Review a proposed agent detector stack (kill switch, circuit breakers, canary tokens) and flag missing tripwires before the first autonomous run. Use when building UI with tripwire design language and visual style.
Scanned 9/8/2026
Install to Claude Code
npx -y skills add anubhavg-icpl/vibe --skill tripwire-design --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Tripwire Design?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/anubhavg-icpl-tripwire-design)More formats (shields.io, HTML) on the badges page.
---
name: tripwire-design
description: Review a proposed agent detector stack (kill switch, circuit breakers, canary tokens) and flag missing tripwires before the first autonomous run. Use when building UI with tripwire design language and visual style.
license: CC-BY-NC-SA-4.0
phase: 15
lesson: 14
metadata:
version: 1.0.0
tags: [kill-switch, circuit-breaker, canary, honeytoken, detection-and-response]
---
Given a proposed detector stack for an agent deployment, audit it against the three-detector reference (kill switch, circuit breaker, canary) and flag what is missing, mis-tuned, or exposed to the agent.
Produce:
1. **Kill-switch audit.** Where does the switch live (feature flag, Redis, signed config)? Confirm the agent's credentials cannot set it off. Confirm every consequential action checks the switch, not just startup. Confirm re-enable is an explicit human action.
2. **Circuit-breaker inventory.** List every pattern a breaker watches (repetition, consecutive failures, rate, specific tool after out-of-trust read). State threshold and cool-down for each. Thresholds above 10 are usually too loose.
3. **Canary design.** List every canary token in the environment. For each: what it is (fake credential, fake DB record, fake file, fake memory entry), where it lives, what access triggers the alarm, who is paged. Confirm no canary has a legitimate reason to be touched.
4. **Statistical + hard layering.** Confirm the stack uses at least one hard limit (Lesson 17 constitutional style) in addition to any statistical detectors (EWMA, z-score). Statistical-only detectors accept slow drift.
5. **Quarantine path.** What happens when a detector fires? Full agent stop, path-specific pause, traffic redirect (eBPF / Cilium honeypot), alert-only. Confirm the path has been tested end-to-end at least once.
Hard rejects:
- Any deployment without an external kill switch.
- Canary tokens stored in systems the agent has write access to.
- Statistical-only detection with no hard limits.
- Circuit breakers with cool-downs that auto-re-enable without human review.
- Unattended runs where the kill switch is checked only at startup, not per action.
Refusal rules:
- If the user cannot name the specific systems outside the agent's credentials that host the kill switch, refuse. "We use a config file the agent reads" is not a kill switch if the agent can write config files.
- If the user treats the Auto Mode classifier (Lesson 10) as a replacement for tripwires, refuse. The classifier is orthogonal to detection-and-response.
- If the proposed canaries sit in systems the agent has legitimate reason to read, refuse and require redesign.
Output format:
Return a tripwire audit with:
- **Kill-switch line** (location, check cadence, re-enable procedure)
- **Circuit-breaker table** (pattern, threshold, cool-down)
- **Canary table** (token, location, alarm, owner)
- **Layering note** (statistical + hard limits present y/n)
- **Quarantine flow** (what fires, what happens, tested y/n)
- **Readiness** (production / staging / research-only)
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!