How to use the Content Audit agent skill (content-audit)

What the content-audit skill makes the agent do

An agent skill is a set of written instructions, here a SKILL.md file, that an AI agent such as Claude Code follows when you ask for a specific task. This one is about content counts in game design. The agent reads the project's systems index and design files, finds the numbers and named lists of planned content, and reports them as a planned-content inventory.

The skill is built to stop rather than guess. If the inputs it needs are missing, it reports NOT ASSESSED - NO DATA and names the missing file and the skill that creates it. It can produce a gap summary, and it can save that summary to a file or only show a summary table.

• Reads design/gdd/systems-index.md for the list of systems and their priority tiers.
• Uses design/registry/entities.yaml first, if it exists, for named content already listed with a source GDD.
• Scans the GDD files for content counts such as 'N enemies', 'N levels' or 'N items', and for explicit lists of named content.
• Builds a content inventory table of system, content type, specified count or list, and source GDD.
• Flags content described without a count as Unspecified, since that is a design gap.

Install the content-audit skill from the ZIP

Download the content-audit ZIP from this page. The skill is a single file, SKILL.md, in the source repository's .claude/skills/content-audit folder. Unzip it into one of two places, depending on how you want to use it.

For a personal install that works across your projects, unzip it into ~/.claude/skills/. To limit it to one game project, unzip it into that project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

• Personal (all projects): ~/.claude/skills/
• Single project: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

What your project needs before it runs

The skill needs Claude Code or a compatible agent that has the Read, Glob, Grep, Write and Bash tools. It also needs the design files it reads. Its own guidance says the systems index is required. Without it, the skill reports no data.

The design files come from other skills in the same studio workflow: /map-systems creates design/gdd/systems-index.md, and /design-system creates the GDDs. design/registry/entities.yaml is optional. If it is missing or empty, the skill falls back to scanning the GDDs, which takes longer but still works.

• Required: design/gdd/systems-index.md
• Required: the GDD files, normally in design/gdd/
• Optional: design/registry/entities.yaml
• Optional: the hooks/yaml-helper.sh config script

How to trigger the audit with a request

The skill is user-invocable and takes one optional argument. With no argument it runs a full audit across all systems. Give it a system name to audit one system only. Give it --summary to get the summary table without writing a file.

The argument format comes from the skill's argument hint. Invoke the skill by its name, content-audit, in your agent session, then add the argument you want.

• No argument: full audit of every system
• A system name: audit only that system's GDD
• --summary: summary table only, no file write

Examples of useful audit requests

Ask for a full audit before a production review to see which content types have planned counts and which have none. Ask for a single system, such as your combat or enemy design, when you only need that area checked. Use --summary when you want a quick view in the chat and do not want a file written to the project.

• Full audit before a milestone review to list every planned count against the implemented content.
• Single-system audit of one GDD to check its enemy or item counts.
• Summary-only run to see the planned-versus-built table in chat without saving a report.
• Review of Unspecified rows to find design areas that describe content but give no number.

Safety notes: scripts and network

The skill's allowed tools are Read, Glob, Grep, Write, and a Bash permission limited to one call: running resolve_config through hooks/yaml-helper.sh. It does not get general shell access in its allowed tools list. The skill's documented network use is none.

The skill can write files. A full audit with the default settings may write a report, so check the project's automation mode and the output location before you run it on a shared repository. Its automation settings say that collaborative mode asks before actions, and autonomous mode logs and proceeds.

• Network: none, according to the skill listing.
• Scripts: runs yaml-helper.sh only for config resolution.
• File writes: the skill can write a report unless you run --summary.
• Automation mode decides whether the agent asks before each write or proceeds.

Tips for getting reliable results

Keep your systems index current, since the audit starts from it. Name your content with explicit counts in the GDDs, such as '12 enemy types', because the skill looks for numbers next to game nouns. Check every Unspecified row, because the skill treats missing counts as a design gap, not as zero.

Read the reports carefully. A NOT ASSESSED result is a correct answer when data is missing, not a pass. The skill's guidance says absence of files is not evidence that nothing is built.

• Keep design/gdd/systems-index.md up to date.
• Write counts as numbers next to content nouns in your GDDs.
• Treat Unspecified rows as design gaps to resolve.
• Treat NOT ASSESSED - NO DATA as a missing-input result, not a clean pass.