Design plan for bash-level parallel specialist reviews with
Read/Grep/Glob grounding, citation verification, and confidence
scoring. Intended for multi-commit and PR-level reviews where
agent-loops single-turn is insufficient.
Priority 1 & 3 code review improvements:
- Add source path existence validation before all symlink operations
- Add (OSError, PermissionError) try/except blocks to symlink creation/removal
- Separate error handling for unlink and symlink_to operations
- Fix uninitialized source_path in _agent_activate_recursive (init to None)
- Add specific error messages for each failure mode
Files changed:
- agents.py: Updated _agent_activate_recursive and agent_deactivate with
proper error handling and source_path initialization fix
- skills.py: Enhanced skill_activate/skill_deactivate with source validation
and specific exception handling
- rules.py: Updated rules_activate/rules_deactivate to catch PermissionError
and provide clear error messages
- asset_installer.py: Apply same patterns to all install functions
This ensures robust symlink operations with clear feedback on failure modes.
Remove cortex-workflow references since the command is not implemented
and workflow orchestration is currently handled through composition of
existing commands (agent, skills, review, plan, export, ai).
- Remove workflow subcommand from validation and generation
- Remove cortex-workflow.1 from manpage bundle
- Delete placeholder cortex-workflow.1 manpage file
This keeps the build clean and only bundles manpages for implemented
commands (cortex.1 and cortex-tui.1).
Nicholas Ferguson