// agent skills for Claude Code · Cursor · Codex
Review IAM permissions, retries, spending limits, and approval steps with instructions your coding agent can load during a task. Each skill includes worked examples and a scored rubric so you can compare results with and without it. Start with the free IAM skill.
13 skills · worked examples · scored eval rubrics · one-time purchase · individual & team licenses
A skill is a portable SKILL.md your coding agent loads contextually.
It gives the agent instructions and reference material for a matching task, such as
drafting an IAM policy or reviewing a retry loop. Check that your tool loaded the skill,
then review its output. These files guide generation; permissions, budgets, and approval
gates still need to be enforced by your application and infrastructure.
Core rules, a workflow, and a review question
("if this agent were compromised right now, what's the worst it could do?").
Drops into .claude/skills/ (Claude Code), .agents/skills/ (Codex) or .cursor/skills/ (Cursor).
Worked examples of a risky pattern and a proposed correction. These are authored fixtures, not measured results from paired agent runs. Use them to understand the checks, then test your own model and task.
A fixed prompt, point-scored rubric with auto-fail conditions, and an LLM-as-judge prompt. Run independent sessions with and without the skill, then score both outputs. Repeat the comparison: your baseline may already do well, and the skill may add no benefit. Review any judge scores yourself. CI integration is up to you.
Eight skills for teams building multi-agent and self-improving systems on AWS. Covers permissions, task dispatch, spending, deployment, execution, retries, tracing, and recursion.
| # | Skill | Review focus | |
|---|---|---|---|
| 01 | agent-iam-least-privilege | Wildcard IAM / PassRole escalation in agent roles | FREE |
| 02 | idempotent-agent-fan-out | Double-processing and dropped tasks on retry | |
| 03 | agent-cost-ceilings | Unbounded token/dollar burn in agent loops | |
| 04 | coordinated-agent-rollout | Deploys that silently break peer agents | |
| 05 | sandboxed-agent-execution | Agent-written code reaching the host / network / creds | |
| 06 | self-healing-agent-retries | Retry storms and cascading dependency failures | |
| 07 | agent-observability-scaffolding | Unreconstructable async multi-agent failures | |
| 08 | bounded-agent-recursion | Self-improving agents that never terminate |
Use one skill for a specific review or follow the cross-references when a task spans several concerns.
Five skills for agents that read untrusted content, hold credentials, and take consequential actions. Covers input validation, prompt injection, secrets, approval gates, and memory. Several evaluations use adversarial inputs to check how the agent responds.
| # | Skill | Review focus |
|---|---|---|
| 01 | tool-output-validation | Acting on a tool / API / sub-agent result that's malformed or hostile |
| 02 | prompt-injection-boundaries | Untrusted content hijacking the agent's tools |
| 03 | agent-secrets-handling | Credentials leaking via prompt, tool args, logs, or memory |
| 04 | human-in-the-loop-gates | "Approval" the agent can grant itself |
| 05 | agent-memory-hygiene | Memory poisoning, cross-tenant bleed, stale re-injected state |
The skills follow the path untrusted input takes through an agent: what it reads, what it holds, what it's allowed to do, and what it remembers.
The free IAM download and every skill in the paid packs include an evaluation prompt, a scored rubric, and a judge prompt. Keep the model version, tools, and settings the same; use independent fresh sessions and repeat each condition.
1. skill off: paste the eval prompt into a fresh session # save the output
2. skill on: fresh session, same prompt and model # confirm skill loaded
3. score both against the rubric, inspect errors, and repeatRecord the model, settings, outputs, and scores. A rubric comparison is limited evidence; it does not establish production safety. Recheck after model or workflow changes. The free IAM skill includes references, worked examples, a checklist, and the evaluation files, so you can run the comparison yourself before buying anything.
Individual — you, across every project and client. Team — one price, every engineer and repo in your org.
One-time purchase, no subscription, and free updates when the skills or evals change. Choose the pack that matches the work you need to review. Buying both separately totals $118 ($49 + $69) for individual licenses.
One-time purchase, licensed org-wide: no seats, no subscription, and free updates when the skills or evals change. Buying both separately totals $228 ($99 + $129) for team licenses, covering every engineer and repo in your organization.
$49$99
security & control
$69$129
reliability & operations
Each button opens a separate Gumroad purchase. Select the matching license at checkout.
Not sure yet? Install the free skill: the skill, references, worked examples, checklist, evaluation prompt, rubric, and judge prompt. No paid pack required.
SKILL.md skills (the open Agent Skills format). Claude Code, Codex and Cursor all load them natively and pick a skill when your task matches its description: install to .claude/skills/, .agents/skills/ or .cursor/skills/ respectively. Other tools that read AGENTS.md can use the same text as always-on instructions.SKILL.md, references and pre-merge checklists, worked examples, and the evals: a fixed prompt and scored rubric per skill, plus an LLM-as-judge prompt. You run and score the evaluations yourself. Discipline also includes a script for checking rollout compatibility.