Compare commits

..

25 Commits

Author SHA1 Message Date
Shannon Sands c1ef64a0ac feat(secrets): add phase 1 secrets tool and redaction hardening
Implements the first pragmatic slice of issue #3627 / #410:
- add agent-facing  tool with list/check/request/delete/inject
  actions
- reuse existing secure CLI secret capture path via getpass-backed callback
  so secret values never enter model context
- support  as an alias for the existing
   skill frontmatter
- redact execute_code stdout/stderr before returning tool output
- expand redaction patterns for Twilio SIDs and JWTs
- register the new tool in discovery/core toolsets and add regression tests

Gateway DM+delete secret capture remains scoped as follow-up work per the
Phase 1 issue discussion.
2026-03-31 10:03:02 +10:00
Teknium f007284d05 fix: rate-limit pairing rejection messages to prevent spam (#4081)
* fix: rate-limit pairing rejection messages to prevent spam

When generate_code() returns None (rate limited or max pending), the
"Too many pairing requests" message was sent on every subsequent DM
with no cooldown. A user sending 30 messages would get 30 rejection
replies — reported as potential hack on WhatsApp.

Now check _is_rate_limited() before any pairing response, and record
rate limit after sending a rejection. Subsequent messages from the
same user are silently ignored until the rate limit window expires.

* test: add coverage for pairing response rate limiting

Follow-up to cherry-picked PR #4042 — adds tests verifying:
- Rate-limited users get silently ignored (no response sent)
- Rejection messages record rate limit for subsequent suppression

---------

Co-authored-by: 0xbyt4 <35742124+0xbyt4@users.noreply.github.com>
2026-03-30 16:48:00 -07:00
Teknium 3d47af01c3 fix(honcho): write config to instance-local path for profile isolation (#4037)
Multiple agents/profiles running 'hermes honcho setup' all wrote to
the shared global ~/.honcho/config.json, overwriting each other's
configuration.

Root cause: _write_config() defaulted to resolve_config_path() which
returns the global path when no instance-local file exists yet (i.e.
on first setup).

Fix: _write_config() now defaults to _local_config_path() which always
returns $HERMES_HOME/honcho.json. Each profile gets its own config file.
Reading still falls back to global for cross-app interop and seeding.

Also updates cmd_setup and cmd_status messaging to show the actual
write path.

Includes 10 new tests verifying profile isolation, global fallback
reads, and multi-profile independence.
2026-03-30 16:41:19 -07:00
SHL0MS 275fcc6673 Merge pull request #4054 from NousResearch/ascii-video/text-readability-and-layout-oracle
ascii-video skill: text readability techniques and external layout oracle
2026-03-30 15:52:14 -07:00
SHL0MS ab62614a89 ascii-video: add text readability techniques and external layout oracle pattern
- composition.md: add text backdrop (gaussian dark mask behind glyphs) and
  external layout oracle pattern (browser-based text layout → JSON → Python
  renderer pipeline for obstacle-aware text reflow)
- shaders.md: add reverse vignette shader (center-darkening for text readability)
- troubleshooting.md: add diagnostic entries for text-over-busy-background
  readability and kaleidoscope-destroys-text pitfall
2026-03-30 18:48:22 -04:00
Teknium de368cac54 fix(tools): show browser and TTS in reconfigure menu (#4041)
* fix(gateway): honor default for invalid bool-like config values

* refactor: simplify web backend priority detection

Replace cascading boolean conditions with a priority-ordered loop.
Same behavior (verified against all 16 env var combinations),
half the lines, trivially extensible for new backends.

* fix(tools): show browser and TTS in reconfigure menu

_toolset_has_keys() returned False for toolsets with no-key providers
(Local Browser, Edge TTS) because it only checked providers with
env_vars. Users couldn't find these tools in the reconfigure list
and had no obvious way to switch browser/TTS backends.

Now treats providers with empty env_vars as always-configured, so
toolsets with free/local options always appear in the reconfigure menu.

---------

Co-authored-by: aydnOktay <xaydinoktay@gmail.com>
2026-03-30 14:11:39 -07:00
Teknium 0d1003559d refactor: simplify web backend priority detection (#4036)
* fix(gateway): honor default for invalid bool-like config values

* refactor: simplify web backend priority detection

Replace cascading boolean conditions with a priority-ordered loop.
Same behavior (verified against all 16 env var combinations),
half the lines, trivially extensible for new backends.

---------

Co-authored-by: aydnOktay <xaydinoktay@gmail.com>
2026-03-30 13:37:25 -07:00
Teknium eba8d52d54 fix: show correct shell config path for macOS/zsh in install script (#4025)
- print_success() hardcoded 'source ~/.bashrc' regardless of user's shell
- On macOS (default zsh), ~/.bashrc doesn't exist, leaving users unable to
  find the hermes command after install
- Now detects $SHELL and shows the correct file (zshrc/bashrc)
- Also captures .[all] install failure output instead of silencing with
  2>/dev/null, so users can diagnose why full extras failed
2026-03-30 13:25:11 -07:00
Teknium 72104eb06f fix(gateway): honor default for invalid bool-like config values (#4029)
Co-authored-by: aydnOktay <xaydinoktay@gmail.com>
2026-03-30 13:24:48 -07:00
Teknium 4b35836ba4 fix(auth): use bearer auth for MiniMax Anthropic endpoints (#4028)
MiniMax's /anthropic endpoints implement Anthropic's Messages API but
require Authorization: Bearer instead of x-api-key. Without this fix,
MiniMax users get 401 errors in gateway sessions.

Adds _requires_bearer_auth() to detect MiniMax endpoints and route
through auth_token in the Anthropic SDK. Check runs before OAuth
token detection so MiniMax keys aren't misclassified as setup tokens.

Co-authored-by: kshitijk4poor <kshitijk4poor@users.noreply.github.com>
2026-03-30 13:21:39 -07:00
Teknium bd376fe976 fix(docs): improve mobile sidebar navigation
The sidebar had all categories expanded by default (collapsed: false),
which on mobile created a 60+ item flat list when opening the sidebar.
Reported by danny on Discord.

Changes:
- Set all top-level categories to collapsed: true (tap to expand)
- Enable autoCollapseCategories: true (accordion — opening one section
  closes others, prevents the overwhelming flat list)
- Enable hideable sidebar (swipe-to-dismiss on mobile)
- Add mobile CSS: larger touch targets (0.75rem padding), bolder
  category headers, visible subcategory indentation with left border,
  wider sidebar (85vw / 360px max), darker backdrop overlay
2026-03-30 13:20:55 -07:00
Teknium f93637b3a1 feat: add /profile slash command to show active profile (#4027)
Adds /profile to COMMAND_REGISTRY (Info category) with handlers in
both CLI and gateway. Shows the active profile name and home directory.

Works on all platforms — CLI, Telegram, Discord, Slack, etc.
Detects profile by checking if HERMES_HOME is under ~/.hermes/profiles/.
Shows 'default' when running without a profile.
2026-03-30 13:20:06 -07:00
Teknium 7b4fe0528f fix(auth): use bearer auth for MiniMax Anthropic endpoints (#4028)
MiniMax's /anthropic endpoints implement Anthropic's Messages API but
require Authorization: Bearer instead of x-api-key. Without this fix,
MiniMax users get 401 errors in gateway sessions.

Adds _requires_bearer_auth() to detect MiniMax endpoints and route
through auth_token in the Anthropic SDK. Check runs before OAuth
token detection so MiniMax keys aren't misclassified as setup tokens.

Co-authored-by: kshitijk4poor <kshitijk4poor@users.noreply.github.com>
2026-03-30 13:19:44 -07:00
Teknium 950f69475f feat(browser): add Camofox local anti-detection browser backend (#4008)
Camofox-browser is a self-hosted Node.js server wrapping Camoufox
(Firefox fork with C++ fingerprint spoofing). When CAMOFOX_URL is set,
all 11 browser tools route through the Camofox REST API instead of
the agent-browser CLI.

Maps 1:1 to the existing browser tool interface:
- Navigate, snapshot, click, type, scroll, back, press, close
- Get images, vision (screenshot + LLM analysis)
- Console (returns empty with note — camofox limitation)

Setup: npm start in camofox-browser dir, or docker run -p 9377:9377
Then: CAMOFOX_URL=http://localhost:9377 in ~/.hermes/.env

Advantages over Browserbase (cloud):
- Free (no per-session API costs)
- Local (zero network latency for browser ops)
- Anti-detection at C++ level (bypasses Cloudflare/Google bot detection)
- Works offline, Docker-ready

Files:
- tools/browser_camofox.py: Full REST backend (~400 lines)
- tools/browser_tool.py: Routing at each tool function
- hermes_cli/config.py: CAMOFOX_URL env var entry
- tests/tools/test_browser_camofox.py: 20 tests
2026-03-30 13:18:42 -07:00
Teknium 7dac75f2ae fix: prevent context pressure warning spam after compression (#4012)
* feat: add /yolo slash command to toggle dangerous command approvals

Adds a /yolo command that toggles HERMES_YOLO_MODE at runtime, skipping
all dangerous command approval prompts for the current session. Works in
both CLI and gateway (Telegram, Discord, etc.).

- /yolo -> ON: all commands auto-approved, no confirmation prompts
- /yolo -> OFF: normal approval flow restored

The --yolo CLI flag already existed for launch-time opt-in. This adds
the ability to toggle mid-session without restarting.

Session-scoped — resets when the process ends. Uses the existing
HERMES_YOLO_MODE env var that check_all_command_guards() already
respects.

* fix: prevent context pressure warning spam (agent loop + gateway rate-limit)

Two complementary fixes for repeated context pressure warnings spamming
gateway users (Telegram, Discord, etc.):

1. Agent-level loop fix (run_agent.py):
   After compression, only reset _context_pressure_warned if the
   post-compression estimate is actually below the 85% warning level.
   Previously the flag was unconditionally reset, causing the warning
   to re-fire every loop iteration when compression couldn't reduce
   below 85% of the threshold (e.g. very low threshold like 15%,
   or system prompt alone exceeds the warning level).

2. Gateway-level rate-limit (gateway/run.py, salvaged from PR #3786):
   Per-chat_id cooldown of 1 hour on compression warning messages.
   Both warning paths ('still large after compression' and 'compression
   failed') are gated. Defense-in-depth — even if the agent-level fix
   has edge cases, users won't see more than one warning per hour.

Co-authored-by: dlkakbs <dlkakbs@users.noreply.github.com>

---------

Co-authored-by: dlkakbs <dlkakbs@users.noreply.github.com>
2026-03-30 13:18:21 -07:00
Teknium ed9af6e589 fix: create AsyncOpenAI lazily in trajectory_compressor to avoid closed event loop (#4013)
The AsyncOpenAI client was created once at __init__ and stored as an
instance attribute. process_directory() calls asyncio.run() which creates
and closes a fresh event loop. On a second call, the client's httpx
transport is still bound to the closed loop, raising RuntimeError:
"Event loop is closed" — the same pattern fixed by PR #3398 for the
main agent loop.

Create the client lazily in _get_async_client() so each asyncio.run()
gets a client bound to the current loop.

Co-authored-by: binhnt92 <binhnt.ht.92@gmail.com>
2026-03-30 13:16:16 -07:00
Teknium 158f49f19a fix: enforce priority order in Telegram menu — core > plugins > skills (#4023)
The menu now has explicit priority tiers:
1. Core CommandDef commands (always included, never bumped)
2. Plugin slash commands (take precedence over skills)
3. Built-in skill commands (fill remaining slots alphabetically)

Only skills get trimmed when the 100-command cap is hit. Adding new
core commands or plugin commands automatically pushes skills out,
not the other way around.
2026-03-30 13:04:06 -07:00
Teknium 86250a3e45 docs: expand terminal backends section + fix docs build (#4016)
* feat(telegram): add webhook mode as alternative to polling

When TELEGRAM_WEBHOOK_URL is set, the adapter starts an HTTP webhook
server (via python-telegram-bot's start_webhook()) instead of long
polling. This enables cloud platforms like Fly.io and Railway to
auto-wake suspended machines on inbound HTTP traffic.

Polling remains the default — no behavior change unless the env var
is set.

Env vars:
  TELEGRAM_WEBHOOK_URL    Public HTTPS URL for Telegram to push to
  TELEGRAM_WEBHOOK_PORT   Local listen port (default 8443)
  TELEGRAM_WEBHOOK_SECRET Secret token for update verification

Cherry-picked and adapted from PR #2022 by SHL0MS. Preserved all
current main enhancements (network error recovery, polling conflict
detection, DM topics setup).

Co-authored-by: SHL0MS <SHL0MS@users.noreply.github.com>

* fix: send_document call in background task delivery + vision download timeout

Two fixes salvaged from PR #2269 by amethystani:

1. gateway/run.py: adapter.send_file() → adapter.send_document()
   send_file() doesn't exist on BasePlatformAdapter. Background task
   media files were silently never delivered (AttributeError swallowed
   by except Exception: pass).

2. tools/vision_tools.py: configurable image download timeout via
   HERMES_VISION_DOWNLOAD_TIMEOUT env var (default 30s), plus guard
   against raise None when max_retries=0.

The third fix in #2269 (opencode-go auth config) was already resolved
on main.

Co-authored-by: amethystani <amethystani@users.noreply.github.com>

* docs: expand terminal backends section + fix feishu MDX build error

---------

Co-authored-by: SHL0MS <SHL0MS@users.noreply.github.com>
Co-authored-by: amethystani <amethystani@users.noreply.github.com>
2026-03-30 12:59:58 -07:00
Teknium ea342f2382 Fix banner alignment in installer script (#4011)
Co-authored-by: Ahmed Khaled <wakeupwithme000@gmail.com>
2026-03-30 11:24:10 -07:00
Teknium 60ecde8ac7 fix: fit all 100 commands in Telegram menu with 40-char descriptions (#4010)
* fix: truncate skill descriptions to 100 chars in Telegram menu

* fix: 40-char desc cap + 100 command limit for Telegram menu

setMyCommands has an undocumented total payload size limit.
50 commands with 256-char descriptions failed, 50 with 100-char
worked, and 100 with 40-char descriptions also works (~5300 total
chars). Truncate skill descriptions to 40 chars in the menu picker
and set cap back to 100. Full descriptions available via /commands.
2026-03-30 11:21:13 -07:00
Teknium f3069c649c fix(cli): add missing subprocess.run() timeouts in doctor and status (#4009)
Add timeout parameters to 4 subprocess.run() calls that could hang
indefinitely if the child process blocks (e.g., unresponsive docker
daemon, systemctl waiting for D-Bus):

- doctor.py: docker info (timeout=10), ssh check (timeout=15)
- status.py: systemctl is-active (timeout=5), launchctl list (timeout=5)

Each call site now catches subprocess.TimeoutExpired and treats it as
a failure, consistent with how non-zero return codes are already handled.

Add AST-based regression test that verifies every subprocess.run() call
in CLI modules specifies a timeout keyword argument.

Co-authored-by: dieutx <dangtc94@gmail.com>
2026-03-30 11:17:15 -07:00
Teknium 0976bf6cd0 feat: add /yolo slash command to toggle dangerous command approvals (#3990)
Adds a /yolo command that toggles HERMES_YOLO_MODE at runtime, skipping
all dangerous command approval prompts for the current session. Works in
both CLI and gateway (Telegram, Discord, etc.).

- /yolo -> ON: all commands auto-approved, no confirmation prompts
- /yolo -> OFF: normal approval flow restored

The --yolo CLI flag already existed for launch-time opt-in. This adds
the ability to toggle mid-session without restarting.

Session-scoped — resets when the process ends. Uses the existing
HERMES_YOLO_MODE env var that check_all_command_guards() already
respects.
2026-03-30 11:17:09 -07:00
Teknium da3e22bcfa fix: cap Telegram menu at 50 commands — API rejects above ~60 (#4006)
* fix: use SKILLS_DIR not repo path for Telegram menu skill filter

Skills are synced to ~/.hermes/skills/ (SKILLS_DIR), not the repo's
skills/ directory. The previous filter compared against the repo path
so no skills matched. Now checks SKILLS_DIR and excludes .hub/
subdirectory (user-installed hub skills).

* fix: cap Telegram menu at 50 commands — API rejects above ~60

Telegram's setMyCommands returns BOT_COMMANDS_TOO_MUCH when
registering close to 100 commands despite docs claiming 100 is the
limit. Metadata overhead causes rejection above ~60. Cap at 50 for
reliability — remaining commands accessible via /commands.
2026-03-30 11:05:20 -07:00
Teknium 9fd78c7a8e fix: use SKILLS_DIR not repo path for Telegram menu skill filter (#4005)
Skills are synced to ~/.hermes/skills/ (SKILLS_DIR), not the repo's
skills/ directory. The previous filter compared against the repo path
so no skills matched. Now checks SKILLS_DIR and excludes .hub/
subdirectory (user-installed hub skills).
2026-03-30 11:01:13 -07:00
Teknium 5ceed021dc feat(gateway): skill-aware slash commands, paginated /commands, Telegram 100-cap (#3934)
* feat(gateway): skill-aware slash commands, paginated /commands, Telegram 100-cap

Map active skills to Telegram's slash command menu so users can
discover and invoke skills directly. Three changes:

1. Telegram menu now includes active skill commands alongside built-in
   commands, capped at 100 entries (Telegram Bot API limit). Overflow
   commands remain callable but hidden from the picker. Logged at
   startup when cap is hit.

2. New /commands [page] gateway command for paginated browsing of all
   commands + skills. /help now shows first 10 skill commands and
   points to /commands for the full list.

3. When a user types a slash command that matches a disabled or
   uninstalled skill, they get actionable guidance:
   - Disabled: 'Enable it with: hermes skills config'
   - Optional (not installed): 'Install with: hermes skills install official/<path>'

Built on ideas from PR #3921 by @kshitijk4poor.

* chore: move 21 niche skills to optional-skills

Move specialized/niche skills from built-in (skills/) to optional
(optional-skills/) to reduce the default skill count. Users can
install them with: hermes skills install official/<category>/<name>

Moved skills (21):
- mlops: accelerate, chroma, faiss, flash-attention,
  hermes-atropos-environments, huggingface-tokenizers, instructor,
  lambda-labs, llava, nemo-curator, pinecone, pytorch-lightning,
  qdrant, saelens, simpo, slime, tensorrt-llm, torchtitan
- research: domain-intel, duckduckgo-search
- devops: inference-sh cli

Built-in skills: 96 → 75
Optional skills: 22 → 43

* fix: only include repo built-in skills in Telegram menu, not user-installed

User-installed skills (from hub or manually added) stay accessible via
/skills and by typing the command directly, but don't get registered
in the Telegram slash command picker. Only skills whose SKILL.md is
under the repo's skills/ directory are included in the menu.

This keeps the Telegram menu focused on the curated built-in set while
user-installed skills remain discoverable through /skills and /commands.
2026-03-30 10:57:30 -07:00
121 changed files with 2731 additions and 1356 deletions
+26 -1
View File
@@ -162,6 +162,21 @@ def _is_oauth_token(key: str) -> bool:
return True
def _requires_bearer_auth(base_url: str | None) -> bool:
"""Return True for Anthropic-compatible providers that require Bearer auth.
Some third-party /anthropic endpoints implement Anthropic's Messages API but
require Authorization: Bearer instead of Anthropic's native x-api-key header.
MiniMax's global and China Anthropic-compatible endpoints follow this pattern.
"""
if not base_url:
return False
normalized = base_url.rstrip("/").lower()
return normalized.startswith("https://api.minimax.io/anthropic") or normalized.startswith(
"https://api.minimaxi.com/anthropic"
)
def build_anthropic_client(api_key: str, base_url: str = None):
"""Create an Anthropic client, auto-detecting setup-tokens vs API keys.
@@ -180,7 +195,17 @@ def build_anthropic_client(api_key: str, base_url: str = None):
if base_url:
kwargs["base_url"] = base_url
if _is_oauth_token(api_key):
if _requires_bearer_auth(base_url):
# Some Anthropic-compatible providers (e.g. MiniMax) expect the API key in
# Authorization: Bearer even for regular API keys. Route those endpoints
# through auth_token so the SDK sends Bearer auth instead of x-api-key.
# Check this before OAuth token shape detection because MiniMax secrets do
# not use Anthropic's sk-ant-api prefix and would otherwise be misread as
# Anthropic OAuth/setup tokens.
kwargs["auth_token"] = api_key
if _COMMON_BETAS:
kwargs["default_headers"] = {"anthropic-beta": ",".join(_COMMON_BETAS)}
elif _is_oauth_token(api_key):
# OAuth access token / setup-token → Bearer auth + Claude Code identity.
# Anthropic routes OAuth requests based on user-agent and headers;
# without Claude Code's fingerprint, requests get intermittent 500s.
+19
View File
@@ -40,6 +40,8 @@ _PREFIX_PATTERNS = [
r"sk_[A-Za-z0-9_]{10,}", # ElevenLabs TTS key (sk_ underscore, not sk- dash)
r"tvly-[A-Za-z0-9]{10,}", # Tavily search API key
r"exa_[A-Za-z0-9]{10,}", # Exa search API key
r"AC[a-fA-F0-9]{32}", # Twilio Account SID
r"SK[a-fA-F0-9]{32}", # Twilio API Key SID / Secret SID-like identifiers
]
# ENV assignment patterns: KEY=value where KEY contains a secret-like name
@@ -68,6 +70,17 @@ _TELEGRAM_RE = re.compile(
r"(bot)?(\d{8,}):([-A-Za-z0-9_]{30,})",
)
# JWTs: three base64url-ish segments separated by dots.
# Keep threshold moderately high to avoid redacting short dotted identifiers.
_JWT_RE = re.compile(
r"\b([A-Za-z0-9_-]{12,}\.[A-Za-z0-9_-]{12,}\.[A-Za-z0-9_-]{12,})\b"
)
# Twilio auth tokens are commonly plain 32-char lowercase hex strings.
# This may also match some MD5-like identifiers, but we prefer false positives
# over leaking a credential into model context.
_TWILIO_AUTH_TOKEN_RE = re.compile(r"\b([a-f0-9]{32})\b")
# Private key blocks: -----BEGIN RSA PRIVATE KEY----- ... -----END RSA PRIVATE KEY-----
_PRIVATE_KEY_RE = re.compile(
r"-----BEGIN[A-Z ]*PRIVATE KEY-----[\s\S]*?-----END[A-Z ]*PRIVATE KEY-----"
@@ -140,6 +153,12 @@ def redact_sensitive_text(text: str) -> str:
return f"{prefix}{digits}:***"
text = _TELEGRAM_RE.sub(_redact_telegram, text)
# JWTs
text = _JWT_RE.sub(lambda m: _mask_token(m.group(1)), text)
# Twilio auth tokens / bare 32-char lowercase hex tokens
text = _TWILIO_AUTH_TOKEN_RE.sub(lambda m: _mask_token(m.group(1)), text)
# Private key blocks
text = _PRIVATE_KEY_RE.sub("[REDACTED PRIVATE KEY]", text)
+41
View File
@@ -492,6 +492,7 @@ from cron import get_job
from tools.terminal_tool import cleanup_all_environments as _cleanup_all_terminals
from tools.terminal_tool import set_sudo_password_callback, set_approval_callback
from tools.skills_tool import set_secret_capture_callback
from tools.secrets_tool import set_secrets_request_callback
from hermes_cli.callbacks import prompt_for_secret
from tools.browser_tool import _emergency_cleanup_all_sessions as _cleanup_all_browsers
@@ -2837,6 +2838,28 @@ class HermesCLI:
print(" Example: python cli.py --toolsets web,terminal")
print()
def _handle_profile_command(self):
"""Display active profile name and home directory."""
from hermes_constants import get_hermes_home, display_hermes_home
home = get_hermes_home()
display = display_hermes_home()
profiles_parent = Path.home() / ".hermes" / "profiles"
try:
rel = home.relative_to(profiles_parent)
profile_name = str(rel).split("/")[0]
except ValueError:
profile_name = None
print()
if profile_name:
print(f" Profile: {profile_name}")
else:
print(" Profile: default")
print(f" Home: {display}")
print()
def show_config(self):
"""Display current configuration with kawaii ASCII art."""
# Get terminal config from environment (which was set from cli-config.yaml)
@@ -3679,6 +3702,8 @@ class HermesCLI:
return False
elif canonical == "help":
self.show_help()
elif canonical == "profile":
self._handle_profile_command()
elif canonical == "tools":
self._handle_tools_command(cmd_original)
elif canonical == "toolsets":
@@ -3836,6 +3861,8 @@ class HermesCLI:
self.console.print(f" Status bar {state}")
elif canonical == "verbose":
self._toggle_verbose()
elif canonical == "yolo":
self._toggle_yolo()
elif canonical == "reasoning":
self._handle_reasoning_command(cmd_original)
elif canonical == "compress":
@@ -4434,6 +4461,17 @@ class HermesCLI:
}
_cprint(labels.get(self.tool_progress_mode, ""))
def _toggle_yolo(self):
"""Toggle YOLO mode — skip all dangerous command approval prompts."""
import os
current = bool(os.environ.get("HERMES_YOLO_MODE"))
if current:
os.environ.pop("HERMES_YOLO_MODE", None)
self.console.print(" ⚠ YOLO mode [bold red]OFF[/] — dangerous commands will require approval.")
else:
os.environ["HERMES_YOLO_MODE"] = "1"
self.console.print(" ⚡ YOLO mode [bold green]ON[/] — all commands auto-approved. Use with caution.")
def _handle_reasoning_command(self, cmd: str):
"""Handle /reasoning — manage effort level and display toggle.
@@ -5550,6 +5588,7 @@ class HermesCLI:
# Single-query and direct chat callers do not go through run(), so
# register secure secret capture here as well.
set_secret_capture_callback(self._secret_capture_callback)
set_secrets_request_callback(self._secret_capture_callback)
# Refresh provider credentials if needed (handles key rotation transparently)
if not self._ensure_runtime_credentials():
@@ -6255,6 +6294,7 @@ class HermesCLI:
set_sudo_password_callback(self._sudo_password_callback)
set_approval_callback(self._approval_callback)
set_secret_capture_callback(self._secret_capture_callback)
set_secrets_request_callback(self._secret_capture_callback)
# Ensure tirith security scanner is available (downloads if needed).
# Warn the user if tirith is enabled in config but not available,
@@ -7458,6 +7498,7 @@ class HermesCLI:
set_sudo_password_callback(None)
set_approval_callback(None)
set_secret_capture_callback(None)
set_secrets_request_callback(None)
# Flush + shut down Honcho async writer (drains queue before exit)
if self.agent and getattr(self.agent, '_honcho', None):
try:
-1
View File
@@ -13,7 +13,6 @@ Core layers:
Concrete environments:
- terminal_test_env/: Simple file-creation tasks for testing the stack
- hermes_swe_env/: SWE-bench style tasks with Modal sandboxes
- endless_terminals/: Terminal tasks from HuggingFace dataset with Apptainer containers
Benchmarks (eval-only):
- benchmarks/terminalbench_2/: Terminal-Bench 2.0 evaluation
@@ -1,5 +0,0 @@
"""Endless Terminals Environment - Terminal task training from HuggingFace dataset."""
from .endless_terminals_env import EndlessTerminalsEnv, EndlessTerminalsEnvConfig
__all__ = ["EndlessTerminalsEnv", "EndlessTerminalsEnvConfig"]
File diff suppressed because it is too large Load Diff
@@ -1,91 +0,0 @@
# Endless Terminals - Qwen3-4B-Instruct-2507
# Single config for both trainer (launch_training.py) and env (endless_terminals_env.py serve)
#
# Usage:
# Terminal 1: run-api
# Terminal 2: cd tinker-atropos && python launch_training.py --config ../environments/endless_terminals/tinker_qwen.yaml
# Terminal 3: python environments/endless_terminals/endless_terminals_env.py serve --config environments/endless_terminals/tinker_qwen.yaml
env:
# Toolsets
enabled_toolsets: ["terminal", "file"]
# Model / tokenizer
tokenizer_name: "Qwen/Qwen3-4B-Instruct-2507"
# Agent configuration
max_agent_turns: 16
max_token_length: 2048
agent_temperature: 0.6
extra_body:
chat_template_kwargs:
enable_thinking: false
tool_call_parser: "hermes"
# Terminal backend
terminal_backend: "docker"
# Dataset settings
use_dataset: true
dataset_name: "obiwan96/endless-terminals"
dataset_split: "train"
dataset_cache_dir: "~/.cache/huggingface/datasets"
tasks_base_dir: "/Users/samherring/Desktop/Projects/Hermes-Agent/endless-terminals"
# Test execution
test_timeout_s: 180
default_docker_image: "ubuntu:22.04"
max_concurrent_containers: 16
# Training configuration
group_size: 16
batch_size: 64 # 4 groups × 16 rollouts per step
total_steps: 500
steps_per_eval: 5
min_items_sent_before_logging: 1
ensure_scores_are_not_same: true
max_num_workers: 2048
worker_timeout: 3600
inference_weight: 1.0
eval_limit_ratio: 0.1
rollout_server_url: "http://localhost:8000"
# Evaluation configuration
num_eval_tasks: 20
eval_split_ratio: 0.1
# Logging
use_wandb: true
wandb_name: "endless-terminals-qwen3-4b"
# System prompt
system_prompt: >
You are a skilled Linux system administrator and programmer.
You have access to a terminal and file tools to complete system administration
and programming tasks. Use the tools effectively to solve the given task,
and verify your solution works correctly before finishing.
Keep each command short and focused — break complex tasks into multiple steps
rather than writing long one-liners.
tinker:
lora_rank: 32
learning_rate: 0.0000005
max_token_trainer_length: 32768
checkpoint_dir: "./temp/"
save_checkpoint_interval: 50
wandb_project: "endless-terminals"
wandb_group: null
wandb_run_name: "qwen3-4b"
tool_call_parser: "hermes"
openai:
- model_name: "Qwen/Qwen3-4B-Instruct-2507"
base_url: "http://localhost:8001/v1"
api_key: "x"
weight: 1.0
num_requests_for_eval: 64
timeout: 600
server_type: "sglang"
slurm: false
testing: false
+1
View File
@@ -298,6 +298,7 @@ class HermesAgentBaseEnv(BaseEnv):
return False
server = self.server.servers[0]
# If the server is an OpenAI server (not VLLM/SGLang), use direct mode
from atroposlib.envs.server_handling.openai_server import OpenAIServer
return not isinstance(server, OpenAIServer)
@@ -48,13 +48,7 @@ class HermesToolCallParser(ToolCallParser):
if not raw_json.strip():
continue
try:
tc_data = json.loads(raw_json)
except json.JSONDecodeError:
# Fix invalid backslash escapes from shell commands in JSON strings
# e.g. \s \w \d \n (unescaped) → \\s \\w \\d \\n
fixed = re.sub(r'\\([^"\\/bfnrtu0-9\n])', r'\\\\\1', raw_json)
tc_data = json.loads(fixed)
tc_data = json.loads(raw_json)
tool_calls.append(
ChatCompletionMessageToolCall(
id=f"call_{uuid.uuid4().hex[:8]}",
+9 -2
View File
@@ -27,9 +27,16 @@ def _coerce_bool(value: Any, default: bool = True) -> bool:
return default
if isinstance(value, bool):
return value
if isinstance(value, int):
return value != 0
if isinstance(value, str):
return value.strip().lower() in ("true", "1", "yes", "on")
return bool(value)
lowered = value.strip().lower()
if lowered in ("true", "1", "yes", "on"):
return True
if lowered in ("false", "0", "no", "off"):
return False
return default
return default
def _normalize_unauthorized_dm_behavior(value: Any, default: str = "pair") -> str:
+11 -2
View File
@@ -622,10 +622,19 @@ class TelegramAdapter(BasePlatformAdapter):
# gateway command there automatically adds it to the Telegram menu.
try:
from telegram import BotCommand
from hermes_cli.commands import telegram_bot_commands
from hermes_cli.commands import telegram_menu_commands
# Telegram allows up to 100 commands but has an undocumented
# payload size limit. Skill descriptions are truncated to 40
# chars in telegram_menu_commands() to fit 100 commands safely.
menu_commands, hidden_count = telegram_menu_commands(max_commands=100)
await self._bot.set_my_commands([
BotCommand(name, desc) for name, desc in telegram_bot_commands()
BotCommand(name, desc) for name, desc in menu_commands
])
if hidden_count:
logger.info(
"[%s] Telegram menu: %d commands registered, %d hidden (over 100 limit). Use /commands for full list.",
self.name, len(menu_commands), hidden_count,
)
except Exception as e:
logger.warning(
"[%s] Could not register Telegram command menu: %s",
+182 -4
View File
@@ -301,6 +301,50 @@ def _resolve_runtime_agent_kwargs() -> dict:
}
def _check_unavailable_skill(command_name: str) -> str | None:
"""Check if a command matches a known-but-inactive skill.
Returns a helpful message if the skill exists but is disabled or only
available as an optional install. Returns None if no match found.
"""
# Normalize: command uses hyphens, skill names may use hyphens or underscores
normalized = command_name.lower().replace("_", "-")
try:
from tools.skills_tool import SKILLS_DIR, _get_disabled_skill_names
disabled = _get_disabled_skill_names()
# Check disabled built-in skills
for skill_md in SKILLS_DIR.rglob("SKILL.md"):
if any(part in ('.git', '.github', '.hub') for part in skill_md.parts):
continue
name = skill_md.parent.name.lower().replace("_", "-")
if name == normalized and name in disabled:
return (
f"The **{command_name}** skill is installed but disabled.\n"
f"Enable it with: `hermes skills config`"
)
# Check optional skills (shipped with repo but not installed)
from hermes_constants import get_hermes_home
repo_root = Path(__file__).resolve().parent.parent
optional_dir = repo_root / "optional-skills"
if optional_dir.exists():
for skill_md in optional_dir.rglob("SKILL.md"):
name = skill_md.parent.name.lower().replace("_", "-")
if name == normalized:
# Build install path: official/<category>/<name>
rel = skill_md.parent.relative_to(optional_dir)
parts = list(rel.parts)
install_path = f"official/{'/'.join(parts)}"
return (
f"The **{command_name}** skill is available but not installed.\n"
f"Install it with: `hermes skills install {install_path}`"
)
except Exception:
pass
return None
def _platform_config_key(platform: "Platform") -> str:
"""Map a Platform enum to its config.yaml key (LOCAL→"cli", rest→enum value)."""
return "cli" if platform == Platform.LOCAL else platform.value
@@ -432,6 +476,13 @@ class GatewayRunner:
self._honcho_managers: Dict[str, Any] = {}
self._honcho_configs: Dict[str, Any] = {}
# Rate-limit compression warning messages sent to users.
# Keyed by chat_id — value is the timestamp of the last warning sent.
# Prevents the warning from firing on every message when a session
# remains above the threshold after compression.
self._compression_warn_sent: Dict[str, float] = {}
self._compression_warn_cooldown: int = 3600 # seconds (1 hour)
# Ensure tirith security scanner is available (downloads if needed)
try:
from tools.tirith_security import ensure_installed
@@ -1651,6 +1702,11 @@ class GatewayRunner:
# In DMs: offer pairing code. In groups: silently ignore.
if source.chat_type == "dm" and self._get_unauthorized_dm_behavior(source.platform) == "pair":
platform_name = source.platform.value if source.platform else "unknown"
# Rate-limit ALL pairing responses (code or rejection) to
# prevent spamming the user with repeated messages when
# multiple DMs arrive in quick succession.
if self.pairing_store._is_rate_limited(platform_name, source.user_id):
return None
code = self.pairing_store.generate_code(
platform_name, source.user_id, source.user_name or ""
)
@@ -1672,6 +1728,8 @@ class GatewayRunner:
"Too many pairing requests right now~ "
"Please try again later!"
)
# Record rate limit so subsequent messages are silently ignored
self.pairing_store._record_rate_limit(platform_name, source.user_id)
return None
# PRIORITY handling when an agent is already running for this session.
@@ -1817,7 +1875,13 @@ class GatewayRunner:
if canonical == "help":
return await self._handle_help_command(event)
if canonical == "commands":
return await self._handle_commands_command(event)
if canonical == "profile":
return await self._handle_profile_command(event)
if canonical == "status":
return await self._handle_status_command(event)
@@ -1830,6 +1894,9 @@ class GatewayRunner:
if canonical == "verbose":
return await self._handle_verbose_command(event)
if canonical == "yolo":
return await self._handle_yolo_command(event)
if canonical == "provider":
return await self._handle_provider_command(event)
@@ -1974,6 +2041,12 @@ class GatewayRunner:
if msg:
event.text = msg
# Fall through to normal message processing with skill content
else:
# Not an active skill — check if it's a known-but-disabled or
# uninstalled skill and give actionable guidance.
_unavail_msg = _check_unavailable_skill(command)
if _unavail_msg:
return _unavail_msg
except Exception as e:
logger.debug("Skill command check failed (non-fatal): %s", e)
@@ -2344,13 +2417,18 @@ class GatewayRunner:
pass
# Still too large after compression — warn user
# Rate-limited to once per cooldown period per
# chat to avoid spamming on every message.
if _new_tokens >= _warn_token_threshold:
logger.warning(
"Session hygiene: still ~%s tokens after "
"compression — suggesting /reset",
f"{_new_tokens:,}",
)
if _hyg_adapter:
_now = time.time()
_last_warn = self._compression_warn_sent.get(source.chat_id, 0)
if _hyg_adapter and _now - _last_warn >= self._compression_warn_cooldown:
self._compression_warn_sent[source.chat_id] = _now
try:
await _hyg_adapter.send(
source.chat_id,
@@ -2372,7 +2450,10 @@ class GatewayRunner:
if _approx_tokens >= _warn_token_threshold:
_hyg_adapter = self.adapters.get(source.platform)
_hyg_meta = {"thread_id": source.thread_id} if source.thread_id else None
if _hyg_adapter:
_now = time.time()
_last_warn = self._compression_warn_sent.get(source.chat_id, 0)
if _hyg_adapter and _now - _last_warn >= self._compression_warn_cooldown:
self._compression_warn_sent[source.chat_id] = _now
try:
await _hyg_adapter.send(
source.chat_id,
@@ -2999,6 +3080,36 @@ class GatewayRunner:
return f"{header}\n\n{session_info}"
return header
async def _handle_profile_command(self, event: MessageEvent) -> str:
"""Handle /profile — show active profile name and home directory."""
from hermes_constants import get_hermes_home, display_hermes_home
from pathlib import Path
home = get_hermes_home()
display = display_hermes_home()
# Detect profile name from HERMES_HOME path
# Profile paths look like: ~/.hermes/profiles/<name>
profiles_parent = Path.home() / ".hermes" / "profiles"
try:
rel = home.relative_to(profiles_parent)
profile_name = str(rel).split("/")[0]
except ValueError:
profile_name = None
if profile_name:
lines = [
f"👤 **Profile:** `{profile_name}`",
f"📂 **Home:** `{display}`",
]
else:
lines = [
"👤 **Profile:** default",
f"📂 **Home:** `{display}`",
]
return "\n".join(lines)
async def _handle_status_command(self, event: MessageEvent) -> str:
"""Handle /status command."""
source = event.source
@@ -3065,12 +3176,69 @@ class GatewayRunner:
from agent.skill_commands import get_skill_commands
skill_cmds = get_skill_commands()
if skill_cmds:
lines.append(f"\n⚡ **Skill Commands** ({len(skill_cmds)} installed):")
for cmd in sorted(skill_cmds):
lines.append(f"\n⚡ **Skill Commands** ({len(skill_cmds)} active):")
# Show first 10, then point to /commands for the rest
sorted_cmds = sorted(skill_cmds)
for cmd in sorted_cmds[:10]:
lines.append(f"`{cmd}` — {skill_cmds[cmd]['description']}")
if len(sorted_cmds) > 10:
lines.append(f"\n... and {len(sorted_cmds) - 10} more. Use `/commands` for the full paginated list.")
except Exception:
pass
return "\n".join(lines)
async def _handle_commands_command(self, event: MessageEvent) -> str:
"""Handle /commands [page] - paginated list of all commands and skills."""
from hermes_cli.commands import gateway_help_lines
raw_args = event.get_command_args().strip()
if raw_args:
try:
requested_page = int(raw_args)
except ValueError:
return "Usage: `/commands [page]`"
else:
requested_page = 1
# Build combined entry list: built-in commands + skill commands
entries = list(gateway_help_lines())
try:
from agent.skill_commands import get_skill_commands
skill_cmds = get_skill_commands()
if skill_cmds:
entries.append("")
entries.append("⚡ **Skill Commands**:")
for cmd in sorted(skill_cmds):
desc = skill_cmds[cmd].get("description", "").strip() or "Skill command"
entries.append(f"`{cmd}` — {desc}")
except Exception:
pass
if not entries:
return "No commands available."
from gateway.config import Platform
page_size = 15 if event.source.platform == Platform.TELEGRAM else 20
total_pages = max(1, (len(entries) + page_size - 1) // page_size)
page = max(1, min(requested_page, total_pages))
start = (page - 1) * page_size
page_entries = entries[start:start + page_size]
lines = [
f"📚 **Commands** ({len(entries)} total, page {page}/{total_pages})",
"",
*page_entries,
]
if total_pages > 1:
nav_parts = []
if page > 1:
nav_parts.append(f"`/commands {page - 1}` ← prev")
if page < total_pages:
nav_parts.append(f"next → `/commands {page + 1}`")
lines.extend(["", " | ".join(nav_parts)])
if page != requested_page:
lines.append(f"_(Requested page {requested_page} was out of range, showing page {page}.)_")
return "\n".join(lines)
async def _handle_provider_command(self, event: MessageEvent) -> str:
"""Handle /provider command - show available providers."""
@@ -3999,6 +4167,16 @@ class GatewayRunner:
else:
return f"🧠 ✓ Reasoning effort set to `{effort}` (this session only)"
async def _handle_yolo_command(self, event: MessageEvent) -> str:
"""Handle /yolo — toggle dangerous command approval bypass."""
current = bool(os.environ.get("HERMES_YOLO_MODE"))
if current:
os.environ.pop("HERMES_YOLO_MODE", None)
return "⚠️ YOLO mode **OFF** — dangerous commands will require approval."
else:
os.environ["HERMES_YOLO_MODE"] = "1"
return "⚡ YOLO mode **ON** — all commands auto-approved. Use with caution."
async def _handle_verbose_command(self, event: MessageEvent) -> str:
"""Handle /verbose command — cycle tool progress display mode.
+68
View File
@@ -71,6 +71,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
aliases=("q",), args_hint="<prompt>"),
CommandDef("status", "Show session info", "Session",
gateway_only=True),
CommandDef("profile", "Show active profile name and home directory", "Info"),
CommandDef("sethome", "Set this chat as the home channel", "Session",
gateway_only=True, aliases=("set-home",)),
CommandDef("resume", "Resume a previously-named session", "Session",
@@ -90,6 +91,8 @@ COMMAND_REGISTRY: list[CommandDef] = [
CommandDef("verbose", "Cycle tool progress display: off -> new -> all -> verbose",
"Configuration", cli_only=True,
gateway_config_gate="display.tool_progress_command"),
CommandDef("yolo", "Toggle YOLO mode (skip all dangerous command approvals)",
"Configuration"),
CommandDef("reasoning", "Manage reasoning effort and display", "Configuration",
args_hint="[level|show|hide]",
subcommands=("none", "low", "minimal", "medium", "high", "xhigh", "show", "hide", "on", "off")),
@@ -118,6 +121,8 @@ COMMAND_REGISTRY: list[CommandDef] = [
"Tools & Skills", cli_only=True),
# Info
CommandDef("commands", "Browse all commands and skills (paginated)", "Info",
gateway_only=True, args_hint="[page]"),
CommandDef("help", "Show available commands", "Info"),
CommandDef("usage", "Show token usage for the current session", "Info"),
CommandDef("insights", "Show usage insights and analytics", "Info",
@@ -361,6 +366,69 @@ def telegram_bot_commands() -> list[tuple[str, str]]:
return result
def telegram_menu_commands(max_commands: int = 100) -> tuple[list[tuple[str, str]], int]:
"""Return Telegram menu commands capped to the Bot API limit.
Priority order (higher priority = never bumped by overflow):
1. Core CommandDef commands (always included)
2. Plugin slash commands (take precedence over skills)
3. Built-in skill commands (fill remaining slots, alphabetical)
Skills are the only tier that gets trimmed when the cap is hit.
User-installed hub skills are excluded — accessible via /skills.
Returns:
(menu_commands, hidden_count) where hidden_count is the number of
skill commands omitted due to the cap.
"""
all_commands = list(telegram_bot_commands())
# Plugin slash commands get priority over skills
try:
from hermes_cli.plugins import get_plugin_manager
pm = get_plugin_manager()
plugin_cmds = getattr(pm, "_plugin_commands", {})
for cmd_name in sorted(plugin_cmds):
tg_name = cmd_name.replace("-", "_")
desc = "Plugin command"
if len(desc) > 40:
desc = desc[:37] + "..."
all_commands.append((tg_name, desc))
except Exception:
pass
# Remaining slots go to built-in skill commands (not hub-installed).
skill_entries: list[tuple[str, str]] = []
try:
from agent.skill_commands import get_skill_commands
from tools.skills_tool import SKILLS_DIR
_skills_dir = str(SKILLS_DIR.resolve())
_hub_dir = str((SKILLS_DIR / ".hub").resolve())
skill_cmds = get_skill_commands()
for cmd_key in sorted(skill_cmds):
info = skill_cmds[cmd_key]
skill_path = info.get("skill_md_path", "")
if not skill_path.startswith(_skills_dir):
continue
if skill_path.startswith(_hub_dir):
continue
name = cmd_key.lstrip("/").replace("-", "_")
desc = info.get("description", "")
# Keep descriptions short — setMyCommands has an undocumented
# total payload limit. 40 chars fits 100 commands safely.
if len(desc) > 40:
desc = desc[:37] + "..."
skill_entries.append((name, desc))
except Exception:
pass
# Skills fill remaining slots — they're the only tier that gets trimmed
remaining_slots = max(0, max_commands - len(all_commands))
hidden_count = max(0, len(skill_entries) - remaining_slots)
all_commands.extend(skill_entries[:remaining_slots])
return all_commands[:max_commands], hidden_count
def slack_subcommand_map() -> dict[str, str]:
"""Return subcommand -> /command mapping for Slack /hermes handler.
+8
View File
@@ -706,6 +706,14 @@ OPTIONAL_ENV_VARS = {
"password": True,
"category": "tool",
},
"CAMOFOX_URL": {
"description": "Camofox browser server URL for local anti-detection browsing (e.g. http://localhost:9377)",
"prompt": "Camofox server URL",
"url": "https://github.com/jo-inc/camofox-browser",
"tools": ["browser_navigate", "browser_click"],
"password": False,
"category": "tool",
},
"FAL_KEY": {
"description": "FAL API key for image generation",
"prompt": "FAL API key",
+15 -8
View File
@@ -406,8 +406,11 @@ def run_doctor(args):
if terminal_env == "docker":
if shutil.which("docker"):
# Check if docker daemon is running
result = subprocess.run(["docker", "info"], capture_output=True)
if result.returncode == 0:
try:
result = subprocess.run(["docker", "info"], capture_output=True, timeout=10)
except subprocess.TimeoutExpired:
result = None
if result is not None and result.returncode == 0:
check_ok("docker", "(daemon running)")
else:
check_fail("docker daemon not running")
@@ -426,12 +429,16 @@ def run_doctor(args):
ssh_host = os.getenv("TERMINAL_SSH_HOST")
if ssh_host:
# Try to connect
result = subprocess.run(
["ssh", "-o", "ConnectTimeout=5", "-o", "BatchMode=yes", ssh_host, "echo ok"],
capture_output=True,
text=True
)
if result.returncode == 0:
try:
result = subprocess.run(
["ssh", "-o", "ConnectTimeout=5", "-o", "BatchMode=yes", ssh_host, "echo ok"],
capture_output=True,
text=True,
timeout=15
)
except subprocess.TimeoutExpired:
result = None
if result is not None and result.returncode == 0:
check_ok(f"SSH connection to {ssh_host}")
else:
check_fail(f"SSH connection to {ssh_host}")
+4 -2
View File
@@ -601,13 +601,15 @@ def _print_setup_summary(config: dict, hermes_home):
Path(__file__).parent.parent / "node_modules" / ".bin" / "agent-browser"
).exists()
)
if get_env_value("BROWSERBASE_API_KEY"):
if get_env_value("CAMOFOX_URL"):
tool_status.append(("Browser Automation (Camofox)", True, None))
elif get_env_value("BROWSERBASE_API_KEY"):
tool_status.append(("Browser Automation (Browserbase)", True, None))
elif _ab_found:
tool_status.append(("Browser Automation (local)", True, None))
else:
tool_status.append(
("Browser Automation", False, "npm install -g agent-browser")
("Browser Automation", False, "npm install -g agent-browser or set CAMOFOX_URL")
)
# FAL (image generation)
+20 -12
View File
@@ -285,23 +285,31 @@ def show_status(args):
_gw_svc = get_service_name()
except Exception:
_gw_svc = "hermes-gateway"
result = subprocess.run(
["systemctl", "--user", "is-active", _gw_svc],
capture_output=True,
text=True
)
is_active = result.stdout.strip() == "active"
try:
result = subprocess.run(
["systemctl", "--user", "is-active", _gw_svc],
capture_output=True,
text=True,
timeout=5
)
is_active = result.stdout.strip() == "active"
except subprocess.TimeoutExpired:
is_active = False
print(f" Status: {check_mark(is_active)} {'running' if is_active else 'stopped'}")
print(" Manager: systemd (user)")
elif sys.platform == 'darwin':
from hermes_cli.gateway import get_launchd_label
result = subprocess.run(
["launchctl", "list", get_launchd_label()],
capture_output=True,
text=True
)
is_loaded = result.returncode == 0
try:
result = subprocess.run(
["launchctl", "list", get_launchd_label()],
capture_output=True,
text=True,
timeout=5
)
is_loaded = result.returncode == 0
except subprocess.TimeoutExpired:
is_loaded = False
print(f" Status: {check_mark(is_loaded)} {'loaded' if is_loaded else 'not loaded'}")
print(" Manager: launchd")
else:
+35 -1
View File
@@ -273,6 +273,16 @@ TOOL_CATEGORIES = {
"browser_provider": "browser-use",
"post_setup": "browserbase",
},
{
"name": "Camofox",
"tag": "Local anti-detection browser (Firefox/Camoufox)",
"env_vars": [
{"key": "CAMOFOX_URL", "prompt": "Camofox server URL", "default": "http://localhost:9377",
"url": "https://github.com/jo-inc/camofox-browser"},
],
"browser_provider": "camofox",
"post_setup": "camofox",
},
],
},
"homeassistant": {
@@ -337,6 +347,28 @@ def _run_post_setup(post_setup_key: str):
elif not node_modules.exists():
_print_warning(" Node.js not found - browser tools require: npm install (in hermes-agent directory)")
elif post_setup_key == "camofox":
camofox_dir = PROJECT_ROOT / "node_modules" / "@askjo" / "camoufox-browser"
if not camofox_dir.exists() and shutil.which("npm"):
_print_info(" Installing Camofox browser server...")
import subprocess
result = subprocess.run(
["npm", "install", "--silent"],
capture_output=True, text=True, cwd=str(PROJECT_ROOT)
)
if result.returncode == 0:
_print_success(" Camofox installed")
else:
_print_warning(" npm install failed - run manually: npm install")
if camofox_dir.exists():
_print_info(" Start the Camofox server:")
_print_info(" npx @askjo/camoufox-browser")
_print_info(" First run downloads the Camoufox engine (~300MB)")
_print_info(" Or use Docker: docker run -p 9377:9377 jo-inc/camofox-browser")
elif not shutil.which("npm"):
_print_warning(" Node.js not found. Install Camofox via Docker:")
_print_info(" docker run -p 9377:9377 jo-inc/camofox-browser")
elif post_setup_key == "rl_training":
try:
__import__("tinker_atropos")
@@ -565,7 +597,9 @@ def _toolset_has_keys(ts_key: str) -> bool:
if cat:
for provider in cat.get("providers", []):
env_vars = provider.get("env_vars", [])
if env_vars and all(get_env_value(e["key"]) for e in env_vars):
if not env_vars:
return True # No-key provider (e.g. Local Browser, Edge TTS)
if all(get_env_value(e["key"]) for e in env_vars):
return True
return False
+22 -8
View File
@@ -10,16 +10,27 @@ import os
import sys
from pathlib import Path
from hermes_constants import get_hermes_home
from honcho_integration.client import resolve_config_path, GLOBAL_CONFIG_PATH
HOST = "hermes"
def _config_path() -> Path:
"""Return the active Honcho config path (instance-local or global)."""
"""Return the active Honcho config path for reading (instance-local or global)."""
return resolve_config_path()
def _local_config_path() -> Path:
"""Return the instance-local Honcho config path for writing.
Always returns $HERMES_HOME/honcho.json so each profile/instance gets
its own config file. The global ~/.honcho/config.json is only used as
a read fallback (via resolve_config_path) for cross-app interop.
"""
return get_hermes_home() / "honcho.json"
def _read_config() -> dict:
path = _config_path()
if path.exists():
@@ -31,7 +42,7 @@ def _read_config() -> dict:
def _write_config(cfg: dict, path: Path | None = None) -> None:
path = path or _config_path()
path = path or _local_config_path()
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(
json.dumps(cfg, indent=2, ensure_ascii=False) + "\n",
@@ -95,13 +106,13 @@ def cmd_setup(args) -> None:
"""Interactive Honcho setup wizard."""
cfg = _read_config()
active_path = _config_path()
write_path = _local_config_path()
read_path = _config_path()
print("\nHoncho memory setup\n" + "" * 40)
print(" Honcho gives Hermes persistent cross-session memory.")
if active_path != GLOBAL_CONFIG_PATH:
print(f" Instance config: {active_path}")
else:
print(" Config is shared with other hosts at ~/.honcho/config.json")
print(f" Config: {write_path}")
if read_path != write_path and read_path.exists():
print(f" (seeding from existing config at {read_path})")
print()
if not _ensure_sdk_installed():
@@ -189,7 +200,7 @@ def cmd_setup(args) -> None:
hermes_host.setdefault("saveMessages", True)
_write_config(cfg)
print(f"\n Config written to {active_path}")
print(f"\n Config written to {write_path}")
# Test connection
print(" Testing connection... ", end="", flush=True)
@@ -237,6 +248,7 @@ def cmd_status(args) -> None:
cfg = _read_config()
active_path = _config_path()
write_path = _local_config_path()
if not cfg:
print(f" No Honcho config found at {active_path}")
@@ -259,6 +271,8 @@ def cmd_status(args) -> None:
print(f" Workspace: {hcfg.workspace_id}")
print(f" Host: {hcfg.host}")
print(f" Config path: {active_path}")
if write_path != active_path:
print(f" Write path: {write_path} (instance-local)")
print(f" AI peer: {hcfg.ai_peer}")
print(f" User peer: {hcfg.peer_name or 'not set'}")
print(f" Session key: {hcfg.resolve_session_name()}")
+1
View File
@@ -150,6 +150,7 @@ def _discover_tools():
"tools.tts_tool",
"tools.todo_tool",
"tools.memory_tool",
"tools.secrets_tool",
"tools.session_search_tool",
"tools.clarify_tool",
"tools.code_execution_tool",
+2 -1
View File
@@ -16,7 +16,8 @@
},
"homepage": "https://github.com/NousResearch/Hermes-Agent#readme",
"dependencies": {
"agent-browser": "^0.13.0"
"agent-browser": "^0.13.0",
"@askjo/camoufox-browser": "^1.0.0"
},
"engines": {
"node": ">=18.0.0"
+12 -5
View File
@@ -5221,11 +5221,8 @@ class AIAgent:
except Exception as e:
logger.warning("Session DB compression split failed — new session will NOT be indexed: %s", e)
# Reset context pressure warning and token estimate — usage drops
# after compaction. Without this, the stale last_prompt_tokens from
# the previous API call causes the pressure calculation to stay at
# >1000% and spam warnings / re-trigger compression in a loop.
self._context_pressure_warned = False
# Update token estimate after compaction so pressure calculations
# use the post-compression count, not the stale pre-compression one.
_compressed_est = (
estimate_tokens_rough(new_system_prompt)
+ estimate_messages_tokens_rough(compressed)
@@ -5233,6 +5230,16 @@ class AIAgent:
self.context_compressor.last_prompt_tokens = _compressed_est
self.context_compressor.last_completion_tokens = 0
# Only reset the pressure warning if compression actually brought
# us below the warning level (85% of threshold). When compression
# can't reduce enough (e.g. threshold is very low, or system prompt
# alone exceeds the warning level), keep the flag set to prevent
# spamming the user with repeated warnings every loop iteration.
if self.context_compressor.threshold_tokens > 0:
_post_progress = _compressed_est / self.context_compressor.threshold_tokens
if _post_progress < 0.85:
self._context_pressure_warned = False
return compressed, new_system_prompt
def _execute_tool_calls(self, assistant_message, messages: list, effective_task_id: str, api_call_count: int = 0) -> None:
+15 -3
View File
@@ -94,7 +94,7 @@ print_banner() {
echo ""
echo -e "${MAGENTA}${BOLD}"
echo "┌─────────────────────────────────────────────────────────┐"
echo "│ ⚕ Hermes Agent Installer │"
echo "│ ⚕ Hermes Agent Installer │"
echo "├─────────────────────────────────────────────────────────┤"
echo "│ An open source AI agent by Nous Research. │"
echo "└─────────────────────────────────────────────────────────┘"
@@ -699,14 +699,19 @@ install_deps() {
# Install the main package in editable mode with all extras.
# Try [all] first, fall back to base install if extras have issues.
if ! $UV_CMD pip install -e ".[all]" 2>/dev/null; then
ALL_INSTALL_LOG=$(mktemp)
if ! $UV_CMD pip install -e ".[all]" 2>"$ALL_INSTALL_LOG"; then
log_warn "Full install (.[all]) failed, trying base install..."
log_info "Reason: $(tail -5 "$ALL_INSTALL_LOG" | head -3)"
rm -f "$ALL_INSTALL_LOG"
if ! $UV_CMD pip install -e "."; then
log_error "Package installation failed."
log_info "Check that build tools are installed: sudo apt install build-essential python3-dev"
log_info "Then re-run: cd $INSTALL_DIR && uv pip install -e '.[all]'"
exit 1
fi
else
rm -f "$ALL_INSTALL_LOG"
fi
log_success "Main package installed"
@@ -1070,7 +1075,14 @@ print_success() {
echo ""
echo -e "${YELLOW}⚡ Reload your shell to use 'hermes' command:${NC}"
echo ""
echo " source ~/.bashrc # or ~/.zshrc"
LOGIN_SHELL="$(basename "${SHELL:-/bin/bash}")"
if [ "$LOGIN_SHELL" = "zsh" ]; then
echo " source ~/.zshrc"
elif [ "$LOGIN_SHELL" = "bash" ]; then
echo " source ~/.bashrc"
else
echo " source ~/.bashrc # or ~/.zshrc"
fi
echo ""
# Show Node.js warning if auto-install failed
@@ -744,3 +744,149 @@ class PixelBlendStack:
result = blend_canvas(result, canvas, mode, opacity)
return result
```
## Text Backdrop (Readability Mask)
When placing readable text over busy multi-grid ASCII backgrounds, the text will blend into the background and become illegible. **Always apply a dark backdrop behind text regions.**
The technique: compute the bounding box of all text glyphs, create a gaussian-blurred dark mask covering that area with padding, and multiply the background by `(1 - mask * darkness)` before rendering text on top.
```python
from scipy.ndimage import gaussian_filter
def apply_text_backdrop(canvas, glyphs, padding=80, darkness=0.75):
"""Darken the background behind text for readability.
Call AFTER rendering background, BEFORE rendering text.
Args:
canvas: (VH, VW, 3) uint8 background
glyphs: list of {"x": float, "y": float, ...} glyph positions
padding: pixel padding around text bounding box
darkness: 0.0 = no darkening, 1.0 = fully black
Returns:
darkened canvas (uint8)
"""
if not glyphs:
return canvas
xs = [g['x'] for g in glyphs]
ys = [g['y'] for g in glyphs]
x0 = max(0, int(min(xs)) - padding)
y0 = max(0, int(min(ys)) - padding)
x1 = min(VW, int(max(xs)) + padding + 50) # extra for char width
y1 = min(VH, int(max(ys)) + padding + 60) # extra for char height
# Soft dark mask with gaussian blur for feathered edges
mask = np.zeros((VH, VW), dtype=np.float32)
mask[y0:y1, x0:x1] = 1.0
mask = gaussian_filter(mask, sigma=padding * 0.6)
factor = 1.0 - mask * darkness
return (canvas.astype(np.float32) * factor[:, :, np.newaxis]).astype(np.uint8)
```
### Usage in render pipeline
Insert between background rendering and text rendering:
```python
# 1. Render background (multi-grid ASCII effects)
bg = render_background(cfg, t)
# 2. Darken behind text region
bg = apply_text_backdrop(bg, frame_glyphs, padding=80, darkness=0.75)
# 3. Render text on top (now readable against dark backdrop)
bg = text_renderer.render(bg, frame_glyphs, color=(255, 255, 255))
```
Combine with **reverse vignette** (see shaders.md) for scenes where text is always centered — the reverse vignette provides a persistent center-dark zone, while the backdrop handles per-frame glyph positions.
## External Layout Oracle Pattern
For text-heavy videos where text needs to dynamically reflow around obstacles (shapes, icons, other text), use an external layout engine to pre-compute glyph positions and feed them into the Python renderer via JSON.
### Architecture
```
Layout Engine (browser/Node.js) → layouts.json → Python ASCII Renderer
↑ ↑
Computes per-frame Reads glyph positions,
glyph (x,y) positions renders as ASCII chars
with obstacle-aware reflow with full effect pipeline
```
### JSON interchange format
```json
{
"meta": {
"canvas_width": 1080, "canvas_height": 1080,
"fps": 24, "total_frames": 1248,
"fonts": {
"body": {"charW": 12.04, "charH": 24, "fontSize": 20},
"hero": {"charW": 24.08, "charH": 48, "fontSize": 40}
}
},
"scenes": [
{
"id": "scene_name",
"start_frame": 0, "end_frame": 96,
"frames": {
"0": {
"glyphs": [
{"char": "H", "x": 287.1, "y": 400.0, "alpha": 1.0},
{"char": "e", "x": 311.2, "y": 400.0, "alpha": 1.0}
],
"obstacles": [
{"type": "circle", "cx": 540, "cy": 540, "r": 80},
{"type": "rect", "x": 300, "y": 500, "w": 120, "h": 80}
]
}
}
}
]
}
```
### When to use
- Text that dynamically reflows around moving objects
- Per-glyph animation (reveal, scatter, physics)
- Variable typography that needs precise measurement
- Any case where Python's Pillow text layout is insufficient
### When NOT to use
- Static centered text (just use PIL `draw.text()` directly)
- Text that only fades in/out without spatial animation
- Simple typewriter effects (handle in Python with a character counter)
### Running the oracle
Use Playwright to run the layout engine in a headless browser:
```javascript
// extract.mjs
import { chromium } from 'playwright';
const browser = await chromium.launch({ headless: true });
const page = await browser.newPage();
await page.goto(`file://${oraclePath}`);
await page.waitForFunction(() => window.__ORACLE_DONE__ === true, null, { timeout: 60000 });
const result = await page.evaluate(() => window.__ORACLE_RESULT__);
writeFileSync('layouts.json', JSON.stringify(result));
await browser.close();
```
### Consuming in Python
```python
# In the renderer, map pixel positions to the canvas:
for glyph in frame_data['glyphs']:
char, px, py = glyph['char'], glyph['x'], glyph['y']
alpha = glyph.get('alpha', 1.0)
# Render using PIL draw.text() at exact pixel position
draw.text((px, py), char, fill=(int(255*alpha),)*3, font=font)
```
Obstacles from the JSON can also be rendered as glowing ASCII shapes (circles, rectangles) to visualize the reflow zones.
@@ -834,6 +834,39 @@ def sh_vignette(c, s=0.22):
return np.clip(c * _vig_cache[k][:,:,None], 0, 255).astype(np.uint8)
```
#### Reverse Vignette
Inverted vignette: darkens the **center** and leaves edges bright. Useful when text is centered over busy backgrounds — creates a natural dark zone for readability without a hard-edged box.
Combine with `apply_text_backdrop()` (see composition.md) for per-frame glyph-aware darkening.
```python
_rvignette_cache = {}
def sh_reverse_vignette(c, strength=0.5):
"""Center darkening, edge brightening. Cached."""
k = ('rv', c.shape[0], c.shape[1], round(strength, 2))
if k not in _rvignette_cache:
h, w = c.shape[:2]
Y = np.linspace(-1, 1, h)[:, None]
X = np.linspace(-1, 1, w)[None, :]
d = np.sqrt(X**2 + Y**2)
# Invert: bright at edges, dark at center
mask = np.clip(1.0 - (1.0 - d * 0.7) * strength, 0.2, 1.0)
_rvignette_cache[k] = mask[:, :, np.newaxis].astype(np.float32)
return np.clip(c.astype(np.float32) * _rvignette_cache[k], 0, 255).astype(np.uint8)
```
| Param | Default | Effect |
|-------|---------|--------|
| `strength` | 0.5 | 0 = no effect, 1.0 = center nearly black |
Add to ShaderChain dispatch:
```python
elif name == "reverse_vignette":
return sh_reverse_vignette(canvas, kwargs.get("strength", 0.5))
```
#### Contrast
```python
def sh_contrast(c, factor=1.3):
@@ -14,6 +14,8 @@
| Random dark holes in output | Font missing Unicode glyphs | Validate palettes at init |
| Audio-visual desync | Frame timing accumulation | Use integer frame counter, compute t fresh each frame |
| Single-color flat output | Hue field shape mismatch | Ensure h,s,v arrays all (rows,cols) before hsv2rgb |
| Text unreadable over busy bg | No contrast between text and background | Use `apply_text_backdrop()` (composition.md) + `reverse_vignette` shader (shaders.md) |
| Text garbled/mirrored | Kaleidoscope or mirror shader applied to text scene | **Never apply kaleidoscope, mirror_h/v/quad/diag to scenes with readable text** — radial folding destroys legibility. Apply these only to background layers or text-free scenes |
Common bugs, gotchas, and platform-specific issues encountered during ASCII video development.
+15
View File
@@ -52,6 +52,21 @@ class TestKnownPrefixes:
result = redact_sensitive_text("fal_abc123def456ghi789jkl")
assert "abc123def456" not in result
def test_twilio_account_sid(self):
sid = "AC" + ("1" * 16) + ("a" * 16)
result = redact_sensitive_text(sid)
assert sid not in result
def test_twilio_auth_token_bare(self):
token = ("0" * 16) + ("a" * 16)
result = redact_sensitive_text(token)
assert token not in result
def test_jwt_bare(self):
jwt = "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJhIjoiYiIsImMiOiJkIiwicm9sZSI6ImFkbWluIn0.c2lnbmF0dXJlMTIzNDU2Nzg5MGFiY2RlZg"
result = redact_sensitive_text(jwt)
assert "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9" not in result
def test_short_token_fully_masked(self):
result = redact_sensitive_text("key=sk-short1234567")
assert "***" in result
+40
View File
@@ -207,6 +207,46 @@ Generate some audio.
assert len(calls) == 1
assert calls[0][0] == "TENOR_API_KEY"
def test_requires_secrets_alias_triggers_secure_capture(self, tmp_path, monkeypatch):
monkeypatch.delenv("TENOR_API_KEY", raising=False)
calls = []
def fake_secret_callback(var_name, prompt, metadata=None):
calls.append((var_name, prompt, metadata))
os.environ[var_name] = "stored-in-test"
return {
"success": True,
"stored_as": var_name,
"validated": False,
"skipped": False,
}
monkeypatch.setattr(
skills_tool_module,
"_secret_capture_callback",
fake_secret_callback,
raising=False,
)
with patch("tools.skills_tool.SKILLS_DIR", tmp_path):
_make_skill(
tmp_path,
"test-skill",
frontmatter_extra=(
"requires_secrets:\n"
" - key: TENOR_API_KEY\n"
" description: Tenor API key\n"
" instructions: Find it in the Tenor dashboard\n"
),
)
scan_skill_commands()
msg = build_skill_invocation_message("/test-skill", "do stuff")
assert msg is not None
assert len(calls) == 1
assert calls[0][0] == "TENOR_API_KEY"
assert "tenor" in (calls[0][2].get("required_for", "") or "").lower()
def test_gateway_still_loads_skill_but_returns_setup_guidance(
self, tmp_path, monkeypatch
):
+43
View File
@@ -212,6 +212,49 @@ class TestSessionHygieneWarnThreshold:
assert post_compress_tokens < warn_threshold
class TestCompressionWarnRateLimit:
"""Compression warning messages must be rate-limited per chat_id."""
def _make_runner(self):
from unittest.mock import MagicMock, patch
with patch("gateway.run.load_gateway_config"), \
patch("gateway.run.SessionStore"), \
patch("gateway.run.DeliveryRouter"):
from gateway.run import GatewayRunner
runner = GatewayRunner.__new__(GatewayRunner)
runner._compression_warn_sent = {}
runner._compression_warn_cooldown = 3600
return runner
def test_first_warn_is_sent(self):
runner = self._make_runner()
now = 1_000_000.0
last = runner._compression_warn_sent.get("chat:1", 0)
assert now - last >= runner._compression_warn_cooldown
def test_second_warn_suppressed_within_cooldown(self):
runner = self._make_runner()
now = 1_000_000.0
runner._compression_warn_sent["chat:1"] = now - 60 # 1 minute ago
last = runner._compression_warn_sent.get("chat:1", 0)
assert now - last < runner._compression_warn_cooldown
def test_warn_allowed_after_cooldown(self):
runner = self._make_runner()
now = 1_000_000.0
runner._compression_warn_sent["chat:1"] = now - 3601 # just past cooldown
last = runner._compression_warn_sent.get("chat:1", 0)
assert now - last >= runner._compression_warn_cooldown
def test_rate_limit_is_per_chat(self):
"""Rate-limiting one chat must not suppress warnings for another."""
runner = self._make_runner()
now = 1_000_000.0
runner._compression_warn_sent["chat:1"] = now - 60 # suppressed
last_other = runner._compression_warn_sent.get("chat:2", 0)
assert now - last_other >= runner._compression_warn_cooldown
class TestEstimatedTokenThreshold:
"""Verify that hygiene thresholds are always below the model's context
limit for both actual and estimated token counts.
@@ -60,6 +60,7 @@ def _make_runner(platform: Platform, config: GatewayConfig):
runner.adapters = {platform: adapter}
runner.pairing_store = MagicMock()
runner.pairing_store.is_approved.return_value = False
runner.pairing_store._is_rate_limited.return_value = False
return runner, adapter
@@ -142,6 +143,56 @@ async def test_unauthorized_whatsapp_dm_can_be_ignored(monkeypatch):
adapter.send.assert_not_awaited()
@pytest.mark.asyncio
async def test_rate_limited_user_gets_no_response(monkeypatch):
"""When a user is already rate-limited, pairing messages are silently ignored."""
_clear_auth_env(monkeypatch)
config = GatewayConfig(
platforms={Platform.WHATSAPP: PlatformConfig(enabled=True)},
)
runner, adapter = _make_runner(Platform.WHATSAPP, config)
runner.pairing_store._is_rate_limited.return_value = True
result = await runner._handle_message(
_make_event(
Platform.WHATSAPP,
"15551234567@s.whatsapp.net",
"15551234567@s.whatsapp.net",
)
)
assert result is None
runner.pairing_store.generate_code.assert_not_called()
adapter.send.assert_not_awaited()
@pytest.mark.asyncio
async def test_rejection_message_records_rate_limit(monkeypatch):
"""After sending a 'too many requests' rejection, rate limit is recorded
so subsequent messages are silently ignored."""
_clear_auth_env(monkeypatch)
config = GatewayConfig(
platforms={Platform.WHATSAPP: PlatformConfig(enabled=True)},
)
runner, adapter = _make_runner(Platform.WHATSAPP, config)
runner.pairing_store.generate_code.return_value = None # triggers rejection
result = await runner._handle_message(
_make_event(
Platform.WHATSAPP,
"15551234567@s.whatsapp.net",
"15551234567@s.whatsapp.net",
)
)
assert result is None
adapter.send.assert_awaited_once()
assert "Too many" in adapter.send.await_args.args[1]
runner.pairing_store._record_rate_limit.assert_called_once_with(
"whatsapp", "15551234567@s.whatsapp.net"
)
@pytest.mark.asyncio
async def test_global_ignore_suppresses_pairing_reply(monkeypatch):
_clear_auth_env(monkeypatch)

Some files were not shown because too many files have changed in this diff Show More