hermes-agent/tools
Teknium 97d54f0e4d
fix(terminal): three-layer defense against watch_patterns notification spam (#15642)
* fix(terminal): three-layer defense against watch_patterns notification spam

Background processes that stack notify_on_complete=True with watch_patterns
can flood the user with duplicate, delayed notifications — matches deliver
asynchronously via the completion queue and continue arriving minutes after
the process has exited. The docstring warning against this (PR #12113) has
proven insufficient; agents still misuse the combination.

Three layered defenses, each sufficient on its own:

1. Mutual exclusion (terminal_tool.py): When both flags are set on a
   background process, drop watch_patterns with a warning. notify_on_complete
   wins because 'let me know when it's done' is the more useful signal and
   fires exactly once. Extracted as _resolve_notification_flag_conflict() so
   the rule is testable in isolation.

2. Suppress-after-exit (process_registry.py): _check_watch_patterns() now
   bails the moment session.exited is True. Post-exit chunks (buffered reads
   draining after the process is gone) no longer produce notifications. This
   is the fix flagged as future work in session 20260418_020302_79881c.

3. Global circuit breaker (process_registry.py): Per-session rate limits don't
   catch the sibling-flood case — N concurrent processes can each stay under
   8/10s and still collectively spam. New WATCH_GLOBAL_MAX_PER_WINDOW=15 cap
   trips a 30-second cooldown across ALL sessions, emits a single
   watch_overflow_tripped event, silently counts dropped events, and emits a
   watch_overflow_released summary when the cooldown ends.

Also updates the tool schema + docstring to document the new behavior.

Tests: 8 new tests covering all three fixes (suppress-after-exit x2,
mutual-exclusion resolver x4, global breaker trip/cooldown/release x2).
All 60 tests across test_watch_patterns.py, test_notify_on_complete.py,
test_terminal_tool.py pass.

Real-world trigger: self-inflicted in session 20260425_051924 — three
concurrent hermes-sweeper review subprocesses each set watch_patterns=
['failed validation', 'errored'] AND notify_on_complete=True, then iterated
over multiple items, producing enough matches per process to defeat the
per-session cap while staying under the global cap that didn't yet exist.

* fix(terminal): aggressive 1-per-15s watch_patterns rate limit + strike-3 promotion

Per Teknium's direction, the watch_patterns rate limit is now much more
aggressive and self-healing.

## New rule — per session

- HARD cap: 1 watch-match notification per 15 seconds per process.
- Any match arriving inside the cooldown window is dropped and counts as
  ONE strike for that window (many drops in the same window still = 1 strike).
- After 3 consecutive strike windows, watch_patterns is permanently disabled
  for the session and the session is auto-promoted to notify_on_complete
  semantics — exactly one notification when the process actually exits.
- A cooldown window that expires with zero drops resets the consecutive
  strike counter — healthy cadence is forgiven.

## Schema + docstring rewritten

The tool schema description now gives the model explicit guidance:
- notify_on_complete is 'the right choice for almost every long-running task'
- watch_patterns is for RARE one-shot signals on LONG-LIVED processes
- Do NOT use watch_patterns with loops/batch jobs — error patterns fire every
  iteration and will hit the strike limit fast
- Mutual exclusion is stated on both parameter descriptions
- 1/15s cooldown and 3-strike promotion are stated in the watch_patterns
  description so the model sees the contract every turn

## Removed

- WATCH_MAX_PER_WINDOW (8/10s) and WATCH_OVERLOAD_KILL_SECONDS (45) — the
  new 1/15s limit subsumes both; keeping them would double-count.
- _watch_window_hits / _watch_window_start / _watch_overload_since fields
  on ProcessSession. Replaced by _watch_last_emit_at / _watch_cooldown_until
  / _watch_strike_candidate / _watch_consecutive_strikes.

## Kept

- Global circuit breaker across all sessions (15/10s → 30s cooldown) as a
  secondary safety net for concurrent siblings. Still valuable when 20
  short-lived processes each fire once — none individually violates the
  per-session limit.
- Suppress-after-exit guard.
- Mutual exclusion resolver at the tool entry point.

## Tests

- 6 new tests in TestPerSessionRateLimit covering: first match delivers,
  second in cooldown suppressed, multi-drop = single strike, 3 strikes
  disables + promotes, clean window resets counter, suppressed count
  carried to next emit.
- Global circuit breaker tests rewritten to use fresh sessions instead of
  hacking removed per-window fields.
- 50/50 watch_patterns + notify_on_complete tests pass.
- 60/60 including test_terminal_tool.py pass.
2026-04-25 06:41:58 -07:00
..
browser_providers feat: ungate Tool Gateway — subscription-based access with per-tool opt-in 2026-04-16 12:36:49 -07:00
environments fix(env): safely quote ~/ subpaths in wrapped cd commands 2026-04-24 15:25:12 -07:00
neutts_samples refactor(tts): replace NeuTTS optional skill with built-in provider + setup flow 2026-03-17 02:33:12 -07:00
__init__.py Merge branch 'main' into rewbs/tool-use-charge-to-subscription 2026-03-31 08:48:54 +09:00
ansi_strip.py fix: strip ANSI at the source — clean terminal output before it reaches the model 2026-03-23 07:43:12 -07:00
approval.py test: cover absolute paths in project env/config approval regex 2026-04-23 14:05:36 -07:00
binary_extensions.py fix(tools): address PR review — remove _extract_raw_output, BudgetConfig everywhere, read_file hardening 2026-04-08 02:24:32 -07:00
browser_camofox.py refactor: remove remaining redundant local imports (comprehensive sweep) 2026-04-21 00:50:58 -07:00
browser_camofox_state.py feat(browser): add persistent Camofox sessions and VNC URL discovery (salvage #4400) (#4419) 2026-04-01 04:18:50 -07:00
browser_cdp_tool.py fix: sanitize tool schemas for llama.cpp backends; restore MCP in TUI (#15032) 2026-04-24 02:44:46 -07:00
browser_dialog_tool.py feat(browser): CDP supervisor — dialog detection + response + cross-origin iframe eval (#14540) 2026-04-23 22:23:37 -07:00
browser_supervisor.py feat(browser): CDP supervisor — dialog detection + response + cross-origin iframe eval (#14540) 2026-04-23 22:23:37 -07:00
browser_tool.py feat(browser): CDP supervisor — dialog detection + response + cross-origin iframe eval (#14540) 2026-04-23 22:23:37 -07:00
budget_config.py fix: preserve existing thresholds, remove pre-read byte guard 2026-04-08 02:24:32 -07:00
checkpoint_manager.py refactor: remove redundant local imports already available at module level 2026-04-21 00:50:58 -07:00
clarify_tool.py refactor: add tool_error/tool_result helpers + read_raw_config, migrate 129 callsites 2026-04-07 13:36:38 -07:00
code_execution_tool.py fix(tools): restrict RPC socket permissions to owner-only 2026-04-22 17:27:18 -07:00
credential_files.py refactor: extract shared helpers to deduplicate repeated code patterns (#7917) 2026-04-11 13:59:52 -07:00
cronjob_tools.py fix(cron): wire context_from through the update action 2026-04-25 04:49:28 -07:00
debug_helpers.py refactor: codebase-wide lint cleanup — unused imports, dead code, and inefficient patterns (#5821) 2026-04-07 10:25:31 -07:00
delegate_tool.py fix(delegate): resolve subagent approval prompts without deadlocking parent TUI (#15491) 2026-04-24 22:37:22 -07:00
discord_tool.py feat(discord): split discord_server into discord + discord_admin tools 2026-04-25 04:50:14 -07:00
env_passthrough.py fix(env_passthrough): reject Hermes provider credentials from skill passthrough (#13523) 2026-04-21 06:14:25 -07:00
feishu_doc_tool.py fix(feishu-comment): use get_hermes_home(); drop dead asyncio wrapper; AUTHOR_MAP 2026-04-17 19:04:11 -07:00
feishu_drive_tool.py fix(feishu-comment): use get_hermes_home(); drop dead asyncio wrapper; AUTHOR_MAP 2026-04-17 19:04:11 -07:00
file_operations.py feat(skills): add design-md skill for Google's DESIGN.md spec (#14876) 2026-04-23 21:51:19 -07:00
file_state.py feat(delegate): cross-agent file state coordination for concurrent subagents (#13718) 2026-04-21 16:41:26 -07:00
file_tools.py fix(file_tools): resolve bookkeeping paths against live terminal cwd 2026-04-23 15:11:52 -07:00
fuzzy_match.py fix(patch): gate 'did you mean?' to no-match + extend to v4a/skill_manage 2026-04-21 02:03:46 -07:00
homeassistant_tool.py fix: clean up description escaping, add string-data tests 2026-04-13 04:45:07 -07:00
image_generation_tool.py fix(image-gen): force-refresh plugin providers in long-lived sessions 2026-04-23 03:01:18 -07:00
interrupt.py fix(interrupt): propagate to concurrent-tool workers + opt-in debug trace (#11907) 2026-04-17 20:39:25 -07:00
managed_tool_gateway.py fix(tools): add debug logging for token refresh and tighten domain check 2026-04-02 12:40:03 +11:00
mcp_oauth.py fix(mcp_oauth): raise RuntimeError instead of asserting OAuth port is set 2026-04-24 05:28:45 -07:00
mcp_oauth_manager.py fix(mcp-oauth): bidirectional auth_flow bridge + absolute expires_at (salvage #12025) (#12717) 2026-04-19 16:31:07 -07:00
mcp_tool.py fix(mcp): auto-reconnect + retry once when the transport session expires (#13383) 2026-04-24 05:28:45 -07:00
memory_tool.py fix: nest msvcrt import inside fcntl except block 2026-04-14 10:18:05 -07:00
mixture_of_agents_tool.py Fix (mixture_of_agents): replace deprecated Gemini model and forward max_tokens to OpenRouter (#6621) 2026-04-23 15:14:11 -07:00
neutts_synth.py fix(tts): document NeuTTS provider and align install guidance (#1903) 2026-03-18 02:55:30 -07:00
openrouter_client.py refactor: route ad-hoc LLM consumers through centralized provider router 2026-03-11 20:02:36 -07:00
osv_check.py feat: OSV malware check for MCP extension packages (#5305) 2026-04-05 12:46:07 -07:00
patch_parser.py fix(patch): gate 'did you mean?' to no-match + extend to v4a/skill_manage 2026-04-21 02:03:46 -07:00
path_security.py refactor: extract shared helpers to deduplicate repeated code patterns (#7917) 2026-04-11 13:59:52 -07:00
process_registry.py fix(terminal): three-layer defense against watch_patterns notification spam (#15642) 2026-04-25 06:41:58 -07:00
registry.py fix: tighten AST check to module-level only 2026-04-14 21:12:29 -07:00
rl_training_tool.py refactor: codebase-wide lint cleanup — unused imports, dead code, and inefficient patterns (#5821) 2026-04-07 10:25:31 -07:00
schema_sanitizer.py fix: sanitize tool schemas for llama.cpp backends; restore MCP in TUI (#15032) 2026-04-24 02:44:46 -07:00
send_message_tool.py refactor: remove remaining redundant local imports (comprehensive sweep) 2026-04-21 00:50:58 -07:00
session_search_tool.py fix(aux): add session_search extra_body and concurrency controls 2026-04-20 00:47:39 -07:00
skill_manager_tool.py feat(skills-guard): gate agent-created scanner on config.skills.guard_agent_created (default off) 2026-04-23 06:20:47 -07:00
skills_guard.py feat(skills-guard): gate agent-created scanner on config.skills.guard_agent_created (default off) 2026-04-23 06:20:47 -07:00
skills_hub.py feat(skills): add MiniMax-AI/cli as default skill tap 2026-04-23 02:35:13 -07:00
skills_sync.py feat(skills_sync): surface collision with reset-hint 2026-04-23 05:09:08 -07:00
skills_tool.py fix(skills): drop raw_content to avoid doubling skill payload 2026-04-24 15:15:07 -07:00
terminal_tool.py fix(terminal): three-layer defense against watch_patterns notification spam (#15642) 2026-04-25 06:41:58 -07:00
tirith_security.py fix: guard against None tirith path in security scanner 2026-04-23 03:08:53 -07:00
todo_tool.py fix(tools): enforce ID uniqueness in TODO store during replace operations 2026-04-11 16:22:50 -07:00
tool_backend_helpers.py fix(fal): extend whitespace-only FAL_KEY handling to all call sites 2026-04-21 02:04:21 -07:00
tool_output_limits.py feat(skills): add design-md skill for Google's DESIGN.md spec (#14876) 2026-04-23 21:51:19 -07:00
tool_result_storage.py fix(tools): neutralize shell injection in _write_to_sandbox via path quoting (#7940) 2026-04-11 14:26:11 -07:00
transcription_tools.py fix(transcription): fall back to CPU when CUDA runtime libs are missing 2026-04-24 02:50:14 -07:00
tts_tool.py fix(tts): use per-provider input-character caps instead of global 4000 (#13743) 2026-04-21 17:49:39 -07:00
url_safety.py feat(security): add global toggle to allow private/internal URL resolution 2026-04-22 14:38:59 -07:00
vision_tools.py fix: vision tool respects auxiliary.vision.temperature from config (#4661) 2026-04-20 00:32:09 -07:00
voice_mode.py fix: point optional-dep install hints at the venv's python (#11938) 2026-04-17 21:16:33 -07:00
web_tools.py feat(web): support TAVILY_BASE_URL env var for custom proxy endpoints 2026-04-22 17:36:33 -07:00
website_policy.py refactor: codebase-wide lint cleanup — unused imports, dead code, and inefficient patterns (#5821) 2026-04-07 10:25:31 -07:00
xai_http.py feat(xai): upgrade to Responses API, add TTS provider 2026-04-16 02:24:08 -07:00