Build a Command Security Pipeline for Your Agent

PROJECT CONTEXT:
My agent executes shell commands guarded by a handful of tangled regex checks that are impossible to extend safely. Claude Code runs 20+ independent validators in a deliberately ordered pipeline where each check does one job and the most severe finding wins.

GOAL:
Build a layered command-validation pipeline: a shared result contract, early fast-path validators, a main chain of single-purpose checks, and severity-aware result selection.

REQUIREMENTS:
- A shared validator contract: every check returns allow, ask, deny, or passthrough
- Early validators that short-circuit common safe patterns before the main chain
- Single-purpose main-chain validators - one attack class per check
- Hard path constraints that no permission rule or prompt can override
- Severity ordering: parser-mismatch findings outrank informational ones

CONSTRAINTS:
- Validators must not share mutable state - independence is the point
- A new validator must be addable without editing existing ones

DEFINITION OF DONE:
- All command execution flows through the pipeline - no bypass path
- Each validator has isolated unit tests with attack and benign cases
- Removal of any protected system path is blocked even with explicit permission rules
- A command triggering multiple findings reports the most severe one

COMMON FAILURES TO AVOID:
- One mega-function of nested ifs where fixing one check breaks another
- Validators that mutate shared parse state, creating order-dependent bugs
- Path checks that trust permission rules over the hard denylist
- First-match-wins reporting that surfaces a warning while hiding a critical finding