feat(plan): persist plans with step-by-step execution

[12britz]

feat(plan): Persist plans with step-by-step execution

Summary

Replaces the bare-prompt /plan command with a full planning workflow: structured checklist plans saved to .claude/plan.md, terminal display, editor integration, step-by-step LLM execution with auto-checkoff, and manual tracking.

Changes Made

New file: code_puppy/command_line/plan_manager.py

Core module for plan file I/O, step parsing, execution prompt building, and editor support.

• PlanManager class manages .claude/plan.md
• Parses - [ ] / - [x] checklist items with 1-based indexing
• mark_step_done(N) toggles completion in the file
• build_run_prompt([N]) generates LLM prompts for executing one or all pending steps
• edit_in_editor() opens the plan in $EDITOR / $VISUAL (creates template if missing)
• create_template() returns a bare-bones plan skeleton
• show_formatted() displays the plan with a progress header

Modified: code_puppy/command_line/core_commands.py

Replaced the simple prompt wrapper with a full subcommand handler:

Subcommand	Action
/plan <goal>	LLM explores codebase, outputs structured checklist, saves to .claude/plan.md
/plan show	Displays current plan with (N/M steps complete)
/plan edit	Opens .claude/plan.md in $EDITOR
/plan run [N]	Sends one or all pending steps to LLM for execution
/plan mark <N>	Toggles step N done/undone

Added alias p → /p works as shorthand.

Modified: code_puppy/cli_runner.py

Added _process_plan_step_markers() — intercepts PLAN_STEP_DONE: N in LLM responses and auto-checks the corresponding step in the plan file. Runs after every turn (main loop + continuation loop).

Modified: code_puppy/help_system/help_content.py

Updated /plan help entry with new syntax and examples.

New file: tests/command_line/test_plan_manager.py

28 tests covering PlanManager: I/O, step parsing, mark-done/undone, display formatting, editor launch (with mocked subprocess), execution prompt builders, and marker stripping.

Modified: tests/command_line/test_core_commands_full_coverage.py

6 tests for the new plan command subcommands: show (with/without plan), edit, mark, goal prompt, and usage error.

Testing

pytest tests/command_line/test_plan_manager.py -v      # 28 passed
pytest tests/command_line/test_core_commands_full_coverage.py -v  # 55 passed
pytest tests/command_line/ -v                           # 1960 passed (no regressions)

Technical Details

• Prompt engineering: The plan prompt explicitly allows Write tool for .claude/plan.md while blocking other modifications
• Step execution: build_run_prompt() includes the full plan for context plus a ## Pending steps section with only unchecked items
• Auto-checkoff: The PLAN_STEP_DONE: N marker is stripped from the user-visible response after processing
• Editor fallback: Uses $VISUAL → $EDITOR → nano in that order; open -t is not used (runs $EDITOR directly)
• No permission mode: Execution guard is prompt-based only (the LLM is instructed not to modify files during /plan, but doesn't enforce via a permission mode)

Files Modified

File	Status
code_puppy/command_line/plan_manager.py	New
code_puppy/command_line/core_commands.py	Modified (+100 / -18)
code_puppy/cli_runner.py	Modified (+35)
code_puppy/help_system/help_content.py	Modified (+10 / -6)
tests/command_line/test_plan_manager.py	New (28 tests)
tests/command_line/test_core_commands_full_coverage.py	Modified (+79 / -1)