m3taz-ahmed/aizee / skills/bmad-toolsmith
bmad-toolsmith
VerifiedSmithy the Toolsmith builds and maintains BMad skills, agents and modules. Use when the user asks for Smithy or says "build a skill", "make an agent", "scaffold a skill", "turn my session logs into a skill", "fix this skill", "convert this prompt to a skill", "review this skill", "optimize this skill", "cut this skill down", "make this skill BMad compliant", "package a module", "validate my module" or "migrate my old module". Not for running evals (`bmad-eval`) or overriding an installed skill (`bmad-customize`).
modes/cut-roles.md
View on GitHub# Cut roles
The subagent prompts `modes/cut.md` and `ship.md` use, one role per heading. Fill each angle-bracket placeholder, an empty one as "none", and write `{target}` and `{skill-root}` as absolute paths. `<old file>` is the file's path in the copy taken before the rewrite.
## Cutter
```text
Rewrite <file>, one file of the skill at {target}, into the leanest prose that steers the model as far as the old text did. The bar is `{skill-root}/canon.md` and `{skill-root}/lenses/lens-leanness.md`; read both. Start each passage from the one sentence that steers the model (canon test 2) and add back only what prevents a wrong action you can name, listing each in `added_back`. Then split what still holds several topics (canon test 9). There is no token target; leave a lean file as it is.
Keep, in the fewest words that fix it:
- Every convention the model cannot know: file and folder names, id formats, frontmatter keys, which step writes which section, what the scripts read, state and status words.
- Every rule and every disambiguation. A list of examples the model could produce itself is neither.
- Every Jinja expression and tag, character for character; every {skill-root} and {project-root}; every command line.
- Every heading, and every name another file of the skill refers to.
- Every cross-file reference.
- These passages, word for word: <protected>. These decisions; nothing you write may contradict them: <decisions>.
Plain declarative sentences; a Markdown paragraph is one line; a TOML comment keeps its comment form; no new terms; add nothing that is not a compression of what was there, apart from a heading a split needs; keep the section order. Edit <file> in place, touch no other file, commit nothing.
Return JSON: { "file": path, "rewrites": [one line per passage removed or recast, beyond trimming a why-clause], "added_back": [{ "clause", "wrong_action" }], "doubts": [anything you were not sure was explanation rather than rule, with your choice] }
```
## Checker
```text
<file> in the skill at {target} was just rewritten to be leaner. Read the old file at <old file>; the new file is on disk. The rewrite had to keep every rule, convention, disambiguation, Jinja construct, heading and cross-file reference, these passages word for word: <protected>, and these decisions: <decisions>. Assume it failed somewhere and find where.
Walk the old file sentence by sentence. List a sentence under "lost" only when the new file no longer states it in any wording and you can name the different action a model would take without it; write that action in "action". A why-clause, a definition of general knowledge, a restatement or an illustration is not a loss. A protected passage not kept word for word is a loss whatever the action. When "lost" has more than five entries, check each again against the action test and drop those that fail.
Separately list where the new wording changes what a rule says, contradicts a decision, or moves a rule so it no longer covers what it covered ("changed_meaning"), and up to five passages that fewer words would still say as well, each with its shorter wording ("remaining_fat"). Edit nothing.
Return JSON: { "file": path, "lost": [{ "old_text", "action" }], "changed_meaning": [{ "old_text", "new_text", "why" }], "remaining_fat": [{ "text", "shorter" }] }
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |