cvsz/zaff

bounded-execution

Use when executing approved implementation or operational plan steps that mutate files, services, repositories, or external state

cost-governance

Use when an agent loop, long-running task, or multi-agent workflow needs explicit budget, context, retry, or escalation control

execution-planning

Use when requirements must become a finite executable plan with acceptance criteria, validation, rollback notes, and stop conditions

independent-verification

Use when checking whether completed work meets acceptance criteria, especially when the implementer should not be the only verifier

loop-orchestration

Use when running a bounded autonomous engineering loop with discovery, planning, execution, verification, repair, and handoff decisions

persistent-memory

Use when a loop or long-running task must persist factual state, decisions, evidence, blockers, or next actions outside model context

repair-loop

Use when verification or review finds failures that need bounded root-cause analysis, correction, and re-verification

repository-discovery

Use when starting work in a repository and needing targeted context on instructions, source, tests, dependencies, CI, or failures

worktree-isolation

Use when multiple agents or concurrent tasks may mutate a repository and need isolated branches, worktrees, or file ownership boundaries