Assemble Tool Prompts Dynamically per Environment

PROJECT CONTEXT:
My agent has one hardcoded system/tool prompt that describes features that are sometimes disabled, wasting tokens and confusing the model. Claude Code generates its BashTool prompt per environment: sandbox rules appear only when sandboxing is on, git protocols only when relevant, and config lists are deduped before injection.

GOAL:
Replace static tool prompts with a generator that assembles blocks conditionally from the live environment and measures the token cost of every block.

REQUIREMENTS:
- A prompt generator function per tool, not a static string
- Feature-gated blocks: sandbox, background tasks, git protocols included only when active
- Live config (allowed paths, hosts, limits) injected from the source of truth, deduped
- A token count logged per assembled prompt so drift is visible
- Different variants for different user classes where workflows differ

CONSTRAINTS:
- The generator must be deterministic for a given environment - same inputs, same bytes
- Never describe a capability the current environment has disabled

DEFINITION OF DONE:
- Toggling a feature flag visibly adds/removes its prompt block
- Injected config matches the enforcing code's actual values
- Token counts per block are logged and reviewed
- No prompt text references a disabled capability

COMMON FAILURES TO AVOID:
- A static prompt describing sandbox rules while sandboxing is off, teaching the model false constraints
- Config duplicated from multiple sources bloating every request by hundreds of tokens
- Prompt and enforcement drifting apart because values are hand-copied
- No measurement, so prompt bloat accumulates invisibly