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.
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 #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.
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>
When a session expires (daily schedule or idle timeout) and is
automatically reset, send a notification to the user explaining
what happened:
◐ Session automatically reset (inactive for 24h).
Conversation history cleared.
Use /resume to browse and restore a previous session.
Adjust reset timing in config.yaml under session_reset.
Notifications are suppressed when:
- The expired session had no activity (no tokens used)
- The platform is excluded (api_server, webhook by default)
- notify: false in config
Changes:
- session.py: _should_reset() returns reason string ('idle'/'daily')
instead of bool; SessionEntry gains auto_reset_reason and
reset_had_activity fields; old entry's total_tokens checked
- config.py: SessionResetPolicy gains notify (bool, default: true)
and notify_exclude_platforms (default: api_server, webhook)
- run.py: sends notification via adapter.send() before processing
the user's message, with activity + platform checks
- 13 new tests
Config (config.yaml):
session_reset:
notify: true
notify_exclude_platforms: [api_server, webhook]
- Download and cache .pdf, .docx, .xlsx, .pptx attachments locally
instead of passing expiring CDN URLs to the agent
- Inject .txt and .md content (≤100 KB) into event.text so the agent
sees file content without needing to fetch the URL
- Add 20 MB size guard and SUPPORTED_DOCUMENT_TYPES allowlist
- Fix: unsupported types (.zip etc.) no longer get MessageType.DOCUMENT
- Add 9 unit tests in test_discord_document_handling.py
Mirrors the Slack implementation from PR #784. Discord CDN URLs are
publicly accessible so no auth header is needed (unlike Slack).
Co-authored-by: Dilee <uzmpsk.dilekakbas@gmail.com>
- test_plugins.py: remove tests for unimplemented plugin command API
(get_plugin_command_handler, register_command never existed)
- test_redact.py: add autouse fixture to clear HERMES_REDACT_SECRETS
env var leaked by cli.py import in other tests
- test_signal.py: same HERMES_REDACT_SECRETS fix for phone redaction
- test_mattermost.py: add @bot_user_id to test messages after the
mention-only filter was added in #2443
- test_context_token_tracking.py: mock resolve_provider_client for
openai-codex provider that requires real OAuth credentials
Full suite: 5893 passed, 0 failed.
* fix: respect DashScope v1 runtime mode for alibaba
Remove the hardcoded Alibaba branch from resolve_runtime_provider()
that forced api_mode='anthropic_messages' regardless of the base URL.
Alibaba now goes through the generic API-key provider path, which
auto-detects the protocol from the URL:
- /apps/anthropic → anthropic_messages (via endswith check)
- /v1 → chat_completions (default)
This fixes Alibaba setup with OpenAI-compatible DashScope endpoints
(e.g. coding-intl.dashscope.aliyuncs.com/v1) that were broken because
runtime always forced Anthropic mode even when setup saved a /v1 URL.
Based on PR #2024 by @kshitijk4poor.
* docs(skill): add split, merge, search examples to ocr-and-documents skill
Adds pymupdf examples for PDF splitting, merging, and text search
to the existing ocr-and-documents skill. No new dependencies — pymupdf
already covers all three operations natively.
* fix: replace all production print() calls with logger in rl_training_tool
Replace all bare print() calls in production code paths with proper logger calls.
- Add `import logging` and module-level `logger = logging.getLogger(__name__)`
- Replace print() in _start_training_run() with logger.info()
- Replace print() in _stop_training_run() with logger.info()
- Replace print(Warning/Note) calls with logger.warning() and logger.info()
Using the logging framework allows log level filtering, proper formatting,
and log routing instead of always printing to stdout.
* fix(gateway): process /queue'd messages after agent completion
/queue stored messages in adapter._pending_messages but never consumed
them after normal (non-interrupted) completion. The consumption path
at line 5219 only checked pending messages when result.get('interrupted')
was True — since /queue deliberately doesn't interrupt, queued messages
were silently dropped.
Now checks adapter._pending_messages after both interrupted AND normal
completion. For queued messages (non-interrupt), the first response is
delivered before recursing to process the queued follow-up. Skips the
direct send when streaming already delivered the response.
Reported by GhostMode on Discord.
---------
Co-authored-by: kshitijk4poor <kshitijk4poor@users.noreply.github.com>
Co-authored-by: memosr.eth <96793918+memosr@users.noreply.github.com>
The /v1/responses endpoint used an in-memory OrderedDict that lost
all conversation state on gateway restart. Replace with SQLite-backed
storage at ~/.hermes/response_store.db.
- Responses and conversation name mappings survive restarts
- Same LRU eviction behavior (configurable max_size)
- WAL mode for concurrent read performance
- Falls back to in-memory SQLite if disk path unavailable
- Conversation name→response_id mapping moved into the store
Reverts the sanitizer addition from PR #2466 (originally #2129).
We already have _empty_content_retries handling for reasoning-only
responses. The trailing strip risks silently eating valid messages
and is redundant with existing empty-content handling.
Add hermes mcp add/remove/list/test/configure CLI for managing MCP
server connections interactively. Discovery-first 'add' flow connects,
discovers tools, and lets users select which to enable via curses checklist.
Add OAuth 2.1 PKCE authentication for MCP HTTP servers (RFC 7636).
Supports browser-based and manual (headless) authorization, token
caching with 0600 permissions, automatic refresh. Zero external deps.
Add ${ENV_VAR} interpolation in MCP server config values, resolved
from os.environ + ~/.hermes/.env at load time.
Core OAuth module from PR #2021 by @imnotdev25. CLI and mcp_tool
wiring rewritten against current main. Closes#497, #690.
Cherry-picked from PR #2017 by @simpolism. Fixes#2011.
Discord slash commands in threads were missing thread_id in the
SessionSource, causing them to route to the parent channel session.
Commands like /usage and /reset returned wrong data or affected the
wrong session.
Detects discord.Thread channels in _build_slash_event and sets
chat_type='thread' with thread_id. Two tests added.
Remove the hardcoded Alibaba branch from resolve_runtime_provider()
that forced api_mode='anthropic_messages' regardless of the base URL.
Alibaba now goes through the generic API-key provider path, which
auto-detects the protocol from the URL:
- /apps/anthropic → anthropic_messages (via endswith check)
- /v1 → chat_completions (default)
This fixes Alibaba setup with OpenAI-compatible DashScope endpoints
(e.g. coding-intl.dashscope.aliyuncs.com/v1) that were broken because
runtime always forced Anthropic mode even when setup saved a /v1 URL.
Based on PR #2024 by @kshitijk4poor.
Co-authored-by: kshitijk4poor <kshitijk4poor@users.noreply.github.com>
Parse thread_id from explicit deliver target (e.g. telegram:-1003724596514:17)
and forward it to _send_to_platform and mirror_to_session.
Previously _resolve_delivery_target() always set thread_id=None when
parsing the platform:chat_id format, breaking cron job delivery to
specific Telegram topics.
Added tests:
- test_explicit_telegram_topic_target_with_thread_id
- test_explicit_telegram_chat_id_without_thread_id
Also updated CRONJOB_SCHEMA deliver description to document the
platform:chat_id:thread_id format.
Co-authored-by: Alex Ferrari <alex@thealexferrari.com>
Five improvements to the /api/jobs endpoints:
1. Startup availability check — cron module imported once at class load,
endpoints return 501 if unavailable (not 500 per-request import error)
2. Input limits — name ≤ 200 chars, prompt ≤ 5000 chars, repeat must be
positive int
3. Update field whitelist — only name/schedule/prompt/deliver/skills/
repeat/enabled pass through to cron.jobs.update_job, preventing
arbitrary key injection
4. Deduplicated validation — _check_job_id and _check_jobs_available
helpers replace repeated boilerplate
5. 32 new tests covering all endpoints, validation, auth, and
cron-unavailable cases
Replace hardcoded 120-second grace period with a dynamic window that
scales with the job's scheduling frequency (half the period, clamped
to [120s, 2h]). Daily jobs now catch up if missed by up to 2 hours
instead of being silently skipped after just 2 minutes.
Changes the policy for agent-created skills with critical security
findings from 'block' (silently rejected) to 'ask' (allowed with
warning logged). The agent created the skill, so blocking it entirely
is too aggressive — let it through but log the findings.
- Policy: agent-created dangerous changed from block to ask
- should_allow_install returns None for 'ask' (vs True/False)
- format_scan_report shows 'NEEDS CONFIRMATION' for ask
- skill_manager_tool.py caller handles None (allows with warning)
- force=True still overrides as before
Based on PR #2271 by redhelix (closed — 3200 lines of unrelated
Mission Control code excluded).
Python 3.12 changed PosixPath.__new__ to ignore the redirected path
argument, breaking the FakePath subclass pattern. Use monkeypatch on
Path.exists instead.
Based on PR #2261 by @dieutx, fixed NameError (bare Path not imported).
Fixes#2234
The placeholder '(No response generated)' was overwriting the actual
final_response, causing it to be delivered to Discord even when the
agent completed work silently via tools.
Changes:
- Separate logged_response for output template display
- Keep final_response clean (empty when agent has no text)
- Delivery logic now correctly skips when final_response is empty
Test added to verify empty response stays empty for delivery.
Co-authored-by: Bartok9 <bartokmagic@proton.me>
Two bugs in the auxiliary provider auto-detection chain:
1. Expired Codex JWT blocks the auto chain: _read_codex_access_token()
returned any stored token without checking expiry, preventing fallback
to working providers. Now decodes JWT exp claim and returns None for
expired tokens.
2. Auxiliary Anthropic client missing OAuth identity transforms:
_AnthropicCompletionsAdapter always called build_anthropic_kwargs with
is_oauth=False, causing 400 errors for OAuth tokens. Now detects OAuth
tokens via _is_oauth_token() and propagates the flag through the
adapter chain.
Cherry-picked from PR #2378 by 0xbyt4. Fixed test_api_key_no_oauth_flag
to mock resolve_anthropic_token directly (env var alone was insufficient).
redact_sensitive_text() now returns early for None and coerces other
non-string values to str before applying regex-based redaction,
preventing TypeErrors in logging/tool-output paths.
Cherry-picked from PR #2369 by aydnOktay.
On the native Anthropic Messages API path, convert_messages_to_anthropic()
moves top-level cache_control on role:tool messages inside the tool_result
block. On OpenRouter (chat_completions), no such conversion happens — the
unexpected top-level field causes a silent hang on the second tool call.
Add native_anthropic parameter to _apply_cache_marker() and
apply_anthropic_cache_control(). When False (OpenRouter), role:tool messages
are skipped entirely. When True (native Anthropic), existing behaviour is
preserved.
Fixes#2362
When 'hermes update' stashes local changes and the restore hits
conflicts, the previous behavior silently ran 'git reset --hard HEAD'
to clean up. This could surprise users who didn't realize their
working tree was being nuked.
Now the conflict handler:
- Lists the specific conflicted files
- Reassures the user their stash is preserved
- Asks before resetting (interactive mode)
- Auto-resets in non-interactive mode (prompt_user=False)
- If declined, leaves the working tree as-is with guidance
* fix: prevent Anthropic token fallback leaking to third-party anthropic_messages providers
When provider is minimax/alibaba/etc and MINIMAX_API_KEY is not set,
the code fell back to resolve_anthropic_token() sending Anthropic OAuth
credentials to third-party endpoints, causing 401 errors.
Now only provider=="anthropic" triggers the fallback. Generalizes the
Alibaba-specific guard from #1739 to all non-Anthropic providers.
* fix: set provider='anthropic' in credential refresh tests
Follow-up for cherry-picked PR #2383 — existing tests didn't set
agent.provider, which the new guard requires to allow Anthropic
token refresh.
---------
Co-authored-by: 0xbyt4 <35742124+0xbyt4@users.noreply.github.com>
The gateway created a fresh AIAgent per message, rebuilding the system
prompt (including memory, skills, context files) every turn. This broke
prompt prefix caching — providers like Anthropic charge ~10x more for
uncached prefixes.
Now caches AIAgent instances per session_key with a config signature.
The cached agent is reused across messages in the same session,
preserving the frozen system prompt and tool schemas. Cache is
invalidated when:
- Config changes (model, provider, toolsets, reasoning, ephemeral
prompt) — detected via signature mismatch
- /new, /reset, /clear — explicit session reset
- /model — global model change clears all cached agents
- /reasoning — global reasoning change clears all cached agents
Per-message state (callbacks, stream consumers, progress queues) is
set on the agent instance before each run_conversation() call.
This matches CLI behavior where a single AIAgent lives across all turns
in a session, with _cached_system_prompt built once and reused.
When `hermes update` stashes local changes and the subsequent
`git stash apply` fails or leaves unmerged files, the conflict markers
(<<<<<<< etc.) were left in the working tree, making Hermes unrunnable
until manually cleaned up.
Now the update command runs `git reset --hard HEAD` to restore a clean
working tree before exiting, and also detects unmerged files even when
git stash apply reports success.
Closes#2348
Add @file:path, @folder:dir, @diff, @staged, @git:N, and @url:
references that expand inline before the message reaches the LLM.
Supports line ranges (@file:main.py:10-50), token budget enforcement
(soft warn at 25%, hard block at 50%), and path sandboxing for gateway.
Core module from PR #2090 by @kshitijk4poor. CLI and gateway wiring
rewritten against current main. Fixed asyncio.run() crash when called
from inside a running event loop (gateway).
Closes#682.
Fixes#1803. send_image_file, send_document, and send_video were missing
message_thread_id forwarding, causing them to fail in Telegram forum/supergroups
where thread_id is required. send_voice already handled this correctly. Adds
metadata parameter + message_thread_id to all three methods, and adds tests
covering the thread_id forwarding path.
Based on PR #1749 by @erosika (reimplemented on current main).
Extracts three protected methods from run() so wrapper CLIs can extend
the TUI without overriding the entire method:
- _get_extra_tui_widgets(): inject widgets between spacer and status bar
- _register_extra_tui_keybindings(kb, input_area): add keybindings
- _build_tui_layout_children(**widgets): full control over ordering
Default implementations reproduce existing layout exactly. The inline
HSplit in run() now delegates to _build_tui_layout_children().
5 tests covering defaults, widget insertion position, and keybinding
registration.
When using Alibaba (DashScope) with an anthropic-compatible endpoint,
model names like qwen3.5-plus were being normalized to qwen3-5-plus.
Alibaba's API expects the dot. Added preserve_dots parameter to
normalize_model_name() and build_anthropic_kwargs().
Also fixed 401 auth: when provider is alibaba or base_url contains
dashscope/aliyuncs, use only the resolved API key (DASHSCOPE_API_KEY).
Never fall back to resolve_anthropic_token(), and skip Anthropic
credential refresh for DashScope endpoints.
Cherry-picked from PR #1748 by crazywriter1. Fixes#1739.
- Add resolve_config_path(): checks $HERMES_HOME/honcho.json first,
falls back to ~/.honcho/config.json. Enables isolated Hermes instances
with independent Honcho credentials and settings.
- Update CLI and doctor to use resolved path instead of hardcoded global.
- Change default session_strategy from per-session to per-directory.
Part 1 of #1962 by @erosika.
Bare strings like "image", "audio", "document" were appended to
media_types, but downstream run.py checks mtype.startswith("image/")
and mtype.startswith("audio/"), which never matched. This caused all
Mattermost file attachments to be silently dropped from vision/STT
processing. Use the actual MIME type from file_info instead.
When streaming is enabled, the base adapter receives None from
_handle_message (already_sent=True) and cannot run auto-TTS for
voice input. The runner was unconditionally skipping voice input
TTS assuming the base adapter would handle it.
Now the runner takes over TTS responsibility when streaming has
already delivered the text response, so voice channel playback
works with both streaming on and off.
Streaming off behavior is unchanged (default already_sent=False
preserves the original code path exactly).
Co-authored-by: 0xbyt4 <35742124+0xbyt4@users.noreply.github.com>
Cron deliveries were mirrored into the target gateway session as
assistant-role messages, causing consecutive assistant messages that
violate message alternation (issue #2221).
Instead of fixing the role, remove the mirror injection entirely.
Cron outputs already live in their own cron session and don't belong
in the interactive conversation history.
Delivered messages are now wrapped with a header (task name) and a
footer noting the agent cannot see or respond to the message, so
users have clear context about what they're reading.
Closes#2221
A single Telegram 409 Conflict from getUpdates permanently killed
Telegram polling with no recovery possible (retryable=False on
first occurrence). This is too aggressive for production use with
process supervisors.
Transient 409s are expected during:
- --replace handoffs where the old long-poll session lingers on
Telegram servers for a few seconds after SIGTERM
- systemd Restart=on-failure respawns that overlap with the dying
instance cleanup
Now _handle_polling_conflict() retries up to 3 times with a
10-second delay between attempts. The 30-second total retry window
lets stale server-side sessions expire. If all retries fail, the
error is still marked as permanently fatal — preserving the original
protection against genuine dual-instance conflicts.
Tests updated: split the single conflict test into two — one verifying
retry on transient conflict, one verifying fatal after exhausted
retries.
Closes#2296
Previously, all project context files (AGENTS.md, .cursorrules, .hermes.md)
were loaded and concatenated into the system prompt. This bloated the prompt
with potentially redundant or conflicting instructions.
Now only ONE project context type is loaded, using priority order:
1. .hermes.md / HERMES.md (walk to git root)
2. AGENTS.md / agents.md (recursive directory walk)
3. CLAUDE.md / claude.md (cwd only, NEW)
4. .cursorrules / .cursor/rules/*.mdc (cwd only)
SOUL.md from HERMES_HOME remains independent and always loads.
Also adds CLAUDE.md as a recognized context file format, matching the
convention popularized by Claude Code.
Refactored the monolithic function into four focused helpers:
_load_hermes_md, _load_agents_md, _load_claude_md, _load_cursorrules.
Tests: replaced 1 coexistence test with 10 new tests covering priority
ordering, CLAUDE.md loading, case sensitivity, injection blocking.
Cherry-picked from PR #2201 by @Gutslabs.
session_search resolved hits to parent/root sessions but only excluded
the exact current_session_id. If the active session was a child
continuation (compression/delegation), its parent could still appear
as a 'past' conversation result.
Fix: resolve current_session_id to its lineage root before filtering,
so the entire active lineage (parent and children) is excluded.
Remove the [Files already read — do NOT re-read these] user message
that was injected into the conversation after context compression.
This message used role='user' for system-generated content, creating
a fake user turn that confused models about conversation state and
could contribute to task-redo behavior.
The file_tools.py read tracker (warn on 3rd consecutive read, block
on 4th+) already handles re-read prevention inline without injecting
synthetic messages.
Closes#2224.
Co-authored-by: Test <test@test.com>
Replace asyncio.run() with thread-local persistent event loops for
worker threads (e.g., delegate_task's ThreadPoolExecutor). asyncio.run()
creates and closes a fresh loop on every call, leaving cached
httpx/AsyncOpenAI clients bound to a dead loop — causing 'Event loop is
closed' errors during GC when parallel subagents clean up connections.
The fix mirrors the main thread's _get_tool_loop() pattern but uses
threading.local() so each worker thread gets its own long-lived loop,
avoiding both cross-thread contention and the create-destroy lifecycle.
Added 4 regression tests covering worker loop persistence, reuse,
per-thread isolation, and separation from the main thread's loop.
- Convert ~~text~~ to ~text~ (MarkdownV2 strikethrough)
- Protect ||text|| from pipe escaping (MarkdownV2 spoiler)
- Preserve > at line start as blockquote instead of escaping it
- Update _strip_mdv2() to strip ~strikethrough~ and ||spoiler|| markers
- Add tests covering new formatting paths and edge cases
Cherry-picked from PR #2146 by @crazywriter1. Fixes#2104.
asyncio.run() creates and closes a fresh event loop each call. Cached
httpx/AsyncOpenAI clients bound to the dead loop crash on GC with
'Event loop is closed'. This hit vision_analyze on first use in CLI.
Two-layer fix:
- model_tools._run_async(): replace asyncio.run() with persistent
loop via _get_tool_loop() + run_until_complete()
- auxiliary_client._get_cached_client(): track which loop created
each async client, discard stale entries if loop is closed
6 regression tests covering loop lifecycle, reuse, and full vision
dispatch chain.
Co-authored-by: Test <test@test.com>
Two fixes for Telegram/gateway-specific bugs:
1. Anthropic adapter: strip orphaned tool_result blocks (mirror of
existing tool_use stripping). Context compression or session
truncation can remove an assistant message containing a tool_use
while leaving the subsequent tool_result intact. Anthropic rejects
these with a 400: 'unexpected tool_use_id found in tool_result
blocks'. The adapter now collects all tool_use IDs and filters out
any tool_result blocks referencing IDs not in that set.
2. Gateway: /reset and /new now bypass the running-agent guard (like
/status already does). Previously, sending /reset while an agent
was running caused the raw text to be queued and later fed back as
a user message with the same broken history — replaying the
corrupted session instead of resetting it. Now the running agent is
interrupted, pending messages are cleared, and the reset command
dispatches immediately.
Tests updated: existing tests now include proper tool_use→tool_result
pairs; two new tests cover orphaned tool_result stripping.
Co-authored-by: Test <test@test.com>
* feat: context pressure warnings for CLI and gateway
User-facing notifications as context approaches the compaction threshold.
Warnings fire at 60% and 85% of the way to compaction — relative to
the configured compression threshold, not the raw context window.
CLI: Formatted line with a progress bar showing distance to compaction.
Cyan at 60% (approaching), bold yellow at 85% (imminent).
◐ context ▰▰▰▰▰▰▰▰▰▰▰▰▱▱▱▱▱▱▱▱ 60% to compaction 100k threshold (50%) · approaching compaction
⚠ context ▰▰▰▰▰▰▰▰▰▰▰▰▰▰▰▰▰▱▱▱ 85% to compaction 100k threshold (50%) · compaction imminent
Gateway: Plain-text notification sent to the user's chat via the new
status_callback mechanism (asyncio.run_coroutine_threadsafe bridge,
same pattern as step_callback).
Does NOT inject into the message stream. The LLM never sees these
warnings. Flags reset after each compaction cycle.
Files changed:
- agent/display.py — format_context_pressure(), format_context_pressure_gateway()
- run_agent.py — status_callback param, _context_50/70_warned flags,
_emit_context_pressure(), flag reset in _compress_context()
- gateway/run.py — _status_callback_sync bridge, wired to AIAgent
- tests/test_context_pressure.py — 23 tests
* Merge remote-tracking branch 'origin/main' into hermes/hermes-7ea545bf
---------
Co-authored-by: Test <test@test.com>
Replace the fragile hardcoded context length system with a multi-source
resolution chain that correctly identifies context windows per provider.
Key changes:
- New agent/models_dev.py: Fetches and caches the models.dev registry
(3800+ models across 100+ providers with per-provider context windows).
In-memory cache (1hr TTL) + disk cache for cold starts.
- Rewritten get_model_context_length() resolution chain:
0. Config override (model.context_length)
1. Custom providers per-model context_length
2. Persistent disk cache
3. Endpoint /models (local servers)
4. Anthropic /v1/models API (max_input_tokens, API-key only)
5. OpenRouter live API (existing, unchanged)
6. Nous suffix-match via OpenRouter (dot/dash normalization)
7. models.dev registry lookup (provider-aware)
8. Thin hardcoded defaults (broad family patterns)
9. 128K fallback (was 2M)
- Provider-aware context: same model now correctly resolves to different
context windows per provider (e.g. claude-opus-4.6: 1M on Anthropic,
128K on GitHub Copilot). Provider name flows through ContextCompressor.
- DEFAULT_CONTEXT_LENGTHS shrunk from 80+ entries to ~16 broad patterns.
models.dev replaces the per-model hardcoding.
- CONTEXT_PROBE_TIERS changed from [2M, 1M, 512K, 200K, 128K, 64K, 32K]
to [128K, 64K, 32K, 16K, 8K]. Unknown models no longer start at 2M.
- hermes model: prompts for context_length when configuring custom
endpoints. Supports shorthand (32k, 128K). Saved to custom_providers
per-model config.
- custom_providers schema extended with optional models dict for
per-model context_length (backward compatible).
- Nous Portal: suffix-matches bare IDs (claude-opus-4-6) against
OpenRouter's prefixed IDs (anthropic/claude-opus-4.6) with dot/dash
normalization. Handles all 15 current Nous models.
- Anthropic direct: queries /v1/models for max_input_tokens. Only works
with regular API keys (sk-ant-api*), not OAuth tokens. Falls through
to models.dev for OAuth users.
Tests: 5574 passed (18 new tests for models_dev + updated probe tiers)
Docs: Updated configuration.md context length section, AGENTS.md
Co-authored-by: Test <test@test.com>
* fix: preserve Ollama model:tag colons in context length detection
The colon-split logic in get_model_context_length() and
_query_local_context_length() assumed any colon meant provider:model
format (e.g. "local:my-model"). But Ollama uses model:tag format
(e.g. "qwen3.5:27b"), so the split turned "qwen3.5:27b" into just
"27b" — which matches nothing, causing a fallback to the 2M token
probe tier.
Now only recognised provider prefixes (local, openrouter, anthropic,
etc.) are stripped. Ollama model:tag names pass through intact.
* fix: update claude-opus-4-6 and claude-sonnet-4-6 context length from 200K to 1M
Both models support 1,000,000 token context windows. The hardcoded defaults
were set before Anthropic expanded the context for the 4.6 generation.
Verified via models.dev and OpenRouter API data.
---------
Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
Co-authored-by: Test <test@test.com>
Cherry-picked from PR #2120 by @unclebumpy.
- from_env() now reads HONCHO_BASE_URL and enables Honcho when base_url
is set, even without an API key
- from_global_config() reads baseUrl from config root with
HONCHO_BASE_URL env var as fallback
- get_honcho_client() guard relaxed to allow base_url without api_key
for no-auth local instances
- Added HONCHO_BASE_URL to OPTIONAL_ENV_VARS registry
Result: Setting HONCHO_BASE_URL=http://localhost:8000 in ~/.hermes/.env
now correctly routes the Honcho client to a local instance.
When the user is on a custom provider (provider=custom, localhost, or
127.0.0.1 endpoint), /model <name> no longer tries to auto-detect a
provider switch. The model name changes on the current endpoint as-is.
To switch away from a custom endpoint, users must use explicit
provider:model syntax (e.g. /model openai-codex:gpt-5.2-codex).
A helpful tip is printed when changing models on a custom endpoint.
This prevents the confusing case where someone on LM Studio types
/model gpt-5.2-codex, the auto-detection tries to switch providers,
fails or partially succeeds, and requests still go to the old endpoint.
Also fixes the missing prompt_toolkit.auto_suggest mock stub in
test_cli_init.py (same issue already fixed in test_cli_new_session.py).
Follow-up to PR #2101 (InB4DevOps). Adds three missing context compressor
resets in reset_session_state():
- compression_count (displayed in status bar)
- last_total_tokens
- _context_probed (stale context-error flag)
Also fixes the test_cli_new_session.py prompt_toolkit mock (missing
auto_suggest stub) and adds a regression test for #2099 that verifies
all token counters and compressor state are zeroed on /new.
The colon-split logic in get_model_context_length() and
_query_local_context_length() assumed any colon meant provider:model
format (e.g. "local:my-model"). But Ollama uses model:tag format
(e.g. "qwen3.5:27b"), so the split turned "qwen3.5:27b" into just
"27b" — which matches nothing, causing a fallback to the 2M token
probe tier.
Now only recognised provider prefixes (local, openrouter, anthropic,
etc.) are stripped. Ollama model:tag names pass through intact.
Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
Place a sentinel in _running_agents immediately after the "already
running" guard check passes — before any await. Without this, the
numerous await points between the guard (line 1324) and agent
registration (track_agent at line 4790) create a window where a
second message for the same session can bypass the guard and start
a duplicate agent, corrupting the transcript.
The await gap includes: hook emissions, vision enrichment (external
API call), audio transcription (external API call), session hygiene
compression, and the run_in_executor call itself. For messages with
media attachments the window can be several seconds wide.
The sentinel is wrapped in try/finally so it is always cleaned up —
even if the handler raises or takes an early-return path. When the
real AIAgent is created, track_agent() overwrites the sentinel with
the actual instance (preserving interrupt support).
Also handles the edge case where a message arrives while the sentinel
is set but no real agent exists yet: the message is queued via the
adapter's pending-message mechanism instead of attempting to call
interrupt() on the sentinel object.
MiniMax's default base URL was /v1 which caused runtime_provider to
default to chat_completions mode (OpenAI-style Authorization: Bearer
header). MiniMax rejects this with a 401 because they require the
Anthropic-style x-api-key header.
Changes:
- auth.py: Change default inference_base_url for minimax and minimax-cn
from /v1 to /anthropic
- runtime_provider.py: Auto-correct stale /v1 URLs from existing .env
files to /anthropic, and always default minimax/minimax-cn providers
to anthropic_messages mode
- Update tests to reflect new defaults, add tests for stale URL
auto-correction and explicit api_mode override
Based on PR #2100 by @devorun. Fixes#2094.
Co-authored-by: Test <test@test.com>
When LM Studio has a model loaded with a custom context size (e.g.,
122K), prefer that over the model's max_context_length (e.g., 1M).
This makes the TUI status bar show the actual runtime context window.
Instead of defaulting to 2M for unknown local models, query the server
API for the real context length. Supports Ollama (/api/show), vLLM
(max_model_len), and LM Studio (/v1/models). Results are cached to
avoid repeated queries.
* fix(codex): treat reasoning-only responses as incomplete, not stop
When a Codex Responses API response contains only reasoning items
(encrypted thinking state) with no message text or tool calls, the
_normalize_codex_response method was setting finish_reason='stop'.
This sent the response into the empty-content retry loop, which
burned 3 retries and then failed — exactly the pattern Nester
reported in Discord.
Two fixes:
1. _normalize_codex_response: reasoning-only responses (reasoning_items_raw
non-empty but no final_text) now get finish_reason='incomplete', routing
them to the Codex continuation path instead of the retry loop.
2. Incomplete handling: also checks for codex_reasoning_items when deciding
whether to preserve an interim message, so encrypted reasoning state is
not silently dropped when there is no visible reasoning text.
Adds 4 regression tests covering:
- Unit: reasoning-only → incomplete, reasoning+content → stop
- E2E: reasoning-only → continuation → final answer succeeds
- E2E: encrypted reasoning items preserved in interim messages
* fix(codex): ensure reasoning items have required following item in API input
Follow-up to the reasoning-only response fix. Three additional issues
found by tracing the full replay path:
1. _chat_messages_to_responses_input: when a reasoning-only interim
message was converted to Responses API input, the reasoning items
were emitted as the last items with no following item. The Responses
API requires a following item after each reasoning item (otherwise:
'missing_following_item' error, as seen in OpenHands #11406). Now
emits an empty assistant message as the required following item when
content is empty but reasoning items were added.
2. Duplicate detection: two consecutive reasoning-only incomplete
messages with identical empty content/reasoning but different
encrypted codex_reasoning_items were incorrectly treated as
duplicates, silently dropping the second response's reasoning state.
Now includes codex_reasoning_items in the duplicate comparison.
3. Added tests for both the API input conversion path and the duplicate
detection edge case.
Research context: verified against OpenCode (uses Vercel AI SDK, no
retry loop so avoids the issue), Clawdbot (drops orphaned reasoning
blocks entirely), and OpenHands (hit the missing_following_item error).
Our approach preserves reasoning continuity while satisfying the API
constraint.
---------
Co-authored-by: Test <test@test.com>
* fix: persist ACP sessions to disk so they survive process restarts
The ACP adapter stored sessions entirely in-memory. When the editor
restarted the ACP subprocess (idle timeout, crash, system sleep/wake,
editor restart), all sessions were lost. The editor's load_session /
resume_session calls would fail to find the session, forcing a new
empty session and losing all conversation history.
Changes:
- SessionManager now persists each session as a JSON file under
~/.hermes/acp_sessions/<session_id>.json
- get_session() transparently restores from disk when not in memory
- update_cwd(), fork_session(), list_sessions() all check disk
- server.py calls save_session() after prompt completion, /reset,
/compact, and model switches
- cleanup() and remove_session() delete disk files too
- Sessions have a 7-day TTL; expired sessions are pruned on startup
- Atomic writes via tempfile + os.replace to prevent corruption
- 11 new tests covering persistence, disk restoration, and TTL expiry
* refactor: use SessionDB instead of JSON files for ACP session persistence
Replace the standalone JSON file persistence layer with SessionDB
(~/.hermes/state.db) integration. ACP sessions now:
- Share the same DB as CLI and gateway sessions
- Are searchable via session_search (FTS5)
- Get token tracking, cost tracking, and session titles for free
- Follow existing session pruning policies
Key changes:
- _get_db() lazily creates a SessionDB, resolving HERMES_HOME
dynamically (not at import time) for test compatibility
- _persist() creates session record + replaces messages in DB
- _restore() loads from DB with source='acp' filter
- cwd stored in model_config JSON field (no schema migration)
- Model values coerced to str to handle mock agents in tests
- Removed: json files, sessions_dir, ttl_days, _expire logic
- Tests updated: DB-backed persistence, FTS search, tool_call
round-tripping, source filtering
---------
Co-authored-by: Test <test@test.com>
Authored by Lovre Pešut (rovle). Migrates from deprecated find_one(labels=...)
to get(sandbox_name) with deterministic naming (hermes-{task_id}), plus legacy
fallback via list(labels=...) for pre-migration sandboxes.
When a cron job references a skill that is no longer installed,
_build_job_prompt() now logs a warning and injects a user-visible notice
into the prompt instead of raising RuntimeError. The job continues with
any remaining valid skills and the user prompt.
Adds 4 regression tests for missing skill handling.
find_one is being deprecated. Primary lookup now uses get() with a
deterministic sandbox name (hermes-{task_id}). A legacy fallback via
list(labels=...) ensures sandboxes created before this migration are
still resumable.
Authored by dusterbloom. Closes#1911.
Pre-computes SQL query strings at class definition time in insights.py,
adds identifier quoting for ALTER TABLE DDL in hermes_state.py, and adds
4 regression tests verifying query construction safety.
The merge at e7844e9c re-introduced a line in _build_child_agent() that
references _saved_tool_names — a variable only defined in _run_single_child().
This caused NameError on every delegate_task call, completely breaking
subagent delegation.
Moves the child._delegate_saved_tool_names assignment to _run_single_child()
where _saved_tool_names is actually defined, keeping the save/restore in the
same scope as the try/finally block.
Adds two regression tests from PR #2038 (YanSte).
Also fixes the same issue reported in PR #2048 (Gutslabs).
Co-authored-by: Yannick Stephan <yannick.stephan@gmail.com>
Co-authored-by: Guts <gutslabs@users.noreply.github.com>
