- Regenerate uv.lock with sha256 hashes for all 2965 package artifacts
- Add python_version marker to yc-bench (requires >=3.12)
- Update setup-hermes.sh to prefer 'uv sync --locked' for hash-verified
installs, with fallback to 'uv pip install' when lockfile is stale
This completes the supply chain hardening: pyproject.toml bounds the
version ranges, and uv.lock pins exact versions with cryptographic
hashes so tampered packages are rejected at install time.
Replace all minisweagent imports with Hermes-Agent's own environment
classes (LocalEnvironment, DockerEnvironment, ModalEnvironment).
mini_swe_runner.py no longer has any dependency on mini-swe-agent.
The runner now uses the same backends as the terminal tool, so Docker
and Modal environments work out of the box without extra submodules.
Tested: local and Docker backends verified working through the runner.
Adds upper-bound version pins (<next_major) to all dependencies in
pyproject.toml — both core and optional. Previously most deps were
unpinned or had only floor bounds, meaning fresh installs would pull
whatever version was latest on PyPI.
This limits blast radius from supply chain attacks like the litellm
1.82.7/1.82.8 credential stealer (BerriAI/litellm#24512). With bounded
ranges, a compromised major version bump won't be pulled automatically.
Floors are set to current known-good installed versions.
* feat: env var passthrough for skills and user config
Skills that declare required_environment_variables now have those vars
passed through to sandboxed execution environments (execute_code and
terminal). Previously, execute_code stripped all vars containing KEY,
TOKEN, SECRET, etc. and the terminal blocklist removed Hermes
infrastructure vars — both blocked skill-declared env vars.
Two passthrough sources:
1. Skill-scoped (automatic): when a skill is loaded via skill_view and
declares required_environment_variables, vars that are present in
the environment are registered in a session-scoped passthrough set.
2. Config-based (manual): terminal.env_passthrough in config.yaml lets
users explicitly allowlist vars for non-skill use cases.
Changes:
- New module: tools/env_passthrough.py — shared passthrough registry
- hermes_cli/config.py: add terminal.env_passthrough to DEFAULT_CONFIG
- tools/skills_tool.py: register available skill env vars on load
- tools/code_execution_tool.py: check passthrough before filtering
- tools/environments/local.py: check passthrough in _sanitize_subprocess_env
and _make_run_env
- 19 new tests covering all layers
* docs: add environment variable passthrough documentation
Document the env var passthrough feature across four docs pages:
- security.md: new 'Environment Variable Passthrough' section with
full explanation, comparison table, and security considerations
- code-execution.md: update security section, add passthrough subsection,
fix comparison table
- creating-skills.md: add tip about automatic sandbox passthrough
- skills.md: add note about passthrough after secure setup docs
Live-tested: launched interactive CLI, loaded a skill with
required_environment_variables, verified TEST_SKILL_SECRET_KEY was
accessible inside execute_code sandbox (value: passthrough-test-value-42).
Complete cleanup after dropping the mini-swe-agent submodule (PR #2804):
- Remove MSWEA_SILENT_STARTUP and MSWEA_GLOBAL_CONFIG_DIR env var
settings from cli.py, run_agent.py, hermes_cli/main.py, doctor.py
- Remove mini-swe-agent health check from hermes doctor
- Remove 'minisweagent' from logger suppression lists
- Remove litellm/typer/platformdirs from requirements.txt
- Remove mini-swe-agent install steps from install.ps1 (Windows)
- Remove mini-swe-agent install steps from website docs
- Update all stale comments/docstrings referencing mini-swe-agent
in terminal_tool.py, tools/__init__.py, code_execution_tool.py,
environments/README.md, environments/agent_loop.py
- Remove mini_swe_runner from pyproject.toml py-modules
(still exists as standalone script for RL training use)
- Shrink test_minisweagent_path.py to empty stub
The orphaned mini-swe-agent/ directory on disk needs manual removal:
rm -rf mini-swe-agent/
When TERMINAL_CWD is set to '.' or any relative path (common when the
CLI config defaults to cwd='.'), container backends (docker, modal,
singularity, daytona) would pass it directly to the container where it's
meaningless. This caused 'docker run -d -w .' to fail.
Now relative paths are caught alongside host paths and replaced with
the default '/root' for container backends.
Drop the mini-swe-agent git submodule. All terminal backends now use
hermes-agent's own environment implementations directly.
Docker backend:
- Inline the `docker run -d` container startup (was 15 lines in
minisweagent's DockerEnvironment). Our wrapper already handled
execute(), cleanup(), security hardening, volumes, and resource limits.
Modal backend:
- Import swe-rex's ModalDeployment directly instead of going through
minisweagent's 90-line passthrough wrapper.
- Bake the _AsyncWorker pattern (from environments/patches.py) directly
into ModalEnvironment for Atropos compatibility without monkey-patching.
Cleanup:
- Remove minisweagent_path.py (submodule path resolution helper)
- Remove submodule init/install from install.sh and setup-hermes.sh
- Remove mini-swe-agent from .gitmodules
- environments/patches.py is now a no-op (kept for backward compat)
- terminal_tool.py no longer does sys.path hacking for minisweagent
- mini_swe_runner.py guards imports (optional, for RL training only)
- Update all affected tests to mock the new direct subprocess calls
- Update README.md, CONTRIBUTING.md
No functionality change — all Docker, Modal, local, SSH, Singularity,
and Daytona backends behave identically. 6093 tests pass.
browser_vision and other browser commands had a hardcoded 30-second
subprocess timeout that couldn't be overridden. Users with slower
machines (local Chromium without GPU) would hit timeouts on screenshot
capture even when setting browser.command_timeout in config.yaml,
because nothing read that value.
Changes:
- Add browser.command_timeout to DEFAULT_CONFIG (default: 30s)
- Add _get_command_timeout() helper that reads config, falls back to 30s
- _run_browser_command() now defaults to config value instead of constant
- browser_vision screenshot no longer hardcodes timeout=30
- browser_navigate uses max(config_timeout, 60) as floor for navigation
Reported by Gamer1988.
Documents the full /model command overhaul across 6 files:
AGENTS.md:
- Add model_switch.py to project structure tree
configuration.md:
- Rewrite General Setup with 3 config methods (interactive, config.yaml, env vars)
- Add new 'Switching Models with /model' section documenting all syntax variants
- Add 'Named Custom Providers' section with config.yaml examples and
custom:name:model triple syntax
slash-commands.md:
- Update /model descriptions in both CLI and messaging tables with
full syntax examples (provider:model, custom:model, custom:name:model,
bare custom auto-detect)
cli-commands.md:
- Add /model slash command subsection under hermes model with syntax table
- Add custom endpoint config to hermes model use cases
faq.md:
- Add config.yaml example for offline/local model setup
- Note that provider: custom is a first-class provider
- Document /model custom auto-detect
provider-runtime.md:
- Add model_switch.py to implementation file list
- Update provider families to show Custom as first-class with named variants
- Added calls to `_fire_first_delta()` in the `AIAgent` class to ensure that the first delta is triggered for both reasoning and thinking updates. This change improves the handling of delta events during streaming, enhancing the responsiveness of the agent's reasoning capabilities.
Phase 4 of the /model command overhaul.
Both the CLI (cli.py) and gateway (gateway/run.py) /model handlers
had ~50 lines of duplicated core logic: parsing, provider detection,
credential resolution, and model validation. This extracts that
pipeline into hermes_cli/model_switch.py.
New module exports:
- ModelSwitchResult: dataclass with all fields both handlers need
- CustomAutoResult: dataclass for bare '/model custom' results
- switch_model(): core pipeline — parse → detect → resolve → validate
- switch_to_custom_provider(): resolve endpoint + auto-detect model
The shared functions are pure (no I/O side effects). Each caller
handles its own platform-specific concerns:
- CLI: sets self.model/provider/etc, calls save_config_value(), prints
- Gateway: writes config.yaml directly, sets env vars, returns markdown
Net result: -244 lines from handlers, +234 lines in shared module.
The handlers are now ~80 lines each (down from ~150+) and can't drift
apart on core logic.
Fixes#2492.
`generate_systemd_unit()` and `get_python_path()` hardcoded `venv`
as the virtualenv directory name. When the virtualenv is `.venv`
(which `setup-hermes.sh` and `.gitignore` both reference), the
generated systemd unit had incorrect VIRTUAL_ENV and PATH variables.
Introduce `_detect_venv_dir()` which:
1. Checks `sys.prefix` vs `sys.base_prefix` to detect the active venv
2. Falls back to probing `.venv` then `venv` under PROJECT_ROOT
Both `get_python_path()` and `generate_systemd_unit()` now use
this detection instead of hardcoded paths.
Co-authored-by: Hermes <hermes@nousresearch.ai>
litellm 1.82.7/1.82.8 contained a credential stealer (.pth auto-exec
payload). PyPI quarantined the entire package, blocking all fresh
hermes-agent installs since litellm was listed as a hard dependency.
These three deps (litellm, typer, platformdirs) are only used by the
mini-swe-agent submodule, which has its own pyproject.toml and manages
its own dependencies. They were redundantly duplicated in hermes-agent's
pyproject.toml.
Also fixes install.sh to not print 'mini-swe-agent installed' on
failure, and updates warning messages in both install scripts to clarify
that only Docker/Modal backends are affected — local terminal is
unaffected.
Ref: https://github.com/BerriAI/litellm/issues/24512
* feat(model): persist base_url on /model switch, auto-detect for bare /model custom
Phase 2+3 of the /model command overhaul:
Phase 2 — Persist base_url on model switch:
- CLI: save model.base_url when switching to a non-OpenRouter endpoint;
clear it when switching away from custom to prevent stale URLs
leaking into the new provider's resolution
- Gateway: same logic using direct YAML write
Phase 3 — Better feedback and edge cases:
- Bare '/model custom' now auto-detects the model from the endpoint
using _auto_detect_local_model() and saves all three config values
(model, provider, base_url) atomically
- Shows endpoint URL in success messages when switching to/from
custom providers (both CLI and gateway)
- Clear error messages when no custom endpoint is configured
- Updated test assertions for the additional save_config_value call
Fixes#2562 (Phase 2+3)
* feat(model): support custom:name:model triple syntax for named custom providers
Phase 5 of the /model command overhaul.
Extends parse_model_input() to handle the triple syntax:
/model custom:local-server:qwen → provider='custom:local-server', model='qwen'
/model custom:my-model → provider='custom', model='my-model' (unchanged)
The 'custom:local-server' provider string is already supported by
_get_named_custom_provider() in runtime_provider.py, which matches
it against the custom_providers list in config.yaml. This just wires
the parsing so users can do it from the /model slash command.
Added 4 tests covering single, triple, whitespace, and empty model cases.
resolve_provider('custom') was silently returning 'openrouter', causing
users who set provider: custom in config.yaml to unknowingly route
through OpenRouter instead of their local/custom endpoint. The display
showed 'via openrouter' even when the user explicitly chose custom.
Changes:
- auth.py: Split the conditional so 'custom' returns 'custom' as-is
- runtime_provider.py: _resolve_named_custom_runtime now returns
provider='custom' instead of 'openrouter'
- runtime_provider.py: _resolve_openrouter_runtime returns
provider='custom' when that was explicitly requested
- Add 'no-key-required' placeholder for keyless local servers
- Update existing test + add 5 new tests covering the fix
Fixes#2562
- Updated `_on_tool_gen_start` method in `HermesCLI` to close open streaming boxes exactly once, preventing potential multiple closures.
- Added a check for `_stream_box_opened` to manage the state of the streaming box more effectively, enhancing user experience during large payload streaming.
- Introduced `_on_tool_gen_start` in `HermesCLI` to indicate when tool-call arguments are being generated, enhancing user feedback during streaming.
- Updated `AIAgent` to support a new `tool_gen_callback`, notifying the display layer when tool generation starts, allowing for better user experience during large payloads.
- Ensured that the callback is triggered appropriately during streaming events to prevent user interface freezing.
On macOS with Homebrew (Apple Silicon), Node.js and agent-browser
binaries live under /opt/homebrew/bin/ which is not included in the
_SANE_PATH fallback used by browser_tool.py and environments/local.py.
When Hermes runs with a filtered PATH (e.g. as a systemd service),
these binaries are invisible, causing 'env: node: No such file or
directory' errors when using browser tools.
Changes:
- Add /opt/homebrew/bin and /opt/homebrew/sbin to _SANE_PATH in both
browser_tool.py and environments/local.py
- Add _discover_homebrew_node_dirs() to find versioned Node installs
(e.g. brew install node@24) that aren't linked into /opt/homebrew/bin
- Extend _find_agent_browser() to search Homebrew and Hermes-managed
dirs when agent-browser isn't on the current PATH
- Include discovered Homebrew node dirs in subprocess PATH when
launching agent-browser
- Add 11 new tests covering all Homebrew path discovery logic
The gateway memory flush agent reviews old conversation history on session
reset/expiry and writes to memory. It had no awareness of memory changes
made after that conversation ended (by the live agent, cron jobs, or other
sessions), causing silent overwrites of newer entries.
Two fixes:
1. Skip memory flush entirely for cron sessions (session IDs starting with
'cron_'). Cron sessions are headless with no meaningful user conversation
to extract memories from.
2. Inject the current live memory state (MEMORY.md + USER.md) directly into
the flush prompt. The flush agent can now see what's already saved and
make informed decisions — only adding genuinely new information rather
than blindly overwriting entries that may have been updated since the
conversation ended.
Addresses the root cause identified in #2670: the flush agent was making
memory decisions blind to the current state of memory, causing stale
context to overwrite newer entries on gateway restarts and session resets.
Co-authored-by: devorun <devorun@users.noreply.github.com>
Co-authored-by: dlkakbs <dlkakbs@users.noreply.github.com>
* feat(config): support ${ENV_VAR} substitution in config.yaml
* fix: extend env var expansion to CLI and gateway config loaders
The original PR (#2680) only wired _expand_env_vars into load_config(),
which is used by 'hermes tools' and 'hermes setup'. The two primary
config paths were missed:
- load_cli_config() in cli.py (interactive CLI)
- Module-level _cfg in gateway/run.py (gateway — bridges api_keys to env vars)
Also:
- Remove redundant 'import re' (already imported at module level)
- Add missing blank lines between top-level functions (PEP 8)
- Add tests for load_cli_config() expansion
---------
Co-authored-by: teyrebaz33 <hakanerten02@hotmail.com>
echo was called with the full unquoted path (~username/suffix), allowing
command substitution in the suffix (e.g. ~user/$(malicious)) to execute
arbitrary shell commands. The fix expands only the validated ~username
portion via the shell and concatenates the suffix as a plain string.
Co-authored-by: Gutslabs <gutslabsxyz@gmail.com>
A bare `except Exception: pass` meant any YAML syntax error, bad value,
or unexpected structure in config.yaml was silently ignored and the
gateway fell back to .env / gateway.json without any indication.
Users had no way to know why their config changes had no effect.
Co-authored-by: sprmn24 <oncuevtv@gmail.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
httpx.AsyncClient awaits event hooks. The sync _ssrf_redirect_guard
returned None, causing 'object NoneType can't be used in await
expression' on any vision_analyze call that followed redirects.
Caught during live PTY testing of the merged SSRF protection.
* fix(security): add SSRF protection to vision_tools and web_tools
Both vision_analyze and web_extract/web_crawl accept arbitrary URLs
without checking if they target private/internal network addresses.
A prompt-injected or malicious skill could use this to access cloud
metadata endpoints (169.254.169.254), localhost services, or private
network hosts.
Adds a shared url_safety.is_safe_url() that resolves hostnames and
blocks private, loopback, link-local, and reserved IP ranges. Also
blocks known internal hostnames (metadata.google.internal).
Integrated at the URL validation layer in vision_tools and before
each website_policy check in web_tools (extract, crawl).
* test(vision): update localhost test to reflect SSRF protection
The existing test_valid_url_with_port asserted localhost URLs pass
validation. With SSRF protection, localhost is now correctly blocked.
Update the test to verify the block, and add a separate test for
valid URLs with ports using a public hostname.
* fix(security): harden SSRF protection — fail-closed, CGNAT, multicast, redirect guard
Follow-up hardening on top of dieutx's SSRF protection (PR #2630):
- Change fail-open to fail-closed: DNS errors and unexpected exceptions
now block the request instead of allowing it (OWASP best practice)
- Block CGNAT range (100.64.0.0/10): Python's ipaddress.is_private
does NOT cover this range (returns False for both is_private and
is_global). Used by Tailscale/WireGuard and carrier infrastructure.
- Add is_multicast and is_unspecified checks: multicast (224.0.0.0/4)
and unspecified (0.0.0.0) addresses were not caught by the original
four-check chain
- Add redirect guard for vision_tools: httpx event hook re-validates
each redirect target against SSRF checks, preventing the classic
redirect-based SSRF bypass (302 to internal IP)
- Move SSRF filtering before backend dispatch in web_extract: now
covers Parallel and Tavily backends, not just Firecrawl
- Extract _is_blocked_ip() helper for cleaner IP range checking
- Add 24 new tests (CGNAT, multicast, IPv4-mapped IPv6, fail-closed
behavior, parametrized blocked/allowed IP lists)
- Fix existing tests to mock DNS resolution for test hostnames
---------
Co-authored-by: dieutx <dangtc94@gmail.com>
Root cause: terminal_tool, execute_code, and process_registry returned raw
subprocess output with ANSI escape sequences intact. The model saw these
in tool results and copied them into file writes.
Previous fix (PR #2532) stripped ANSI at the write point in file_tools.py,
but this was a band-aid — regex on file content risks corrupting legitimate
content, and doesn't prevent ANSI from wasting tokens in the model context.
Source-level fix:
- New tools/ansi_strip.py with comprehensive ECMA-48 regex covering CSI
(incl. private-mode, colon-separated, intermediate bytes), OSC (both
terminators), DCS/SOS/PM/APC strings, Fp/Fe/Fs/nF escapes, 8-bit C1
- terminal_tool.py: strip output before returning to model
- code_execution_tool.py: strip stdout/stderr before returning
- process_registry.py: strip output in poll/read_log/wait
- file_tools.py: remove _strip_ansi band-aid (no longer needed)
Verified: `ls --color=always` output returned as clean text to model,
file written from that output contains zero ESC bytes.
* fix(gateway): added MATRIX_ALLOWED_USERS to list of env vars checked by gateway
* fix(gateway): add all missing platform allowlist env vars to startup check
The startup warning for 'No user allowlists configured' was only checking
TELEGRAM, DISCORD, WHATSAPP, SLACK, and SMS — missing SIGNAL, EMAIL,
MATTERMOST, and DINGTALK. Users of those platforms would see a spurious
warning even with their platform-specific allowlist configured.
Now matches the canonical platform_env_map in _is_user_authorized().
---------
Co-authored-by: SteelPh0enix <wojciech_olech@hotmail.com>
Cherry-picked from PR #2576 by ereid7, plus read-side fix from 173a5c62.
Both fixes were originally landed in 173a5c62 but were inadvertently
reverted by commit 34be3f8b (a squash-merge that bundled unrelated
tools_config.py changes).
Save side (_save_platform_tools): exclude platform default toolset
names (hermes-cli, hermes-telegram) from preserved entries so they
don't silently re-enable everything.
Read side (_get_platform_tools): when the saved list contains explicit
configurable keys, use direct membership instead of subset inference.
The subset approach is broken when composite toolsets like hermes-cli
resolve to ALL tools.
Cherry-picked from PR #2583 by Glucksberg.
The MEDIA: regex used \S+ which truncated paths at the first space.
Added a space-aware alternative anchored to known media extensions.
Also updated extract_local_files to allow spaces in path segments.
Follow-up fix: changed \s to [^\S\n] in the space-matching group
so the regex doesn't greedily match across newlines (broke multi-line
MEDIA: tags).
Cherry-picked from PR #2575 by ticketclosed-wontfix.
Filters out Discord system messages (thread renames, pins, member joins,
boosts) that were being treated as regular user messages.
Follow-up fix: also allow MessageType.reply (value 19) — the original
filter only allowed MessageType.default, which would silently drop all
reply-based interactions.
Added pytest.importorskip for discord dependency in tests.
An agent session killed the systemd-managed gateway (PID 1605) and restarted
it with '&disown', taking it outside systemd's Restart= management. When the
orphaned process later received SIGTERM, nothing restarted it.
Add dangerous command patterns to detect:
- 'gateway run' with & (background), disown, nohup, or setsid
- These should use 'systemctl --user restart hermes-gateway' instead
Also applied directly to main repo and fixed the systemd service:
- Changed Restart=on-failure to Restart=always (clean SIGTERM = exit 0 = not
a 'failure', so on-failure never triggered)
- RestartSec=10 for reasonable restart delay
Previously 'Activated skills: xxx' was printed above the banner in
show_banner(). Now it prints directly after the 'Welcome to Hermes
Agent!' line in run(), which is a more natural placement.
Path('~/.hermes/image.png').is_file() returns False because Path
doesn't expand tilde. This caused the tool to fall through to URL
validation, which also failed, producing a confusing error:
'Invalid image source. Provide an HTTP/HTTPS URL or a valid local
file path.'
Fix: use os.path.expanduser() before constructing the Path object.
Added two tests for tilde expansion (success and nonexistent file).
When a messaging platform fails to connect at startup (e.g. transient DNS
failure) or disconnects at runtime with a retryable error, the gateway now
queues it for background reconnection instead of giving up permanently.
- New _platform_reconnect_watcher background task runs alongside the
existing session expiry watcher
- Exponential backoff: 30s, 60s, 120s, 240s, 300s cap
- Max 20 retry attempts before giving up on a platform
- Non-retryable errors (bad auth token, etc.) are not retried
- Runtime disconnections via _handle_adapter_fatal_error now queue
retryable failures instead of triggering gateway shutdown
- On successful reconnect, adapter is wired up and channel directory
is rebuilt automatically
Fixes the case where a DNS blip during gateway startup caused Telegram
and Discord to be permanently unavailable until manual restart.
Local LLM servers (llama.cpp, ollama, vLLM, etc.) typically don't
require authentication. When a custom base_url is configured but no
API key is found, use a placeholder instead of failing with
'Provider resolver returned an empty API key.'
The OpenAI SDK accepts any string as api_key, and local servers
simply ignore the Authorization header.
Fixes issue reported by @ThatWolfieGuy — llama.cpp stopped working
after updating because the new runtime provider resolver enforces
non-empty API keys even for keyless local endpoints.
Added explicit closing of the OpenAI/httpx client in the background review process to prevent "Event loop is closed" errors. This change ensures that the client is properly cleaned up when the review agent is no longer needed, enhancing stability and resource management.
When AsyncOpenAI clients are garbage-collected after the event loop
closes, their AsyncHttpxClientWrapper.__del__ tries to schedule
aclose() on the dead loop, causing RuntimeError: Event loop is closed.
prompt_toolkit catches this as an unhandled exception and shows
'Press ENTER to continue...' which blocks CLI exit.
Fix: Add shutdown_cached_clients() to auxiliary_client.py that marks
all cached async clients' underlying httpx transport as CLOSED before
GC runs. This prevents __del__ from attempting the aclose() call.
- _force_close_async_httpx(): sets httpx AsyncClient._state to CLOSED
- shutdown_cached_clients(): iterates _client_cache, closes sync clients
normally and marks async clients as closed
- Also fix stale client eviction in _get_cached_client to mark evicted
async clients as closed (was just del-ing them, triggering __del__)
- Call shutdown_cached_clients() from _run_cleanup() in cli.py
The previous commit capped the 1.4x at 95% of context, but the multiplier
itself is unnecessary and confusing:
85% threshold × 1.4 = 119% of context → never fires
95% warn × 1.4 = 133% of context → never warns
The 85% hygiene threshold already provides ample headroom over the agent's
own 50% compressor. Even if rough estimates overestimate by 50%, hygiene
would fire at ~57% actual usage — safe and harmless.
Remove the multiplier entirely. Both actual and estimated token paths
now use the same 85% / 95% thresholds. Update tests and comments.
Three bugs in gateway session hygiene pre-compression caused 'Session too
large' errors for ~200K context models like GLM-5-turbo on z.ai:
1. Gateway hygiene called get_model_context_length(model) without passing
config_context_length, provider, or base_url — so user overrides like
model.context_length: 180000 were ignored, and provider-aware detection
(models.dev, z.ai endpoint) couldn't fire. The agent's own compressor
correctly passed all three (run_agent.py line 1038).
2. The 1.4x safety factor on rough token estimates pushed the compression
threshold above the model's actual context limit:
200K * 0.85 * 1.4 = 238K > 200K (model limit)
So hygiene never compressed, sessions grew past the limit, and the API
rejected the request.
3. Same issue for the warn threshold: 200K * 0.95 * 1.4 = 266K.
Fix:
- Read model.context_length, provider, and base_url from config.yaml
(same as run_agent.py does) and pass them to get_model_context_length()
- Resolve provider/base_url from runtime when not in config
- Cap the 1.4x-adjusted compress threshold at 95% of context_length
- Cap the 1.4x-adjusted warn threshold at context_length
Affects: z.ai GLM-5/GLM-5-turbo, any ~200K or smaller context model
where the 1.4x factor would push 85% above 100%.
Ref: Discord report from Ddox — glm-5-turbo on z.ai coding plan
* fix(mcp-oauth): port mismatch, path traversal, and shared state in OAuth flow
Three bugs in the new MCP OAuth 2.1 PKCE implementation:
1. CRITICAL: OAuth redirect port mismatch — build_oauth_auth() calls
_find_free_port() to register the redirect_uri, but _wait_for_callback()
calls _find_free_port() again getting a DIFFERENT port. Browser redirects
to port A, server listens on port B — callback never arrives, 120s timeout.
Fix: share the port via module-level _oauth_port variable.
2. MEDIUM: Path traversal via unsanitized server_name — HermesTokenStorage
uses server_name directly in filenames. A name like "../../.ssh/config"
writes token files outside ~/.hermes/mcp-tokens/.
Fix: sanitize server_name with the same regex pattern used elsewhere.
3. MEDIUM: Class-level auth_code/state on _CallbackHandler causes data
races if concurrent OAuth flows run. Second callback overwrites first.
Fix: factory function _make_callback_handler() returns a handler class
with a closure-scoped result dict, isolating each flow.
* test: add tests for MCP OAuth path traversal, handler isolation, and port sharing
7 new tests covering:
- Path traversal blocked (../../.ssh/config stays in mcp-tokens/)
- Dots/slashes sanitized and resolved within base dir
- Normal server names preserved
- Special characters sanitized (@, :, /)
- Concurrent handler result dicts are independent
- Handler writes to its own result dict, not class-level
- build_oauth_auth stores port in module-level _oauth_port
---------
Co-authored-by: 0xbyt4 <35742124+0xbyt4@users.noreply.github.com>
Models occasionally copy ANSI escape sequences from terminal output
or display formatting into file content, breaking shebangs and
injecting binary characters into scripts.
Strip ANSI codes (CSI, OSC, simple escapes) from:
- write_file content
- patch old_string, new_string, and V4A patch content
The check is fast (skips entirely if no ESC byte present).
Reported by Andi Jaeger.
