Solo
No subagents, no delegation. Read the full architecture before touching it — fix or propose solutions based on root causes at the right layer, not symptoms at the wrong one. For parallel independent tasks, use /subagents. For persistent multi-task coordination, use /team.
Research
- •Use Read for research, not Grep. Grep is acceptable for locating files and symbols — once found, Read the file
- •Read whole files. Never use offset/limit on files under 500 lines
- •Follow every reference — imports, callers, siblings, tests, configs
- •Read the surrounding architecture before proposing changes. The file with the symptom is rarely the file with the problem
Hard Rules
- •No Agent tool — enforced by the
enforce-solo-modehook - •No skimming — if a file matters, read it fully
- •No hedging — never say "likely" or "probably" about code you can read
- •Assume existing code is intentional — if the reason for something is unclear, research more before changing or proposing changes
Process
- •Locate — find the relevant files
- •Read — every relevant file fully, plus surrounding architecture
- •Assess — is this a real issue? Trace the code path to confirm. If not, report findings and stop
- •Understand — which layer owns this responsibility? Fix it there, not where the symptom surfaces
- •Propose or act — at the correct abstraction layer
- •Verify — trace changes through the code paths you read. Run tests