Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By netzkontrast
Agency framework: 54 imported skills (39 sc-* + 15 superpowers-*) + 16 sub-agents (sc-pm-agent excluded — /sc:pm-only per CLAUDE.md §13.1) + 5 D.7-compliant event hooks (planned ST-3) + the agency governance substrate (Task / Prompt / Research / ADR layers).
npx claudepluginhub netzkontrast/agency-backupsApply Dramatica narrative theory (Phillips & Huntley, *Dramatica*, 4th ed., 2001) to story analysis, storyforming, drafting, and draft diagnosis. Dramatica models a complete story as one mind solving one problem, viewed through four throughlines (Overall Story, Main Character, Impact Character, Subjective/Relationship), with a structural model of 4 Classes / 16 Types / 64 Variations / 64 Elements selected to form a "storyform". Ships the full source book as nine thematic reference chunks plus an in-skill conceptual overview and storyforming quick-reference. Trigger phrases include — dramatica, story mind, storyform, throughline, grand argument story, archetype, protagonist antagonist guardian contagonist sidekick skeptic reason emotion, MC resolve, mental sex, story outcome judgment driver limit, signposts and journeys, crucial element, plot dynamics, phillips huntley, dramatica anwenden, throughlines bestimmen. Also for drafts that feel flat, characters feel unmotivated, or act structure is unclear.
Aktive Dramatica-Theorie für Storyform-Aufbau, Encoding und Storyweaving — kein passives Dictionary, sondern Werkzeug. Trigger explizit bei Dramatica, Storyform, Throughline, Class, Type, Variation, Element, Archetype, Dynamic Pair, MC, IC, Goal, Consequence, Cost, Dividend, Driver, Outcome, Judgment, Limit, sowie bei Archetypen-Namen Protagonist, Antagonist, Guardian, Contagonist, Sidekick, Skeptic, Reason, Emotion. Trigger proaktiv in Narrativ-Kontexten — bei novel-architect-Arbeit (Kohärenz Protokoll), Agency System Triptychon-Tracks (Album 1/2/3), Suno-Lyric-Arbeit mit klarem Charakterbogen, oder Diskussionen über Resolve (Steadfast/Change), Approach (Be-er/Do-er), Mental Sex (Linear/Holistic), Growth (Stop/Start). Liefert präzise Definitionen mit Dynamic Pairs, strukturelle Verortung in der Dramatica-Hierarchie, Encoding-Vorschläge und Konsistenz-Checks gegen die 75 Dynamic Pairs. Nicht greifen bei Hero's Journey, Save the Cat, Beat Sheets oder anderen explizit benannten Story-Modellen.
Use when the user wants to convert Google Docs or PDFs in a Google Drive folder to Markdown and upload the results to another Drive folder — without letting file contents land in the main context window. Supports two execution modes: Artifact-Generator (interactive React UI) and Subagent-Prompt (isolated API call for pipeline integration). Triggers on: drive to markdown, convert drive files, google docs to markdown, pdf to markdown drive, drive conversion, drive batch convert, context-safe drive, drive markdown pipeline.
MCP-driven Drive-zu-Notion Curator. Vier Operations: setup (komplett-Listing Source-Folder paginated, alle Files verarbeiten, DBs initial anlegen), routine (incremental, filter createdTime gt newest_in_last_run, cap 50 ASC, Status-Report bei Backlog), review-inbox (askuser-driven, on-demand), repair (komplett-Listing /curated/ paginated, diff gegen Notion). Kopiert Drive-Files EINMAL nach /Claude/curated/ flach, klassifiziert via Title plus Content gegen Topics-DB, indexiert klassifizierte Files als Sub-Pages mit Auszug, schickt unklare Klassifikationen in Inbox-DB. Notion ist State-Source. Sensitive Topics werden NICHT inhaltlich gelesen. Trigger bei drive aufräumen, drive sortieren, drive curate, kuratiere meine docs, neue docs einsortieren, drive cleanup, repair drive notion, drive scan, drive routine, drive sweep, files in topic einsortieren auch ohne explizites skill-Wort. v0.4.0 nutzt Drive-MCP plus Notion-MCP.
Schema cheatsheet, canonical vocabulary (463 appreciations + 144 narrative_functions), validation rules, 10-stage authoring workflow, and runnable schema validator for NCP (ncp-schema.json v1.3.0). Actively co-invokes dramatica-theory (storyform decisions, why a Class/Type/Variation is correct) and dramatica-vocabulary (Dynamic-Pair validation, KTAD coherence, Element-Quad checks) at explicit workflow checkpoints — this skill owns JSON-IO and enum-compliance; the two Dramatica skills own meaning. Use when the user mentions NCP, narrative-context-protocol, ncp-schema, .ncp.json, 'convert to NCP', 'validate NCP', 'storyform JSON', or Subtxt export. Trigger auch auf Deutsch — Storyform anlegen, NCP-Datei validieren, NCP-Skelett, Storyform aus Outline. Do NOT use for general Dramatica theory (defer to dramatica-theory or dramatica-vocabulary) or prose drafting. Path A per TODO T-1.
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Frontend design skill for UI/UX implementation
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Memory compression system for Claude Code - persist context across sessions
Standalone image generation plugin using Nano Banana MCP server. Generates and edits images, icons, diagrams, patterns, and visual assets via Gemini image models. No Gemini CLI dependency required.
Marketing skills for AI agents — conversion optimization, copywriting, SEO, paid ads, ad creative, and growth
Implements the CONTEXT.md workflow system for managing human-authored context and agent-ready briefings in Claude Code
Harness-in-harness agency engine: code-mode capabilities (macroskills) over one provenance graph.
Skillstash workflows and skills for authoring agent skills
Decoupling Machine, Actor, Space, and Capability for Long-Horizon AI Agency.
agency is a governance and orchestration repository for long-horizon work performed by AI agents (Claude Code, Gemini, Jules, and humans). It is not an application. It is the substrate on which agentic work is planned, instructed, executed, audited, and continuously improved — engineered so that work done across many sessions, by many agents, in many contexts, stays coherent.
LLM agents lose coherence over long horizons. They forget intent, drift from prior decisions, blur the line between what should be done, what the agent was told to do, and what running it produced. They re-author prompts mid-research, inline instructions inside tasks, append follow-up questions to closed deliverables, and silently re-interpret governance.
This repository is an opinionated answer to that drift. It treats agentic work as a system with four intentionally decoupled concerns:
| Concept | Question it answers | Lives in |
|---|---|---|
| Machine — the Task | What should be done? (orchestration, plan, todo, ownership) | /tasks/ |
| Actor — the Prompt | What is the agent told to do? (executable instruction set) | /prompts/ |
| Space — the Research | What did running it produce? (evidence, synthesis, output) | /research/ |
| Capability — the Skill | What does the agent know how to do? (reusable, container-loaded competence) | /skills/ |
The decoupling is enforced both socially (via specs) and mechanically (via linters and a pre-commit hook). A Task MUST NOT inline a prompt. Research MUST NOT author its own instructions. A Skill MUST declare its capability surface in SKILLS.md's skill_* namespace and MUST NOT be invoked from a Task or Prompt without an explicit skill_references_* link. Follow-up questions discovered during a run MUST NOT be appended to a closed research workspace — they MUST be filed as new prompts. The audit graph that links the four concerns is the source of truth.
/tasks/<NNN>-<slug>/task.md
│ task_uses_prompts ──► /prompts/<slug>/prompt.md
│ executed by agent ──► /research/<slug>/output/SPEC.md
│ open questions ──► /prompts/<new-slug>/ (prompt_kind: follow-up)
You do not need to install or run anything to read this repository — most of the value is in the specs and the artifacts they govern. To commit changes you need Python 3 (with PyYAML, jsonschema, pytest) and the pre-commit hook installed; the canonical session bootstrap is ./install.sh followed by tools/check-governance.sh (see §9 Quick start and AGENTS.md "Session Setup").
The slogan "Decoupling Machine, Actor, Space, and Capability" is a design constraint, not branding. Each layer has its own filesystem home, its own governance spec, its own frontmatter namespace, and its own pre-commit checks. Crossing the boundaries is an anti-pattern that the linters catch.
/tasks/ (orchestration)A Task is a bounded, named unit of coordination work. It carries a Goal, a Plan, a Todo checklist, and explicit links to the prompts it executes and the research it spawns. Tasks live in /tasks/<NNN>-<slug>/task.md and are governed by TASK.md.
A Task is not an instruction set. It says what should happen; it links to a Prompt that says how.
/prompts/ (instruction)A Prompt is a self-contained, executable instruction set written for a specific agent. Prompts live in /prompts/<slug>/prompt.md alongside an immutable brief.md capturing the original user request. Prompts are governed by PROMPT.md and MUST satisfy seven engineering principles (self-containedness, framework declaration, RFC 2119 normativity, deliverable lock, anti-ambiguity, constraint isolation, failure handling).
A Prompt is not a Task. It does not coordinate; it instructs.
/research/ (execution)