An additive Claude Code library — skills, agents, and rules. Covers what generic workflows don't: session journaling, learned-correction persistence, prompt expansion, context discipline (token budgets, strategic /compact), TypeScript/web rule baselines, plus a handful of safety and lookup helpers.
This is not a workflow framework. It does not enforce phases, gate your work, or auto-inject context. Skills load on demand via the Skill tool. Use what you need, ignore the rest.
/plugin install void-grimoire@claude-plugins-official/plugin marketplace add dmltdev/void-grimoire
/plugin install void-grimoire@dmltdevDownload only the skill you need with npx skills:
npx skills add dmltdev/void-grimoire --skill session-summaryReplace session-summary with any skill name from the list below.
34 skills across 7 domains, plus a 3-agent toolkit and a plugin-local rules/ reference tree. Pick by name:
| Domain | Skills | Description |
|---|---|---|
| void-grimoire | expand-prompt, learn-correction, autoresearch, skill-forge, infer-patterns, strategic-compact, unslop, unslop-design | Prompt expansion, self-learning, skill optimization, high-quality skill authoring, repo-pattern skill pack generation, context discipline, code slop cleanup with agent-instruction memory, and product UI unslopping |
| tools | using-herdr, using-codex, using-omp, omp-plugins, using-adhd, using-agent-browser, using-chrome-devtools-mcp, using-elevenlabs-tts | External CLI/MCP wrappers — preflight, invocation, fallbacks, and plugin install/update workflows for herdr, Codex, oh-my-pi, adhd, agent-browser, chrome-devtools-mcp, and ElevenLabs TTS |
| qa | test-with-browser | Evidence-based UI verification — drive a browser against acceptance criteria, capture screenshots/console/network, write a report under .test-results/ |
| workflow | session-summary, session-usage-summary, session-friction, quick-recap, human-typed-plan, atlas-research, ideal-example-clone, refactor-transaction, failure-memory-compiler, audio-plan, audio-recap | Session journaling, AI-usage feedback, append-only friction log for correction events, final red/yellow/green response status, human-owned implementation planning, evidence-backed Atlas research, ideal-example cloning, clean refactor cutovers, operational failure memory, and TTS-ready plan/recap scripts |
| docs | lookup-docs, index-docs | Documentation search via qmd, with first-class openspec/specs awareness |
| git | enforce-git-safety, commit-push-pr, create-pr | Block destructive git ops; commit/push/PR helper; concise risk-sized PR body rubric |
| npm | enforce-release-safety | Pre-publish safety checks |
Read-on-demand subagents under agents/:
silent-failure-hunter— zero-tolerance review for swallowed errors, empty catch blocks, dangerous fallbacks, broken error propagation.type-design-analyzer— evaluates type design across encapsulation, invariant expression, usefulness, and enforcement.blast-radius-cartographer— read-only impact mapper for implementation planning. Produces DOCS/CODE/TESTS/MEMORY impact maps with evidence anchors before edits.
Plugin-local rules/ reference tree under three buckets: common/ (language-agnostic baselines), typescript/ (TS/JS-specific guidance with the matching common baseline inlined), and web/ (frontend/web-specific guidance with the same flattening). Read on demand; never auto-injected.
session-summary— Write a session journal: TL;DR, decisions with trade-offs, accomplishments, unfinished work, files touched. Use before/compactor at session end.session-usage-summary— Retrospective on the human-AI loop in this session. Scores spec clarity, decision ownership, verification depth, and correction loops.quick-recap— Adds the final red/yellow/green status-line convention for finished, pending, or blocked responses.learn-correction— When you correct the AI ("don't mock the DB", "always use snake_case"), persists the correction to your project'sAGENTS.md/CLAUDE.mdso it survives future sessions.expand-prompt— Turn a terse request ("add dark mode") into a structured intent: relevant docs, learned rules, decomposed sub-tasks. Requires explicit user approval before any action.strategic-compact— Suggests manual/compactat phase boundaries (planning -> implementing -> verifying) so context survives the next phase rather than waiting for arbitrary auto-compaction.autoresearch— Run a skill repeatedly, score outputs against binary evals, mutate the prompt, keep improvements. Karpathy-style autonomous skill optimization.
Each skill is a SKILL.md file with frontmatter (name, description; some older skills also include depends-on, chains-to, suggests). Skills are loaded on demand — there is no startup hook, no .void-grimoire/ directory, no forced gate flow.
Composition still works:
depends-on— listed skills must run firstchains-to— the named skill is invoked after this one completessuggests— soft recommendation, agent checks if relevant
skills/registry.json is a domain → skills catalog used for documentation. Claude Code loads relevant skills automatically via their descriptions; invoke any skill by name when you want it explicitly.
It deliberately does not enforce workflows by default. Agentic SDLC orchestration lives outside this plugin; Void Grimoire stays a utility/helper library with no mandatory gates or hidden state.
Most design/audit/polish skills still live in the Impeccable plugin and belong with their authors. unslop-design is the narrow exception: a product-workflow redesign helper for rough SaaS/admin screens.
The original architecture spec described a multi-gate flow with .void-grimoire/ config, rules, and registry-driven routing. That system has been retired — the spec is kept in docs/specs/ for historical reference but no longer reflects the plugin.
If you're into fantasy, see docs/LORE.md for why this plugin is a grimoire.
MIT