content-review

Run the full Millennium Dawn content review checklist against a file or the current branch diff.

Syntax: /content-review [file_path]

• With file_path: review that specific file against all content guidelines.
• Without argument: review all changed files on the current branch against main.

The authoritative checklist lives in docs/src/content/resources/content-review-guide.md and docs/src/content/resources/new-general-guidelines.md. Read both before starting if not yet read this session. The condensed .claude/docs/content-guidelines.md is a quick summary but not exhaustive; always defer to the full docs.

Execution

1. Gather context

File mode (path provided):

• Read the file and identify its type (focus tree, events, characters/generals, OOB, decisions, etc.).
• Read docs/src/content/resources/content-review-guide.md, docs/src/content/resources/new-general-guidelines.md, and .claude/docs/content-guidelines.md.

Branch mode (no argument):

• Run git diff origin/main...HEAD and git log origin/main..HEAD --oneline.
• Identify the changed files and their types.
• Read docs/src/content/resources/content-review-guide.md, docs/src/content/resources/new-general-guidelines.md, and .claude/docs/content-guidelines.md.

2. Apply the checklist

For each file in scope, check the categories in .claude/docs/content-guidelines.md (Economic, Political, Visual, Military, AI, Code, Variety, Miscellaneous). Skip categories that don't apply to the file type (no admiral counts in a decisions file, no building costs in a character file).

For focus-tree files, run the Variety check as a cross-comparison, not in isolation: pick a reference tree known for variety (Iran common/national_focus/05_iran.txt, Spain common/national_focus/05_spain.txt) and judge reward-shape diversity against it. A tree can pass every balance check and still be formulaic. See .claude/docs/content-guidelines.md → Variety & Anti-Formulaic Content.

Detection notes

Three failure classes that spot-checking reliably undercounts. Sweep the whole file for these, don't sample:

• Treasury double-charge (block-aware). MD's self-charging build effects (one_random_arms_factory, one_random_industrial_complex, two_random_*, one_random_infrastructure, one_office_construction, and similar in common/scripted_effects/) already deduct treasury internally unless set_temp_variable = { skip_payment = 1 } is set beforehand. A focus that calls one of these and adds an explicit set_temp_variable = { treasury_change = -X } modify_treasury_effect = yes with no add_building_construction to justify the second charge pays twice. Distinguish from the legitimate pattern: a raw non-charging engine effect (add_building_construction) paired with one explicit charge is correct. Grep every focus that calls a self-charging effect and check each for a redundant explicit charge; per-branch sampling misses most of them.
• Loc completeness checks referenced keys, not just new ones. Verify that every tooltip = <key>, idea key, and modifier _tt key referenced from the tree resolves in localisation/english/, not merely that the diff's newly-added keys resolve. A modernization pass routinely references a generic _tt key that never existed, and a diff-only check will not catch it.
• Dynamic-modifier first-add guard, per reachable path. A focus that does add_dynamic_modifier without a NOT = { has_dynamic_modifier = { modifier = X } } guard double-applies when a second focus in the same plan adds the same modifier. Check every add_dynamic_modifier against all other focuses reachable in the same branch, not just the one focus in front of you.

3. Output

For each file reviewed, report:

1. File: path and file type.
2. Issues: numbered list with category labels ([Economic], [Political], [Visual], [Military], [AI], [Code], [Variety], [Misc]) and line numbers where applicable.
3. Mark anything that must be fixed before merge as [blocker].

End with a total issue count per category, or "No content issues found."