Summary

I clarified code-execution artifact guidance so failed executions are not confused with broken /mnt/data persistence, and tightened bash PTC output handling based on live testing.

• Added a shared failed-execution reminder when code references /mnt/data, so direct code, bash, and PTC errors tell the model that files from the failed call were not registered.
• Updated artifact path guidance to say failed executions do not register new files while preserving the /tmp scratch guidance.
• Tightened bash PTC jq guidance for JSON-string fields and arrays that may contain strings.
• Updated bash PTC examples to capture tool output with command substitution and printf before saving files, avoiding direct tool-call redirection that can create empty files.
• Added coverage for bash PTC error output and description guidance.

Change Type

• Bug fix (non-breaking change which fixes an issue)
• Documentation update

Testing

• npx jest src/tools/__tests__/ProgrammaticToolCalling.test.ts src/tools/__tests__/BashExecutor.test.ts src/tools/__tests__/CodeApiAuthHeaders.test.ts --runInBand
• npx eslint src/tools/CodeExecutor.ts src/tools/BashExecutor.ts src/tools/ProgrammaticToolCalling.ts src/tools/BashProgrammaticToolCalling.ts src/tools/__tests__/ProgrammaticToolCalling.test.ts src/tools/__tests__/BashExecutor.test.ts src/tools/__tests__/CodeApiAuthHeaders.test.ts
• npx tsc --noEmit
• Live direct CodeAPI smoke against http://localhost:3112/v1/exec/programmatic verified failed /mnt/data writes include the registration reminder.
• Live Claude Sonnet 4.6 agent run verified bash PTC used command substitution for tool output, handled double-encoded JSON with jq, and did not treat failed-call files as persistent.

Test Configuration:

• Local agents worktree on danny-avila/code-error-persistence-guidance
• Existing repository dependencies via node_modules
• Local CodeAPI at http://localhost:3112/v1
• Provider credentials loaded from the main worktree .env

Checklist

• My code adheres to this project's style guidelines
• I have performed a self-review of my own code
• I have made pertinent documentation changes
• My changes do not introduce new warnings
• I have written tests demonstrating that my changes are effective or that my feature works
• Local unit tests pass with my changes