Skip to content

nWave-ai/nWave / nWave/skills/nw-rigor

nw-rigor

Verified

Use when the user wants to see or change how much rigor a project gets (review rounds, examine level), which provider/model a role uses, or whether readable documents are written. The LLM reads the current settings, asks only the missing choices in plain words, and runs `nwave-ai rigor` or `nwave-ai model set` itself. Never derives a model from a competence label.

NW-RIGOR

The user collaborates with you. You run every command; the user never types terminal commands. nwdes does not execute models, and the CLI neither maps competence to a model, checks availability, nor falls back.

1. Read what is already set

Run nwave-ai rigor --show (read only) and read its values and their sources. Also read the project .nwave/config.json and the global .nwave/config.json (under a valid absolute NWAVE_AGENTS_HOME, otherwise the home directory). Reuse anything the user already said in this conversation; never ask them to repeat a known value or type an identifier already available.

2. Ask once, in one batch

Ask ONE batched question set in plain words (short sentences, everyday words), not one round trip per item. Say what each item changes for the user's work. Keep tool names (option names, producers, reviewers, executors, role names, charter, slice, lane, handoff) in your commands only, never in a sentence the user reads. Cover only what is missing or what the user wants to change. Each item gives: the current value and source as the suggested answer, the cost and benefit of the options, and your recommendation. End with one direct question ("keep these, or change which?").

  1. Review rounds (1-5). How many times a second person, not the author, may check the same piece of work. Cost: one more check per round. Benefit: fewer mistakes stay unseen. Too few rounds can miss mistakes; many rounds can keep finding small things and never stop.

  2. Examine level. A final check of the finished work, by someone who has not read the code; it finds what automatic tests miss. Explain "examination" as one such check and "repair round" as one chance to fix what it found. Present the tool's own text, copied here once and pinned by a test:

    Levelgivesnot covered at this level
    quick1 examination, no repair roundany repair round
    standard (default)2 examinations, 1 repair roundmore than 1 repair round
    extended3 examinations, 2 repair roundsmore than 2 repair rounds

    Not covered at any level: security, load or checks on other machines. The tool also says: a row not examined is never green. A longer level is not safer.

  3. Which AI model does which kind of work. Say it as: a stronger model for hard work, a faster one for routine work, for all work of one kind or for one named helper. Offer the model already in effect or known from the config or the conversation; keeping it is the default. Map the answer to --model-set with a group (producers, reviewers, executors) or an exact published role name.

  4. Readable summary pages. A web page in plain language, for people who do not read code, written when a piece of work is finished: yes or no. When yes, also the language (a code such as en, de, it).

  5. Other AI tools on this computer. Whether they should also get the model choices. Writing changes files in the user's home folder; a backup copy is kept next to each. Two tools (codex, devin) cannot take a model per kind of work and refuse. Default: change nothing.

Rounds and levels bound the rework loops. A finding still open at the limit returns to its owner; reaching the limit is never an approval and nothing passes silently.

3. Apply with ONE command

The command takes every chosen value together. State it before running, run it, report the result, then run nwave-ai rigor --show and show the new values. Use current or keep for what does not change, and --yes: the user's answers are the confirmation. Give each option once.

nwave-ai rigor --yes --review-rounds 3 --examine-level extended --models keep --documents yes --document-language en --hosts keep

That example sets 3 rounds, the extended level and English documents, and changes no models or host file. Adapt only the values. The options:

  • --show: print the values in effect; writes nothing.
  • --yes: the user's answers are the confirmation.
  • --review-rounds: 1 to 5, or current.
  • --examine-level: quick, standard, extended, or current.
  • --models: keep; or use --model-set once per target, as --model-set reviewers=PROVIDER/MODEL.
  • --documents: yes, no or keep; --document-language LANG only with yes.
  • --hosts: host names separated by commas, none or keep.

After a refusal, report its WHAT/WHY/HOW and ask the user; never substitute another value, provider, model or scope.

Provider/model selection

Model facts for native-host role invocation are set per target: --model-set above (groups producers, reviewers, executors, or a published role name) or, for the default or one role, nwave-ai model set:

Scope and targetCommand
Global defaultnwave-ai model set --provider PROVIDER --model MODEL
Current-project defaultnwave-ai model set --provider PROVIDER --model MODEL --project
Global published-role overridenwave-ai model set --provider PROVIDER --model MODEL --role ROLE
Current-project published-role overridenwave-ai model set --provider PROVIDER --model MODEL --role ROLE --project
  • Suggest the effective current pair (project role override beats global default) with its source; the user accepts or changes it.
  • A requested competence level helps the user choose a pair; the CLI has no competence profile or mapping. Keep the current pair until the user selects another.
  • PROVIDER is codex or claude; MODEL the accepted exact string; ROLE only an exact published role name. Never invent an identifier, a competence mapping or a fallback; the stored value is not checked, so a wrong name shows only when the role runs.
receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementationStars: 322kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 322kUpdated Claude CodeCodexVerifiedInstall
using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallbackStars: 322kUpdated Claude CodeCodexVerifiedInstall
verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions alwaysStars: 322kUpdated Claude CodeCodexVerifiedInstall