Guard

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

MANDATORY PREPARATION  Invoke /agent-workflow — it contains workflow principles, anti-patterns, and the **Context Gathering Protocol**.

Audit 0.4.1 · source sha 4e163a7a8095 · 19 Oct 2026Report an issue or request a re-scan

• Publishersharpdeveye GitHub userOwner of the repository this skill is read from
• Source foldersharpdeveye/maestro/source/skills/guardThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict4e163a7a8095A new audit runs when the source changes
• Repository statusNot archivedLast push 173 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Use when deploying to production, handling sensitive data, or the workflow needs safety constraints, input validation, and security boundaries.

---
name: guard
description: "Use when deploying to production, handling sensitive data, or the workflow needs safety constraints, input validation, and security boundaries."
argument-hint: "[threat or area]"
category: enhancement
version: 2.0.0
user-invocable: true
---

## MANDATORY PREPARATION

Invoke /agent-workflow — it contains workflow principles, anti-patterns, and the **Context Gathering Protocol**. Follow the protocol before proceeding — if no workflow context exists yet, you MUST run /teach-maestro first.

Consult the guardrails-safety reference in the agent-workflow skill for the full defense-in-depth framework.

---

Add safety boundaries to a workflow. Guards protect against malicious inputs, unintended outputs, data leakage, cost explosion, and all the ways an autonomous system can go wrong in the real world.

### Threat Assessment

Before adding guards, understand what you're protecting against:

| Threat | Risk Level | Guard Type |
|--------|-----------|-----------|
| Prompt injection | High | Input sanitization, instruction hierarchy |
| PII leakage | High | Output filtering, data masking |
| Cost explosion | High | Token budgets, rate limits |
| Unauthorized actions | Medium | Permission scoping, confirmation gates |
| Hallucination | Medium | Source attribution, fact checking |
| Service abuse | Medium | Rate limiting, authentication |

### Guard Implementation

**Input Guards**

```text
Before processing any input:
1. Validate against schema (reject malformed)
2. Check size limits (reject oversized)
3. Sanitize for injection patterns
4. Rate limit check (reject if exceeded)
5. Authentication/authorization check
```

**Output Guards**

```text
Before returning any output:
1. Schema validation (format correct?)
2. PII scan (names, emails, SSNs, etc.)
3. Content policy check
4. Confidence threshold check
5. Source attribution present?
```

**Cost Guards**

```text
Before every model/API call:
1. Check remaining budget
2. Estimate request cost
3. If estimate > remaining budget → reject or use cheaper alternative
4. After call → update spent amount
5. Circuit breaker check (too many failures?)
```

**Permission Guards**

```text
For every tool call:
1. Is this tool allowed for this user/context?
2. Is this a destructive operation? → require confirmation
3. Is this accessing data the user is authorized for?
4. Log the access for audit trail
```

### Guard Checklist

- [ ] All inputs validated before processing
- [ ] PII detection on all outputs
- [ ] Cost ceiling set with enforcement
- [ ] Prompt injection defenses active
- [ ] Destructive operations require confirmation
- [ ] All access logged for audit
- [ ] Circuit breakers on external services
- [ ] Rate limits on all endpoints

### Recommended Next Step

After adding guards, run `/evaluate` with adversarial test scenarios to verify guards hold under attack.

**NEVER**:

- Deploy without input validation
- Trust model output for high-stakes decisions without verification
- Run without cost controls
- Skip logging (you need the audit trail)
- Assume the model will follow safety instructions 100% of the time

Excerpt, first 3,130 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/sharpdeveye/maestro
git -C maestro checkout 4e163a7a8095

Works in

1. 2026-10-19
   !CAUTION B 89 / 100

   first audit · source 4e163a7a8095

Questions about Guard

• github.com/sharpdeveye/maestro/tree/main/source/skills/guard — the audited source at commit 4e163a7a8095 and every finding aboveObserved · 19 Oct 2026

Spotted something wrong? Report a correction · Methodology