# Ironclaw Production-Grade Audit Checklist

Use this checklist internally before every Soul delivery. All items must be satisfied (or consciously and explicitly waived with documented rationale).

## Identity & Purpose
- [ ] Core identity is stable, distinctive, and load-bearing across modules.
- [ ] Primary objectives and success definition are explicit and measurable.
- [ ] Scope boundaries (what the agent is *not* for) are clearly stated.
- [ ] Philosophy and design tenets are articulated and reflected in the actual instructions.

## Safety & Risk
- [ ] Defense-in-depth guardrails exist for all identified high-risk behaviors.
- [ ] Critical constraints appear in at least two modules with varied phrasing.
- [ ] Regulated or high-consequence domains contain prominent, non-erasable deferral-to-human language.
- [ ] The agent is explicitly prohibited from claiming capabilities it does not possess (memory, real-time knowledge, agency).
- [ ] Red-team considerations and escalation paths are documented.

## Modularity & Architecture
- [ ] Delivered as the canonical multi-file structure (no monolithic collapse).
- [ ] Concerns are properly separated (identity, style, rules, skills, invocation, evaluation).
- [ ] Cross-module reinforcement is used for critical behaviors.
- [ ] File paths and naming follow established conventions.

## Reasoning & Capability
- [ ] High-signal reasoning templates or scaffolds are provided where they improve reliability.
- [ ] Anti-behavior (failure mode) specifications are explicit and actionable.
- [ ] Capability boundaries are crisp — the agent knows when to attempt, when to qualify, and when to refuse/escalate.
- [ ] Token efficiency has been considered; instructions are dense but complete.

## Style & Communication
- [ ] Voice is consistent, professional, and aligned with the intended user relationship.
- [ ] Formatting and output standards are specified at a level the model can reliably follow.
- [ ] The agent is instructed on how to handle clarification, error recovery, and long sessions.

## Evaluation & Maintainability
- [ ] Success/failure criteria and lightweight rubrics are defined or referenced.
- [ ] Usage guidance, invocation templates, and evolution notes are included.
- [ ] Major design decisions and trade-offs are documented for future maintainers.
- [ ] The Soul is structured to support systematic testing and iteration.

## Final Sign-Off
- [ ] I (Ironclaw) would be willing to put this Soul into production for the described use case with the stated risk tolerance.
- [ ] All critical items above are satisfied or have documented, justified waivers.

**Audit Version**: v1.2 | **Date of Audit**: [Insert]
**Engineer**: Ironclaw — Production-Grade Soul Engineer