methodology-skill-creator

Methodology Skill Creator

Turn one selected method and its canonical eight-field contract into one reviewable methodology skill. This is not a general skill creator.

Installed dependency

The renderer imports the manifest and selected method reference from the sibling methodology-selector skill. Confirm that both skills are installed under the same skills root. If the sibling is missing, stop and report the required pair; do not copy or recreate the canonical method.

Preconditions

Read references/creator-contract.md.

• methodology-selector has returned a manifest methodology, not none;
• the canonical contract contains all eight shared fields;
• selection.method_ref matches the selected methodology;
• the caller supplies a new lowercase, hyphenated skill name that does not collide with an existing skill;
• the caller supplies skill_summary required;
• one method and one task scope are requested.

Stop on a failed precondition. Do not choose a replacement method or broaden the skill.

Discovery vs runtime

Contract checklist (exact):

1. skill_summary required
2. description formula uses skill_summary not task
3. Triggers suffix max 3
4. ## Invocation holds full alias list
5. Fit stripped; When to use is fit surface
6. summary ≤ 160; description ≤ 400; alias ≤ 48
7. newline rejected; whitespace normalized
8. decapitalizeForGlue on summary and best_when before description glue
9. nested Apply-formula reject only for Apply {manifest display name} to …
10. limit failures throw before checks; no new CREATOR_CHECKS keys

Frontmatter description is cold discovery only. The eight fields stay in the body. Optional invocation aliases: first three may appear as a Triggers suffix in description; the full list renders under ## Invocation.

Workflow

1. Preserve Task, Audience, Context, Output Format, Validation, Method Fit, and Human Review / Stop without changing their meaning.

2. Read only the canonical reference named by the selection.

3. Render one SKILL.md:

   node /absolute/path/to/methodology-skill-creator/scripts/render-methodology-skill.mjs /absolute/path/to/input.json
   

   Resolve the first path from the directory containing this SKILL.md. The agent prepares the JSON input; do not ask the human to hand-write it.

4. Check that the output reads as direct runtime instructions rather than an assignment report: Objective, Operating instructions, Required deliverable, When to use, optional Invocation, Workflow (without ## Fit), Execution checks, Evidence to return, and the original human-return boundary. It must also have repo-compatible frontmatter, one methodology, and the canonical quality questions. Description must follow the skill_summary formula and must not leak task, audience, or embedded-method boilerplate. character_count is evidence for review, not a 4,000-character gate. Keep the skill focused, but allow it to exceed 4,000 characters when the task contract and canonical method content require that space.

5. Confirm the target directory still does not exist immediately before any authorized write.

6. Return the proposed file and evidence. Write it only when the caller asked for repository changes and the target is inside the current repository.

Boundaries

• Do not absorb skill-draft-ship draft/ship/purpose gates; this skill only renders methodology SKILL.md content.
• Do not select a methodology or merge multiple methods.
• Do not invent method principles outside the canonical reference.
• Do not generate itself.
• Do not add the full methodology catalog in one run.
• Do not execute the task described by the generated skill.
• Do not judge loop readiness or approve the generated skill.
• Do not truncate or reject a skill merely because it exceeds 4,000 characters; that hard limit belongs only to generated loop goals.
• Do not put contract.task or audience into the frontmatter description.

Stop

Stop after one generated skill or one explicit validation failure. Human review remains required wherever the canonical contract requires it.