bug: terminal() tool hard-caps foreground timeout at 600s, needs 1200+ #65

New issue

Closed

opened 2026-05-13 20:06:03 +00:00 by Mark · 1 comment

Mark commented

2026-05-13 20:06:03 +00:00

Owner

Description

terminal_tool.py line 105-108 hardcodes FOREGROUND_MAX_TIMEOUT = 600 with an env var override (TERMINAL_MAX_FOREGROUND_TIMEOUT). The tool rejects any foreground timeout= parameter above 600s.

FOREGROUND_MAX_TIMEOUT = _safe_parse_import_env(
    "TERMINAL_MAX_FOREGROUND_TIMEOUT",
    600,   # hardcoded default
    int,
)

GSD pipelines, provider calls, and long-running analysis commands routinely exceed 600s. The 600s cap forces either:

Background mode with polling (extra complexity)
Splitting work into artificial chunks
Running commands outside the hermes tool framework

Impact

GSD runs with multiple slices routinely hit this wall
Pipelines forced into background+notify pattern when a simple foreground call with generous timeout would suffice
No config.yaml setting exists to adjust this — only an env var, which is invisible and fragile across deployments

Request

Raise the default from 600 to 1200 (600 is clearly too tight for any multi-step pipeline)
Better: make it configurable via config.yaml under the terminal: section instead of a source-code env var

Location

~/.hermes/hermes-agent/tools/terminal_tool.py lines 105-111

## Description `terminal_tool.py` line 105-108 hardcodes `FOREGROUND_MAX_TIMEOUT = 600` with an env var override (`TERMINAL_MAX_FOREGROUND_TIMEOUT`). The tool rejects any foreground `timeout=` parameter above 600s. ```python FOREGROUND_MAX_TIMEOUT = _safe_parse_import_env( "TERMINAL_MAX_FOREGROUND_TIMEOUT", 600, # hardcoded default int, ) ``` GSD pipelines, provider calls, and long-running analysis commands routinely exceed 600s. The 600s cap forces either: - Background mode with polling (extra complexity) - Splitting work into artificial chunks - Running commands outside the hermes tool framework ## Impact - GSD runs with multiple slices routinely hit this wall - Pipelines forced into background+notify pattern when a simple foreground call with generous timeout would suffice - No config.yaml setting exists to adjust this — only an env var, which is invisible and fragile across deployments ## Request 1. **Raise the default** from 600 to **1200** (600 is clearly too tight for any multi-step pipeline) 2. **Better: make it configurable via config.yaml** under the `terminal:` section instead of a source-code env var ## Location `~/.hermes/hermes-agent/tools/terminal_tool.py` lines 105-111

Mark added the

bug

label

2026-05-13 20:06:03 +00:00

Mark commented

2026-05-13 20:11:34 +00:00

Author

Owner

Fix Applied

Added TERMINAL_MAX_FOREGROUND_TIMEOUT=1200 to ~/.hermes/.env.

Hermes loads .env via load_hermes_dotenv() before tool modules are imported, so this env var is picked up at import time by _safe_parse_import_env() in terminal_tool.py.

Test

Before fix (current session, still at 600 cap):

terminal(timeout=1200) -> Error: "Foreground timeout 1200s exceeds the maximum of 600s."

After next process restart, the env var will be read and the cap will be 1200.

Resolution

Immediate fix: env var in .env — survives agent restarts, no code change needed
Long-term fix (still worth doing): make this configurable via config.yaml under the terminal: section so it's discoverable, not an invisible env var. Accepting close on the env var workaround; reopen if you want the config.yaml route.

## Fix Applied Added `TERMINAL_MAX_FOREGROUND_TIMEOUT=1200` to `~/.hermes/.env`. Hermes loads `.env` via `load_hermes_dotenv()` before tool modules are imported, so this env var is picked up at import time by `_safe_parse_import_env()` in `terminal_tool.py`. ## Test Before fix (current session, still at 600 cap): ``` terminal(timeout=1200) -> Error: "Foreground timeout 1200s exceeds the maximum of 600s." ``` After next process restart, the env var will be read and the cap will be 1200. ## Resolution - **Immediate fix**: env var in `.env` — survives agent restarts, no code change needed - **Long-term fix** (still worth doing): make this configurable via `config.yaml` under the `terminal:` section so it's discoverable, not an invisible env var. Accepting close on the env var workaround; reopen if you want the config.yaml route.

Mark closed this issue

2026-05-13 20:11:38 +00:00