Shell commands fail to execute properly when requested via Telegram

[theurgistk]

Summary

When sending shell commands via Telegram, ZeroClaw fails to execute or process them correctly.

Environment & Configuration

• OS / Platform: [Linux rpi 6.12.96+rpt-rpi-v8]
• ZeroClaw Version: [e.g., v0.8.5]
• Channel: Telegram

When asked to execute specific shell commands like df -lh or transmission-remote -n test:test -l, the AI agent does not return the actual command output. Instead, it hallucinates or returns incorrect/fake results (mocked data).

Click to expand config.toml

schema_version = 3

[gateway]
port = 3000
host = "0.0.0.0"
allow_public_bind = true

[agents.gk.a2a]
exposed_skills = []
published = false

[agents.gk.identity]
format = "openclaw"

[agents.gk.memory]
backend = "sqlite"

[agents.gk.precheck]
enabled = true
timeout_secs = 5

[agents.gk.workspace]
read_memory_from = []
unrestricted_filesystem = true

[peer_groups.telegram_default]
admin_for_agent_scope = true
agents = ["gk"]
channel = "telegram.default"

ignore = []
output_modality = "mirror"

[agents.gk.workspace.access]

[channels.telegram.default]
api_base_url = "[https://api.telegram.org](https://api.telegram.org)"
approval_timeout_secs = 120
draft_update_interval_ms = 1000
enabled = true
interrupt_on_new_message = false
mention_only = false
reply_min_interval_secs = 0
reply_queue_depth_max = 0
stream_mode = "off"
bot_token = "<YOUR_TELEGRAM_BOT_TOKEN>"
excluded_tools = ["browser_open", "screenshot", "cron_list", "cron_add", "cron_remove", "cron_update", "cron_run", "cron_runs", "memory_store"]

[web_search]
tavily_api_key = "<YOUR_TAVILY_API_KEY>"

[risk_profiles.balanced]
allowed_commands = ["*"]
block_high_risk_commands = false
forbidden_paths = []
level = "supervised"
require_approval_for_medium_risk = false
sandbox_enabled = false
workspace_only = false
allowed_roots = ["/"]
auto_approve = ["shell", "memory_recall", "web_search_tool", "web_fetch", "calculator", "glob_search", "content_search", "image_info", "weather", "tool_search", "browser", "browser_open", "file_write", "file_edit", "file_read"]
allowed_tools = []
always_ask = []
excluded_tools = []
firejail_args = []
shell_env_passthrough = []

[runtime_profiles.unbounded]
agentic = true
agentic_timeout_secs = 1800
compact_context = false
delegation_timeout_secs = 900
keep_tool_context_turns = 8
max_actions_per_hour = 4294967295
max_context_tokens = 128000
max_cost_per_day_cents = 4294967295
max_delegation_depth = 8
max_history_messages = 200
max_system_prompt_chars = 64000
max_tool_iterations = 100
max_tool_result_chars = 64000
memory_recall_limit = 10
parallel_tools = true
shell_timeout_secs = 600
strict_tool_parsing = false
tool_call_dedup_exempt = []
tool_filter_groups = []

[providers]

[risk_profiles.balanced.delegation_policy]
mode = "allow"

[runtime_profiles]

[runtime_profiles.unbounded.context_compression]
enabled = false
identifier_policy = "strict"
max_passes = 3
protect_first_n = 3
protect_last_n = 4
source_max_chars = 50000
summary_max_chars = 4000
threshold_ratio = 0.5
timeout_secs = 60
tool_result_retrim_chars = 2000
summary_provider = ""
tool_result_trim_exempt = []

[providers.models]

[runtime_profiles.unbounded.eval]
enabled = false
max_retries = 1
min_quality_score = 0.5

[providers.models.lmstudio]

[runtime_profiles.unbounded.history_pruning]
collapse_tool_results = true
enabled = false
keep_recent = 4
max_tokens = 8192

[providers.models.lmstudio.lm]

[providers.models.custom]

[agents]

[agents.gk]
risk_profile = "balanced"
model_provider = "custom.default"
runtime_profile = "unbounded"
acp_enable_mcp = false
channels = ["telegram.default"]
classifier_provider = ""
cron_jobs = []
delegate_same_risk_profile = true
delegates = []
enabled = true
knowledge_bundles = []
mcp_bundles = []
skill_bundles = []
summary_provider = ""
transcription_provider = ""
tts_provider = ""

[risk_profiles]

[runtime_profiles.unbounded.thinking]
default_level = "medium"
native_thinking = false
display = "off"

[runtime_profiles.unbounded.tool_receipts]
enabled = true
inject_system_prompt = true
show_in_response = true

[memory]
backend = "sqlite.sqlite"

[onboard_state]
quickstart_completed = true

[channels]

[channels.telegram]