count = 0
elif not _activity_hold:
@@ -287,8 +283,7 @@ class CLIVoiceMixin:
self._voice_continuous
and not submitted
and not self._voice_recording
and not stop_continuous_restart
):
and not stop_continuous_restart):
self._voice_restart_recording_async()
def _voice_speak_response_async(self, text: str) -> None:
@@ -419,8 +414,7 @@ class CLIVoiceMixin:
# Generation phase: no audio to cut — interrupt the in-flight agent turn.
logger.debug(
"full-duplex listener tripped during generation — "
"interrupting agent turn"
)
"interrupting agent turn")
if _pipe_stop is not None:
_pipe_stop.set() # never let the stale reply speak
try:
@@ -432,8 +426,7 @@ class CLIVoiceMixin:
wav_path = full_duplex_listen(
_should_stop, is_playing=is_audio_output_active, on_trigger=_on_trigger,
multiplier=_mult or None, grace_ms=max(0, _grace_ms),
)
multiplier=_mult or None, grace_ms=max(0, _grace_ms))
if wav_path and self._voice_barge_capture.is_set():
self._voice_submit_barge_utterance(wav_path)
else:
@@ -464,8 +457,7 @@ class CLIVoiceMixin:
from tools.voice_mode import is_tts_echo
if is_tts_echo(transcript, getattr(self, "_voice_last_tts_text", "")):
logger.debug(
"Dropping playback-phase barge transcript as TTS echo: %r", transcript
)
"Dropping playback-phase barge transcript as TTS echo: %r", transcript)
_cprint(f"\n{_DIM}Ignored likely TTS echo (not queued).{_RST}")
return
self._pending_input.put(_VoiceInputMessage(transcript))
@@ -617,8 +609,7 @@ class CLIVoiceMixin:
say = _cprint if announce else (lambda *_a: None)
try:
from tools.wake_word import (
check_wake_word_requirements, load_wake_word_config, owns_listener, start_listening
)
check_wake_word_requirements, load_wake_word_config, owns_listener, start_listening)
except Exception as e:
say(f"{_DIM}Wake word unavailable: {e}{_RST}")
return False
@@ -749,8 +740,7 @@ class CLIVoiceMixin:
self._agent_running
or self._voice_recording
or getattr(self, "_voice_processing", False)
or not self._pending_input.empty()
)
or not self._pending_input.empty())
if busy:
idle_polls = 0
continue
@@ -777,8 +767,7 @@ class CLIVoiceMixin:
from cli import _ACCENT, _BOLD, _DIM, _RST, _cprint
from tools.wake_word import (
audio_is_silent, check_wake_word_requirements, is_listening, load_wake_word_config,
owns_listener,
)
owns_listener)
cfg = load_wake_word_config()
reqs = check_wake_word_requirements(cfg)
hermes_cli/clipboard.py
+2 -5
View File
@@ -50,8 +50,7 @@ def _linux_backends():
return (
(_is_wsl(), _wsl_has_image, _wsl_save),
(bool(os.environ.get("WAYLAND_DISPLAY")), _wayland_has_image, _wayland_save),
(True, _xclip_has_image, _xclip_save),
)
(True, _xclip_has_image, _xclip_save))
def save_clipboard_image(dest: Path) -> bool:
@@ -206,9 +205,7 @@ _PS_IMAGE_STRATEGIES = (
_PS_FILEDROP_HIT
+ "if ($null -eq $hit) { exit 1 }"
"[System.Convert]::ToBase64String([System.IO.File]::ReadAllBytes($hit))"
"} catch { exit 1 }",
),
)
"} catch { exit 1 }"))
def _ps_clipboard(exe: str, timeout: int, label: str, dest: Path | None = None) -> bool:
hermes_cli/codex_models.py
+5 -10
View File
@@ -28,8 +28,7 @@ DEFAULT_CODEX_MODELS: List[str] = [
# not in the public API, so it stays out of the "openai" catalog in hermes_cli/models.py.
# The backend reports ``supported_in_api: false`` for it; that flag describes API
# availability, not Codex availability, so fetch/cache paths must not filter on it.
"gpt-5.3-codex-spark",
]
"gpt-5.3-codex-spark"]
_FORWARD_COMPAT_TEMPLATE_MODELS: List[tuple[str, tuple[str, ...]]] = [
("gpt-5.6-sol", ("gpt-5.5", "gpt-5.4")),
@@ -40,8 +39,7 @@ _FORWARD_COMPAT_TEMPLATE_MODELS: List[tuple[str, tuple[str, ...]]] = [
("gpt-5.4", ("gpt-5.3-codex",)),
# Spark surfaces whenever a compatible template is present; the backend (not Hermes)
# gates real availability by ChatGPT Pro entitlement.
("gpt-5.3-codex-spark", ("gpt-5.3-codex",)),
]
("gpt-5.3-codex-spark", ("gpt-5.3-codex",))]
def _dedupe(model_ids) -> List[str]:
@@ -103,8 +101,7 @@ def _extract_chatgpt_account_id(access_token: str) -> Optional[str]:
acct_id = (
claims.get("https://api.openai.com/auth", {}).get("chatgpt_account_id")
if isinstance(claims, dict)
else None
)
else None)
return acct_id if isinstance(acct_id, str) and acct_id else None
except Exception:
return None
@@ -145,8 +142,7 @@ def _fetch_models_from_api(access_token: str) -> List[str]:
resp = httpx.get(
"https://chatgpt.com/backend-api/codex/models?client_version=1.0.0",
headers=headers,
timeout=10,
)
timeout=10)
if resp.status_code != 200:
return []
data = resp.json()
@@ -194,5 +190,4 @@ def get_codex_model_ids(access_token: Optional[str] = None) -> List[str]:
default_model = _read_default_model(codex_home)
return _finalize_codex_models(_dedupe([
*([default_model] if default_model else []), *_read_cache_models(codex_home),
*DEFAULT_CODEX_MODELS,
]))
*DEFAULT_CODEX_MODELS]))
hermes_cli/codex_runtime_plugin_migration.py
+2 -4
View File
@@ -70,8 +70,7 @@ _KEYS_DROPPED_WITH_WARNING = {"sampling"}
# (hermes key, codex key, skip note) — timeouts are emitted as floats or skipped when non-numeric.
_TIMEOUT_KEYS = (
("timeout", "tool_timeout_sec", "timeout (not numeric)"),
("connect_timeout", "startup_timeout_sec", "connect_timeout (not numeric)"),
)
("connect_timeout", "startup_timeout_sec", "connect_timeout (not numeric)"))
def _str_map(d: dict) -> dict[str, str]:
@@ -130,8 +129,7 @@ def _translate_one_server(name: str, hermes_cfg: dict) -> tuple[Optional[dict],
# env-var passthrough (HERMES_HOME, PYTHONPATH) could carry one in pathological cases.
_TOML_ESCAPES = (
("\\", "\\\\"), ('"', '\\"'), ("\b", "\\b"), ("\t", "\\t"),
("\n", "\\n"), ("\f", "\\f"), ("\r", "\\r"),
)
("\n", "\\n"), ("\f", "\\f"), ("\r", "\\r"))
def _escape_toml_string(value: str) -> str: