Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Description Setting tools.exec.security: full and tools.exec.ask: off in config.yaml does NOT enable inline Python execution. Any attempt to use exec() or evaluate inline code fails with "Obfuscated command detected" or strictInlineEval policy violation.

Expected Behavior With security: full and ask: off, the agent should be able to execute inline Python code without approval prompts or security blocks.

Actual Behavior Inline Python execution is still blocked by an undocumented strictInlineEval policy.

Steps to reproduce

Steps to Reproduce Set config:

yaml tools: exec: security: full ask: off Restart gateway: openclaw gateway restart

Attempt inline Python: openclaw agent --agent main -m "run python: print('test')"

See error: "Obfuscated command detected" or strictInlineEval violation

Environment text OpenClaw: 2026.4.11 (769908e) OS: Ubuntu on WSL2 (6.6.87.2-microsoft-standard-WSL2) Install: npm global Attempted Fixes openclaw approvals allowlist add --agent "*" "**"

openclaw approvals allowlist add --agent "main" "**"

Adding "strictInlineEval": false to config (invalid key error)

Question for Developers Is there ANY configuration to disable strictInlineEval? If not, is this a bug or intentional design?

Expected behavior

Expected Behavior With security: full and ask: off, the agent should be able to execute inline Python code without approval prompts or security blocks.

Actual behavior

Actual Behavior Inline Python execution is still blocked by an undocumented strictInlineEval policy.

OpenClaw version

OpenClaw: 2026.4.11 (769908e)

Operating system

OS: Ubuntu on WSL2 (6.6.87.2-microsoft-standard-WSL2)

Install method

Install: npm global

Model

Gemini 2.5 Flash · google

Provider / routing chain

openclaw -> direct -> google/gemini-2.5-flash (default route, no gateway proxy)

Additional provider/model setup details

Default configuration with no custom routing. Model defined in config.yaml under agents.defaults.model.primary: "google/gemini-2.5-flash". No auth profiles or proxies configured.

Logs, screenshots, and evidence

Obfuscated command detected

Impact and severity

Error observed when attempting inline Python:

text Obfuscated command detected Config file after modifications shows:

yaml tools: exec: security: full ask: off Attempted config addition (failed - invalid key):

yaml tools: exec: security: full ask: off strictInlineEval: false # Results in "Unrecognized key" error Approvals configuration:

text Allowlist:

• local | * | ** | just now
• local | main | ** | just now

Additional information

NOT_ENOUGH_INFO for regression testing (no prior working version documented). Workaround: None identified. Manual copy-paste of processed data required.