department-of-prose
CLAUDE.md2 starsChanged 57 days agoMITView on GitHub
A Claude Code plugin / cross-agent skill family for contextual writing review and revision. Three sibling skills share one versioned editorial catalogue. The optional lexical scorer and rhythm checker produce raw measurements, not authorship verdicts.
In this repository3 other files it gives its agents.
- slop-checkSkill
- slop-explainSkill
- slop-senseSkill
The file
We split this CLAUDE.md into the rules it gives an agent
6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 30
Trigger disambiguation matters: All three skills share vocabulary domain.
- Very specific: line 35
Run npm ci && npm run evaluate for deterministic regression coverage.
- Very specific: line 36
Read the diff, run bash scripts/link-skills.sh, and try affected skills in a session for changes outside the deterministic scope.
- Very specific: line 41
Run python scripts/plugin_version.py bump X.Y.Z with a higher semantic version.
- Very specific: line 42
Run python scripts/plugin_version.py check --base origin/main and the packaging checks.
- Distinctive: line 43
Commit, PR to main. Both marketplaces pick up the new version on push.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md 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.
Files
The Department of Prose
A Claude Code plugin / cross-agent skill family for contextual writing review and revision. Three sibling skills share one versioned editorial catalogue. The optional lexical scorer and rhythm checker produce raw measurements, not authorship verdicts.
Structure
skills/slop-sense/SKILL.md— the rewriter (36 patterns, workflow, ai;dr mode). Edit this for rewrite-behavior changes.skills/slop-sense/scripts/score.sh— optional algorithmic scorer wrapper aroundnpx slop-detector. Falls back to qualitative-only if Node is missing. Measures lexical tells only (slop words, trigrams, contrast phrases). Reused byslop-checkvia the sibling path../slop-sense/scripts/score.sh.skills/slop-sense/scripts/rhythm.py— pure-Python (stdlib only) structural checker for dimensionsscore.shdoes not measure: burstiness, contraction ratio, aphoristic closers, anaphora, raw punctuation counts, and conservative prose cadence candidates. These are editorial heuristics rather than authorship evidence. Reused byslop-checkvia the sibling path../slop-sense/scripts/rhythm.pyand vendored into the claude.ai bundle.skills/slop-check/SKILL.md— review-only skill (no rewrite). Uses the shared versioned catalogue; allowed-tools deliberately exclude Write/Edit.skills/slop-explain/SKILL.md— per-pattern educational skill. Resolves user request to a pattern number 1-36 via the lookup table, then reads the matchingpatterns/NN-name.mdfile.skills/slop-explain/patterns/— 36 generated per-pattern evidence notes (01-significance-inflation.mdthrough36-reflexive-formality.md). The versioned catalogue is their policy source.docs/pattern-catalogue-spec.mdanddocs/pattern-catalogue-source-audit.md— editorial decisions and primary-source limits.scripts/sync-catalogue.pygenerates the four runtime catalogue copies and 36 evidence notes;--checkdetects drift..claude-plugin/plugin.json— plugin manifest. Lists all three skills inskills[]sonpx skills@latestcan enumerate them together..claude-plugin/marketplace.json— Claude Code marketplace listing. Mirrors the shared plugin version fromplugin.json..codex-plugin/plugin.json— Codex plugin manifest. Points at the same top-levelskills/directory and includes Codex install-surface metadata..agents/plugins/marketplace.json— Codex repository marketplace. Exposes the repository root as thedepartment-of-proseplugin.scripts/plugin_version.py— checks or bumps the shared version across both plugin formats and enforces version bumps for plugin changes.scripts/link-skills.sh— dogfooding helper; auto-discovers everySKILL.mdunderskills/and symlinks each into~/.claude/skills(override withSKILLS_DEST).dist-src/slop-sense-bundle/SKILL.md— hand-crafted merged super-skill for the claude.ai / Claude Desktop upload path (one ZIP per skill, so all three modes collapse into oneslop-senseskill). Source of truth for the bundle.scripts/package-claude-ai.sh— buildsdist/slop-sense.zipfrom the bundle SKILL.md + vendoredscore.sh+ vendoredrhythm.py+ vendoredpatterns/. Asserts exactly 36 pattern files. Output is gitignored.evaluation/— two-layer quality suite.fixtures/deterministic.jsondrives executable regressions,coverage.jsonmaps in-scope behaviors,baseline.jsontracks strict expected failures, andgolden/holds rubric-reviewed editorial contracts.package.json/package-lock.json— exactslop-detectordevelopment dependency used only for reproducible real-scorer integration tests..github/workflows/evaluation.yml— installs the locked scorer and runsnpm run evaluatewithout model calls.
When making changes
- Bump the shared plugin version with
python scripts/plugin_version.py bump X.Y.Zfor any user-visible change. The helper updates both plugin manifests plus the Claude marketplace mirrors; installed clients use this version as the update cache key. - The plugin is distributed through Claude Code and Codex marketplaces plus
npx skills@latest add CyranoB/department-of-prose(50+ agents). Don't add client-specific paths in any SKILL.md — they have to work cross-agent. - Trigger disambiguation matters. All three skills share vocabulary domain. When editing any SKILL.md description, keep the "Prefer X if..." cross-references intact so the agent's router picks the right skill.
- The scorer is optional by design. Don't make
score.sha hard dependency; the skills must still do useful qualitative analysis when Node is unavailable. - The
slop-checkskill calls the scorer via the sibling path../slop-sense/scripts/score.sh. If a cross-agent install ever flattens directory structure such that the relative path breaks, vendor a per-skill copy insideslop-check/scripts/rather than refactoring the path scheme. - Pattern numbering (1-36) is the load-bearing convention across all three skills. Change policy in the spec and source audit, then run
python3 scripts/sync-catalogue.py; update both lookup tables and packaging rules when IDs change. Re-runbash scripts/package-claude-ai.shafter any catalogue change. - Packaging checks use the Python standard library. Run
python -m unittest discover -s tests -v, both plugin validators, andbash scripts/package-claude-ai.shbefore release. - Run
npm ci && npm run evaluatefor deterministic regression coverage. Expected failures are executable assertions: an unexpected pass fails CI until its exemption is deliberately removed. - Read the diff, run
bash scripts/link-skills.sh, and try affected skills in a session for changes outside the deterministic scope.
Release flow
- Edit
SKILL.mdfiles / scripts / deep-dives. - Run
python scripts/plugin_version.py bump X.Y.Zwith a higher semantic version. - Run
python scripts/plugin_version.py check --base origin/mainand the packaging checks. - Commit, PR to
main. Both marketplaces pick up the new version on push.
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
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.

