🗺️ AI Engineering · Journey Playbooks

Playbook: Harden an agent before giving it tools

Ship an autonomous agent whose failure modes are cheap, visible, and reversible.

journey~125 minAI EngineersML EngineersLLM App Developers

Journey Steps

1Step 1 — Design tools an agent can use reliably: start with "One tool = one verb-noun responsibility; split anything doing two jobs"
2Step 2 — Add guardrails to autonomous agent loops: start with "Set hard max-step and max-token budgets; treat exhaustion as a normal outcome"
3Step 3 — Defend pipelines against prompt injection: start with "Separate instruction and data channels structurally, not just by prose"
4How it fits together: Design tools and guardrails together; retrofitting safety after autonomy is how headlines happen.

Commands

Install all referenced skills

$ npx skills add aniruddhaadak80/skills --skill agents-tool-design && npx skills add aniruddhaadak80/skills --skill agents-agent-loop-guardrails && npx skills add aniruddhaadak80/skills --skill inference-mlops-prompt-injection-defense

Tags

#playbook#journey#ai-engineering

Install this skill

$ npx skills add aniruddhaadak80/skills --skill playbook-ai-engineering-agent-safety -a opencode

SKILL.md

# Playbook: Harden an agent before giving it tools

> Ship an autonomous agent whose failure modes are cheap, visible, and reversible.

## When to Use This Skill

Ship an autonomous agent whose failure modes are cheap, visible, and reversible.

## Steps

1. Step 1 — Design tools an agent can use reliably: start with "One tool = one verb-noun responsibility; split anything doing two jobs"
2. Step 2 — Add guardrails to autonomous agent loops: start with "Set hard max-step and max-token budgets; treat exhaustion as a normal outcome"
3. Step 3 — Defend pipelines against prompt injection: start with "Separate instruction and data channels structurally, not just by prose"
4. How it fits together: Design tools and guardrails together; retrofitting safety after autonomy is how headlines happen.

View source on GitHub ↗

Take retrieval-augmented answers from empty repo to evaluated production feature.

🗺️ AI Engineering·~135m

Open →

🗺️Playbook: Cut LLM costs without quality collapsejourney

Reduce monthly inference spend measurably while keeping answer quality within tolerance.

🗺️ AI Engineering·~95m

Open →

🗺️Playbook: Rescue a slow frontendjourney

Move a lagging app's Core Web Vitals into green without a rewrite.

🗺️ Frontend Engineering·~105m

Open →