Using Skills

Overview

This skill enforces the discipline of skill invocation, preventing the agent from shortcutting workflows or rationalizing its way out of using available frameworks. It is the core defensive mechanism against AI overconfidence and drift.

Iron Law

IF THERE IS EVEN A 1% CHANCE A SKILL APPLIES, INVOKE IT AI models are biased toward immediate generation over reference-checking. This rule forces a "pause and check" behavior that prevents low-quality, ungrounded output.

State Machine

digraph using_skills_flow {
    "Task Received" [shape=doublecircle];
    "Step 1: Check Skill Database" [shape=box];
    "Decision: Is it ≥ 1%?" [shape=diamond];
    "Step 2: Propose Skill to User" [shape=box];
    "Step 3: Apply Skill Logic" [shape=box];
    "Done" [shape=doublecircle];

    "Task Received" -> "Step 1: Check Skill Database";
    "Step 1: Check Skill Database" -> "Decision: Is it ≥ 1%?";
    "Decision: Is it ≥ 1%?" -> "Step 2: Propose Skill to User" [label="yes"];
    "Decision: Is it ≥ 1%?" -> "Done" [label="no"];
    "Step 2: Propose Skill to User" -> "Step 3: Apply Skill Logic";
    "Step 3: Apply Skill Logic" -> "Done";
}

When to Use This Skill

• Whenever a new request is received.
• When an ongoing project shifts into a new phase (e.g., from Analysis to Deliverable).
• When you find yourself "guessing" the right approach.

When NOT to Use This Skill

• Simple, factual lookups with no strategic or creative component.
• Basic conversational chitchat.

Core Process

Step 1: Scan for Applicable Skills

Check the taxonomy of 41 skills across 6 families. Do not rely on memory; use the taxonomy to identify potential matches. (Source: system design)

Step 2: Apply the 1% Rule

Ask: "Is there even a 1% chance that [skill-name] would improve the outcome of this task?" If yes, it is mandatory to mention it. (Source: system design)

Step 3: Present Options to User

State clearly which skills apply and what value they add. Do not make the decision for the user; present it as a professional recommendation. (Source: system design)

Step 4: Execute with Full Rigor

Once a skill is selected, you MUST follow its state machine and iron laws without exception. (Source: system design)

Cross-Skill Invocations

REQUIRED SUB-SKILL: getting-started — for initial session context. RECOMMENDED SUB-SKILL: prompt-optimizer — to refine the invocation of other skills.

Rationalization Table

| Thought | Reality | |---------|---------| | "The user didn't ask for a skill, so I won't use one." | The user may not know which skills are available. It is your job to recommend them. | | "This task is slightly different from the skill description." | Use the 1% rule; even a partial match provides better structure than no structure. | | "I've already used three skills, adding another will be too complex." | Complexity is often exactly what a complex problem needs. | | "I can just apply the principles of the skill without formally invoking it." | Formal invocation is the only way to ensure the state machine and iron laws are followed. |

Red Flags

These thoughts mean STOP — you are about to shortcut:

• "I'll just handle this one manually" → You are avoiding the 1% rule.
• "This skill is overkill for this small request" → Small requests are where frameworks catch the most hidden errors.
• "I don't have the sources for that skill handy" → You should use the extraction methodology to find them.

Diagnostic Checklist

• [ ] Have I checked the full list of 41 skills for this task?
• [ ] Did I apply the 1% rule correctly?
• [ ] Have I explained the "why" behind the recommended skill(s)?
• [ ] Am I following the state machine of the invoked skill?
• [ ] Did I resist the urge to "wing it"?

Sources

• System design — The 1% Rule.
• System design — Rationalization Defense.
• System design — Iron Law Registry.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: JoelLewis
• Source: JoelLewis/skill-library
• License: MIT

Install and usage instructions live in the source repository linked above.