Skill: prompt-transparency

Purpose

Inspired by the CL4R1T4S project (github.com/elder-plinius/CL4R1T4S), this skill audits and surfaces all implicit behavioral instructions, system prompts, and agent directives embedded in this repository. It generates a human-readable transparency report so anyone can understand exactly how the AI agents in this repo are instructed to behave.

When to Use

• When onboarding new contributors who want to understand agent behavior
• Before a release to document what behavioral rules are active
• When debugging unexpected agent behavior
• Periodically as a governance/transparency checkpoint

Steps

1. Collect All Agent & Skill Definitions

Scan and read every file that contains behavioral instructions:

• .claude/agents/*.md — named agent personas and their directives
• .agents/skills/*/SKILL.md — skill behavioral definitions
• .claude/commands/*.md — slash command behaviors
• CLAUDE.md (root) — global project-level instructions (if present)
• .claude/state/* — current runtime state

2. Extract Key Behavioral Dimensions

For each file found, extract:

• Role/Persona: What role does this agent/skill play?
• Constraints: What is it explicitly told NOT to do?
• Capabilities: What is it allowed/instructed to do?
• Tone/Style: Any communication style directives?
• Decision Rules: Any if/then behavioral rules?

3. Generate Transparency Report

Output a structured report to docs/prompt-transparency-report.md with:

• Summary table of all active agents and skills
• Full behavioral inventory per component
• Conflict detection (contradictory instructions across agents)
• Coverage gaps (areas with no behavioral guidance)

4. Flag Risks

Highlight any instructions that:

• Grant broad permissions without guardrails
• Could conflict with each other
• Are ambiguous or underspecified

5. Commit the Report

Use the smart-commit skill to push the report with a clear commit message.

Output Format

docs/prompt-transparency-report.md

Example Usage

/project:prompt-transparency

Inspiration

CL4R1T4S by Elder Plinius — a project dedicated to publishing AI system prompts
for public transparency. This skill applies that same philosophy internally:
every repo using AI agents should be able to show exactly what behavioral
instructions are active at any given time.