risks.md

cli — Risk Vectors & Known Issues

CLI-R-001: Task passthrough from CLI to TUI

File: cli/_handlers/_chat.py Risk: The initial task string from teaagent chat "task" must be passed through to the TUI REPL. This was the subject of TASK-DD2-001 (recently fixed, commit 47710d9). Regression risk if the TUI initialization path changes. Failure mode: Initial task silently ignored; user must retype.

CLI-R-002: Handlers call sys.exit() in some paths

File: Multiple handlers Risk: Some older handlers may call sys.exit() directly rather than returning an integer. This bypasses the top-level exit code handler and prevents cleanup.

CLI-R-003: argparse --help in subcommand exposes internal options

File: cli/__init__.py Risk: Internal/developer flags (e.g., --debug-audit, --skip-plan-check) are exposed in --help output to all users. Failure mode: Users accidentally use dangerous flags.

CLI-R-004: No input sanitization for task strings

File: cli/_handlers/_chat.py, cli/_handlers/_agent.py Risk: The task string from the CLI is passed directly to the agent and LLM. A crafted task string could attempt prompt injection. Failure mode: Prompt injection if task is treated as trusted input.

CLI-R-005: Config loading from multiple sources with no precedence documentation

File: cli/execution.py Risk: Config is loaded from env vars, ~/.teaagent/config.toml, .teaagent/config.toml, and CLI flags. Precedence order is implicit. Failure mode: Unexpected config values; hard to debug.

CLI-R-006: completion_command may write to stdout, breaking shell integration

File: cli/_handlers/_misc.py Risk: completion_command outputs shell completion scripts to stdout. If any other code in the startup path writes to stdout (e.g., a log message), the completion script is corrupted.