nacharbeit

Use to apply the haiku- and sonnet-tier backlog of a completed nacharbeit review: one remediator per file at the file's tier, a blind sonnet verifier that re-reads the file and runs the post-checks its kind needs (py_compile, node --check, the hook test, the viewer standard), one repair round on rejection, all under a PreToolUse hook that denies every edit outside the fix lock. Trigger on 'apply the nacharbeit findings', 'fix what the review found', 'rework these plugins to the standard', 'nacharbeit fix'. Requires a persisted, calibrated review in analysis/nacharbeit/; opus- and human-tier entries are listed for a person and never applied; never commits.

Use to run the mechanical half of the nacharbeit rubric — 90 script-checkable rules over a plugin's skills, agents, commands, workflows, hooks, scripts, report viewers, manifest, README, CHANGELOG and docs wiring — with no model in the loop and no tokens spent. Trigger on 'lint this plugin', 'does this plugin meet the Anthropic standard mechanically', 'check the frontmatter and hooks.json', 'nacharbeit lint'. Reports findings by rule; never fixes anything (that is nacharbeit-fix) and never judges prose (that is nacharbeit-review).

Use FIRST, before any other nacharbeit skill, to inventory what the target plugin directories contain (skills, agents, hooks, scripts, viewers, manifest, docs wiring), which checkers this environment can run, which other plugins' guard hooks are live in the repository, and whether a nacharbeit fix lock is already open. Trigger on 'what would nacharbeit check here', 'is this plugin ready for a review', 'nacharbeit preflight', or as the automatic first step of nacharbeit-review and nacharbeit-fix. Read-only — writes nothing, grades nothing; grading is nacharbeit-lint and nacharbeit-review.

Use to measure whether a documented example prompt actually fires the skill it names, by running it headless in fresh claude -p processes — the real answer the review's haiku routing simulation only approximates. Trigger on 'does this prompt actually fire the skill', 'did my description fix work', 'why does this skill never trigger', 're-measure routing after the fix'. Costs real money per cell, so it always dry-runs and states the estimate first. Not for grading prompt quality (nacharbeit-review) or for benchmarking plugin-present versus plugin-absent sweeps (arbeitsplan-matrix).

Use to run the calibrated judgement review of one or more Claude Code plugins against the nacharbeit rubric: a sonnet finder tuned on planted-defect fixtures until it clears a recall floor for every rule family, measured once on a sealed hold-out, then two-lens finders with gap rounds per (plugin, kind) batch, a haiku routing simulation for cannibalization, an adversarial refuter, and an opus synthesis into a tiered backlog and a findings report. Trigger on 'review this plugin against the Anthropic standard', 'where do these skills and hooks fall short', 'run the nacharbeit review', 'prompt-quality review'. Runs nacharbeit-lint first; never applies a fix (that is nacharbeit-fix). Writes only analysis/nacharbeit/ and, inside werkstoff, the generated findings report.

Use to report what nacharbeit has already run in this repository — the last lint, the last calibrated review with its sealed recall per family and blind spots, the backlog by tier with every opus- and human-tier entry named, the last fix pass and its post-check result, and an open fix lock with its age. Trigger on 'nacharbeit status', 'what did the plugin review find', 'is a fix pass still open', 'which findings need a human'. Read-only; reports on nacharbeit's own state only and has nothing to say about a repository nacharbeit has not run in.