agentleFS
gdhub-iogdhub-io/character-forge-mcp/skills/forge-character/SKILL.md

forge-character

Skill1 starsChanged 4 days agoMITView on GitHub

Create a strong, coherent character (film, game, ad, book) with the character-forge MCP. Runs a variant tournament — 3 variants of the same concept, independent coherence reviewers and critics argue against each, the author defends once, a judge picks the strongest, which gets an A-pose concept. Use when the user asks for a character, hero, NPC, mascot, protagonist or villain, random or from a brief.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 2 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Call character_method once (method + protocol).

    line 12
  2. Very specific:

    Quick: only if the user explicitly asks for one fast character — variants: 1, skip steps 3–7, but still run character_check and the …

    line 18
  3. Distinctive:

    All variants share the anchor and world but have different cores — "all fishermen, but different fishermen".

    line 22
  4. Very specific:

    Write 3 full cards and give each a full, honest effort (a tournament with a straw man is useless).

    line 25
  5. Very specific:

    Call character_role with role: "author" and follow it: for each critique, fix the card (re-save with overwrite) or rebut in one line with a …

    line 38
  6. Very specific:

    Call character_concept for the WINNER only.

    line 49
  7. Very specific:

    Run the translate-character skill for each extra language (the card is free.

    line 55
  8. Very specific:

    Call character_get for the winner and show the card: it comes back as a Markdown document the app can display.

    line 58
  9. Distinctive:

    Keep the report short: the card is shown above.

    line 61

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Forge a character (variant tournament)

You are the AUTHOR. Independent agents review your work; they have fresh context and owe you nothing. Requires the character-forge MCP server (tools named character_*).

0. Setup

  • Call character_method once (method + protocol).
  • Languages: English is always the original. If the user names other languages (e.g. "also in Russian"), note them for step 8c.
  • Decide the mode:
    • Random: the user gave little or nothing. Anchor = the role they named, if any.
    • Brief: the user described what matters. Every variant must keep ALL of it. Anchor = the role from the brief; fix medium/genre/setting from the brief.
    • Concept, or concept + brief: the user gave their own concept image (path or URL), with or without a description. character_create tells you what to do with it: look at it first, every variant must look exactly like it and explain what it shows, and the winner keeps it as its concept.
    • Quick: only if the user explicitly asks for one fast character — variants: 1, skip steps 3–7, but still run character_check and the designer.

1. Create

character_create with whatever the user gave: brief (their words, verbatim) and/or concept (their image), plus anchor, medium, genre, setting when known. It rolls 3 variants by default and starts its answer with the MODE and how to treat the brief and the image: follow it. All variants share the anchor and world but have different cores — "all fishermen, but different fishermen".

2. Write

Write 3 full cards and give each a full, honest effort (a tournament with a straw man is useless). Each card: meta.group = the roll seed, meta.variant = A/B/C, meta.seed = variant seed, meta.mode, meta.brief if any. Save each with character_save; note each id and its card.json path.

3. Coherence review — 3 agents in parallel

Spawn the character-coherence agent once per variant, all in one message (never more than 4 at once). Prompt: the card.json path, the variant letter and the brief/anchor. Nothing else — do not tell them what you think of the card. Fix every BROKEN link they report; re-save with overwrite: true.

4. Prosecution — 3 agents in parallel

Spawn the character-critic agent once per variant (same rules: path + brief/anchor only).

5. Defense — you

Call character_role with role: "author" and follow it: for each critique, fix the card (re-save with overwrite) or rebut in one line with a quote. One round only. Keep the rebuttals.

6. Judgment — 1 agent

Spawn the character-judge agent with the 3 card.json paths, the brief/anchor, all critics' cases and your rebuttals. Optionally fold the judge's "steal" idea into the winner and re-save.

7. Record

Variants are saved as drafts (_drafts/ in the library). Write meta.verdict (winner, score, summary) into the winner and re-save with overwrite, then call character_pick with the winner's id: it moves the winner into the library and deletes the other variants. Losers are not kept, so note who they were for the report first.

8. Look & concept — winner only

Spawn the character-designer agent with the winner's card.json path. Merge its look (and any description.physical changes), re-save with overwrite. Call character_concept for the WINNER only. If the user gave their own image, give the designer the image path too (it must describe exactly what the image shows) and don't generate: character_pick has already made the image the concept.

8b. Edit — winner only

Spawn the character-editor agent with the winner's card.json path. It returns the card rewritten for a reader who knows nothing of the world (plain narration in a voice tuned to the character, no facts changed). Validate with character_check, re-save with overwrite.

8c. Languages — only if the user asked for more than English

Run the translate-character skill for each extra language (the card is free; the voice is paid and needs a yes). When the voice is made later, pass translations: [{ lang: "ru", text }] to character_voice so every language is recorded in one call, each performed fresh from the description.

9. Show the card and report (in the user's language)

  • Call character_get for the winner and show the card: it comes back as a Markdown document the app can display.
  • Winner: name, logline, why it won (2–3 sentences), signature item, path to card.md (and concept.png).
  • One line per losing variant: who they were and their fatal flaw.
  • Keep the report short: the card is shown above.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.