The mock handler checked for function_name == 'search' but the RPC
sends 'search_files'. Any test exercising search_files through the
mock would get 'Unknown tool' instead of the canned response.
The _TOOL_STUBS dict in code_execution_tool.py was out of sync with the
actual tool schemas, causing TypeErrors when the LLM used parameters it
sees in its system prompt but the sandbox stubs didn't accept:
search_files:
- Added missing params: context, offset, output_mode
- Fixed target default: 'grep' → 'content' (old value was obsolete)
patch:
- Added missing params: mode, patch (V4A multi-file patch support)
Also added 4 drift-detection tests (TestStubSchemaDrift) that will
catch future divergence between stubs and real schemas:
- test_stubs_cover_all_schema_params: every schema param in stub
- test_stubs_pass_all_params_to_rpc: every stub param sent over RPC
- test_search_files_target_uses_current_values: no obsolete values
- test_generated_module_accepts_all_params: generated code compiles
All 28 tests pass.
Authored by rovle. Adds Daytona as the sixth terminal execution backend
with cloud sandboxes, persistent workspaces, and full CLI/gateway integration.
Includes 24 unit tests and 8 integration tests.
The execute_code sandbox generates a hermes_tools.py stub module for LLM
scripts. Three common failure modes keep tripping up scripts:
1. json.loads(strict=True) rejects control chars in terminal() output
(e.g., GitHub issue bodies with literal tabs/newlines)
2. Shell backtick/quote interpretation when interpolating dynamic content
into terminal() commands (markdown with backticks gets eaten by bash)
3. No retry logic for transient network failures (API timeouts, rate limits)
Adds three convenience helpers to the generated hermes_tools module:
- json_parse(text) — json.loads with strict=False for tolerant parsing
- shell_quote(s) — shlex.quote() for safe shell interpolation
- retry(fn, max_attempts=3, delay=2) — exponential backoff wrapper
Also updates the EXECUTE_CODE_SCHEMA description to document these helpers
so LLMs know they're available without importing anything extra.
Includes 7 new tests (unit + integration) covering all three helpers.
The original implementation only supported xclip (X11), which silently
fails on WSL2 (can't access Windows clipboard for images), Wayland
desktops (xclip is X11-only), and VSCode terminal on WSL2.
Clipboard backend changes (hermes_cli/clipboard.py):
- WSL2: detect via /proc/version, use powershell.exe with .NET
System.Windows.Forms.Clipboard to extract images as base64 PNG
- Wayland: use wl-paste with MIME type detection, auto-convert BMP
to PNG for WSLg environments (via Pillow or ImageMagick)
- Dispatch order: WSL → Wayland → X11 (xclip), with fallthrough
- New has_clipboard_image() for lightweight clipboard checks
- Cache WSL detection result per-process
CLI changes (cli.py):
- /paste command: explicit clipboard image check for terminals where
BracketedPaste doesn't fire (image-only clipboard in VSCode/WinTerm)
- Ctrl+V keybinding: fallback for Linux terminals where Ctrl+V sends
raw byte instead of triggering bracketed paste
Tests: 80 tests (up from 37) covering WSL, Wayland, X11 dispatch,
BMP conversion, has_clipboard_image, and /paste command.
Copy an image to clipboard (screenshot, browser, etc.) and paste into
the Hermes CLI. The image is saved to ~/.hermes/images/, shown as a
badge above the input ([📎 Image #1]), and sent to the model as a
base64-encoded OpenAI vision multimodal content block.
Implementation:
- hermes_cli/clipboard.py: clean module with platform-specific extraction
- macOS: pngpaste (if installed) → osascript fallback (always available)
- Linux: xclip (apt install xclip)
- cli.py: BracketedPaste key handler checks clipboard on every paste,
image bar widget shows attached images, chat() converts to multimodal
content format, Ctrl+C clears attachments
Inspired by @m0at's fork (https://github.com/m0at/hermes-agent) which
implemented image paste support for local vision models. Reimplemented
cleanly as a separate module with tests.
On top of PR #460: self-hosted Firecrawl instances don't require an API
key (USE_DB_AUTHENTICATION=false), so don't force users to set a dummy
FIRECRAWL_API_KEY when FIRECRAWL_API_URL is set. Also adds a proper
self-hosting section to the configuration docs explaining what you get,
what you lose, and how to set it up (Docker stack, tradeoffs vs cloud).
Added 2 more tests (URL-only without key, neither-set raises).
Replaces the unsafe 128K fallback for unknown models with a descending
probe strategy (2M → 1M → 512K → 200K → 128K → 64K → 32K). When a
context-length error occurs, the agent steps down tiers and retries.
The discovered limit is cached per model+provider combo in
~/.hermes/context_length_cache.yaml so subsequent sessions skip probing.
Also parses API error messages to extract the actual context limit
(e.g. 'maximum context length is 32768 tokens') for instant resolution.
The CLI banner now displays the context window size next to the model
name (e.g. 'claude-opus-4 · 200K context · Nous Research').
Changes:
- agent/model_metadata.py: CONTEXT_PROBE_TIERS, persistent cache
(save/load/get), parse_context_limit_from_error(), get_next_probe_tier()
- agent/context_compressor.py: accepts base_url, passes to metadata
- run_agent.py: step-down logic in context error handler, caches on success
- cli.py + hermes_cli/banner.py: context length in welcome banner
- tests: 22 new tests for probing, parsing, and caching
Addresses #132. PR #319's approach (8K default) rejected — too conservative.
Adds optional FIRECRAWL_API_URL environment variable to support
self-hosted Firecrawl deployments alongside the cloud service.
- Add FIRECRAWL_API_URL to optional env vars in hermes_cli/config.py
- Update _get_firecrawl_client() in tools/web_tools.py to accept custom API URL
- Add tests for client initialization with/without URL
- Document new env var in installation and config guides
The Daytona SDK's process.exec(timeout=N) parameter is not enforced —
the server-side timeout never fires and the SDK has no client-side
fallback, causing commands to hang indefinitely.
Fix: wrap commands with timeout N sh -c '...' (coreutils) which
reliably kills the process and returns exit code 124. Added
shlex.quote for proper shell escaping and a secondary deadline (timeout + 10s) that force-stops the sandbox if the shell timeout somehow fails.
Signed-off-by: rovle <lovre.pesut@gmail.com>
state
- Replace logger.warning with warnings.warn for the disk cap so users
actually see it (logger was suppressed by CLI's log level config)
- Use SandboxState enum instead of string literals in
_ensure_sandbox_ready
Signed-off-by: rovle <lovre.pesut@gmail.com>
Authored by 0xbyt4. Wraps commands with unique fence markers to isolate real output
from shell init/exit noise (oh-my-zsh, macOS session restore, etc.). Falls back to
expanded pattern-based cleaning. Also fixes BSD find fallback and test module shadowing.
Adds a /update command to Telegram, Discord, and other gateway platforms
that runs `hermes update` to pull the latest code, update dependencies,
sync skills, and restart the gateway.
Implementation:
- Spawns `hermes update` in a separate systemd scope (systemd-run --user
--scope) so the process survives the gateway restart that hermes update
triggers at the end. Falls back to nohup if systemd-run is unavailable.
- Writes a marker file (.update_pending.json) with the originating
platform and chat_id before spawning the update.
- On gateway startup, _send_update_notification() checks for the marker,
reads the captured update output, sends the results back to the user,
and cleans up.
Also:
- Registers /update as a Discord slash command
- Updates README.md, docs/messaging.md, docs/slash-commands.md
- Adds 18 tests covering handler, notification, and edge cases
Authored by FarukEst. Fixes#392.
1. Initialize data={} before health-check loop to prevent NameError when
resp.json() raises after http_ready is set to True.
2. Extract _close_bridge_log() helper and call on all return False paths
to prevent file descriptor leaks on failed connection attempts.
Refactors disconnect() to reuse the same helper.
Authored by 0xbyt4.
The italic regex [^*]+ matched across newlines, corrupting bullet lists
using * markers (e.g. '* Item one\n* Item two' became italic garbage).
Fixed by adding \n to the negated character class: [^*\n]+.
Authored by 0xbyt4.
The dedup logic in GitHubSource.search() and unified_search() used
'r.trust_level == "trusted"' which let trusted results overwrite builtin
ones. Now uses ranked comparison: builtin (2) > trusted (1) > community (0).
Authored by 0xbyt4.
Two fixes:
- extract_images(): only remove extracted image tags, not all markdown image
tags. Previously  was silently dropped when real images
were also present.
- truncate_message(): walk chunk_body not full_chunk when tracking code block
state, so the reopened fence prefix doesn't toggle in_code off and leave
continuation chunks with unclosed code blocks.
Authored by 0xbyt4.
144 new tests covering gateway/pairing.py, tools/skill_manager_tool.py,
tools/skills_tool.py, honcho_integration/session.py, and
agent/auxiliary_client.py.
Authored by Farukest. Fixes#389.
Replaces hardcoded forward-slash string checks ('/.git/', '/.hub/') with
Path.parts membership test in _find_all_skills() and scan_skill_commands().
On Windows, str(Path) uses backslashes so the old filter never matched,
causing quarantined skills to appear as installed.
Authored by Farukest. Fixes#387.
Removes 'and not force' from the dangerous verdict check so --force
can never install skills with critical security findings (reverse shells,
data exfiltration, etc). The docstring already documented this behavior
but the code didn't enforce it.
Authored by Farukest. Fixes#385.
Replaces startswith() with Path.is_relative_to() in _check_structure()
symlink escape check — same fix pattern as skill_view() (PR #352).
Prevents symlinks escaping to sibling directories with shared name prefixes.
Fixes a bug where the refresh token was not persisted when the API key
mint failed (e.g., 402 insufficient credits, timeout). The rotated
refresh token was lost, causing subsequent auth attempts to fail with
a stale token.
Changes:
- Persist auth state immediately after each successful token refresh,
before attempting the mint
- Use latest in-memory refresh token on mint-retry paths (was using
the stale original)
- Atomic durable writes for auth.json (temp file + fsync + replace)
- Opt-in OAuth trace logging (HERMES_OAUTH_TRACE=1, fingerprint-only)
- 3 regression tests covering refresh+402, refresh+timeout, and
invalid-token retry behavior
Author: Robin Fernandes <rewbs>
Authored by PercyDikec. Fixes#394.
The transcript extraction used len(history) to find new messages, but
history includes session_meta entries stripped before reaching the agent.
This caused 1 message lost per turn from turn 2 onwards. Fix returns
history_offset (filtered length) from _run_agent and uses it for the slice.
Two fixes for the case where a user switches to a model with a smaller
context window while having a large existing session:
1. Preflight compression in run_conversation(): Before the main loop,
estimate tokens of loaded history + system prompt. If it exceeds the
model's compression threshold (85% of context), compress proactively
with up to 3 passes. This naturally handles model switches because
the gateway creates a fresh AIAgent per message with the current
model's context length.
2. Error handler reordering: Context-length errors (400 with 'maximum
context length' etc.) are now checked BEFORE the generic 4xx handler.
Previously, OpenRouter's 400-status context-length errors were caught
as non-retryable client errors and aborted immediately, never reaching
the compression+retry logic.
Reported by Sonicrida on Discord: 840-message session (2MB+) crashed
after switching from a large-context model to minimax via OpenRouter.
get_definitions() already wrapped check_fn() calls in try/except,
but is_toolset_available() did not. A failing check (network error,
missing import, bad config) would propagate uncaught and crash the
CLI banner, agent startup, and tools-info display.
Now is_toolset_available() catches all exceptions and returns False,
matching the existing pattern in get_definitions().
Added 4 tests covering exception handling in is_toolset_available(),
check_toolset_requirements(), get_definitions(), and
check_tool_availability().
Closes#402
The transcript extraction used len(history) to find new messages, but
history includes session_meta entries that are stripped before passing
to the agent. This mismatch caused 1 message to be lost from the
transcript on every turn after the first, because the slice offset
was too high. Use the filtered history length (history_offset) returned
by _run_agent instead.
Also changed the else branch from returning all agent_messages to
returning an empty list, so compressed/shorter agent output does not
duplicate the entire history into the transcript.
The hidden directory filter used hardcoded forward-slash strings like
'/.git/' and '/.hub/' to exclude internal directories. On Windows,
Path returns backslash-separated strings, so the filter never matched.
This caused quarantined skills in .hub/quarantine/ to appear as
installed skills and available slash commands on Windows.
Replaced string-based checks with Path.parts membership test which
works on both Windows and Unix.
The docstring states --force should never override dangerous verdicts,
but the condition `if result.verdict == "dangerous" and not force`
allowed force=True to skip the early return. Execution then fell
through to `if force: return True`, bypassing the policy block.
Removed `and not force` so dangerous skills are always blocked
regardless of the --force flag.
The symlink escape check in _check_structure() used startswith()
without a trailing separator. A symlink resolving to a sibling
directory with a shared prefix (e.g. 'axolotl-backdoor') would pass
the check for 'axolotl' since the string prefix matched.
Replaced with Path.is_relative_to() which correctly handles directory
boundaries and is consistent with the skill_view path check.
session_search was returning the current session if it matched the
query, which is redundant — the agent already has the current
conversation context. This wasted an LLM summarization call and a
result slot.
Added current_session_id parameter to session_search(). The agent
passes self.session_id and the search filters out any results where
either the raw or parent-resolved session ID matches. Both the raw
match and the parent-resolved match are checked to handle child
sessions from delegation.
Two tests added verifying the exclusion works and that other
sessions are still returned.
Replace the string-based startswith + os.sep approach with
Path.is_relative_to() (Python 3.9+, we require 3.10+). This is
the idiomatic pathlib way to check path containment — it handles
separators, case sensitivity, and the equal-path case natively
without string manipulation.
Simplified tests to match: removed the now-unnecessary
test_separator_is_os_native test since is_relative_to doesn't
depend on separator choice.
The session key construction logic was duplicated in 4 places
(session.py + 3 inline copies in run.py), which is exactly the
kind of drift that caused issue #349 in the first place.
Extracted build_session_key() as a public function in session.py.
SessionStore._generate_session_key() now delegates to it, and all
inline key construction in run.py has been replaced with calls to
the shared function. Tests updated to test the function directly.
The previous implementation used `len(self._entries) > 1` to check if any
sessions had ever been created. This failed for single-platform users because
when sessions reset (via /reset, auto-reset, or gateway restart), the entry
for the same session_key is replaced in _entries, not added. So len(_entries)
stays at 1 for users who only use one platform.
Fix: Query the SQLite database's session count instead. The database preserves
historical session records (marked as ended), so session_count() correctly
returns > 1 for returning users even after resets.
This prevents the agent from reintroducing itself to returning users after
every session reset.
Fixes#351
