48 independently installable specialist packs · 54 skill entrypoints · 530 optional reference sheets
Focused contracts and failure checks for engineering, AI/ML, design, writing and delivery. Skills assume a capable model: routine requests need no routing ceremony, references load selectively, and reviewers report evidence and uncertainty without quotas.
In Claude Code, add the marketplace, then install the packs you need:
/plugin marketplace add tachyon-beep/skillpacks
/plugin install axiom-python-engineering
/plugin install yzmir-deep-rl
For local development, clone this repository and use /plugin marketplace add . from its root. Repository shortcuts under .claude/commands/ point to the same source contracts; plugin commands live inside each independently installable pack. See command shortcuts.
Counts describe actual SKILL.md entrypoints, not reference sheets. Each pack links to its source and carries its own metadata, optional references, commands and agents. The faction names are organizing themes; see FACTIONS.md.
| Pack | Skills | Purpose |
|---|---|---|
| axiom-audit-pipelines | 1 | Use when decisions require verifiable provenance, tamper-evident exports, or defensible retention and replay limits. Ordinary application logging does not require this workflow. |
| axiom-contract-engineering | 1 | Use when typed records cross subsystem authority boundaries and absence, meaning changes, blinded views, or derived decisions must remain explicit and reproducible. |
| axiom-determinism-and-replay | 1 | Use when past stateful execution must be reproduced or investigated: RNG isolation, snapshots, external-effect substitution, scheduling, rollback or first-divergence localization. |
| axiom-devops-engineering | 1 | Use when building or fixing delivery and operations: artifact identity, rollout health, rollback/restore evidence, infrastructure and runtime reliability. |
| axiom-distributed-systems | 1 | Use when multiple processes or failure domains must remain correct through crashes, partitions, retries, duplicates, clock changes or overload. |
| axiom-embedded-database | 1 | Use when SQLite or DuckDB application behavior depends on connection configuration, write concurrency, atomic claims, migrations, backup/restore, FTS or query/storage boundaries. |
| axiom-experiment-formalisation | 1 | Use when experiment records need ontology interoperability, verified EXPO/SUMO/BFO/PROV mappings, executable competency questions or auditable description. Do not activate for ordinary experiment execution. |
| axiom-mcp-engineering | 1 | Use when designing or reviewing MCP server surfaces: model-visible capability contracts, host negotiation, retries, authorization, bounded output and real task evaluation. |
| axiom-planning | 2 | Use when reviewing an implementation plan for source accuracy, missing dependencies, failure paths or acceptance evidence. |
| axiom-procedural-architecture | 1 | Use when designing or reviewing procedures, runbooks or interactive flows whose stages, decisions and handoffs must be correct for a particular audience. |
| axiom-product-management | 1 | Use when deciding product value or maintaining product ownership across sessions: durable state, decision provenance, falsifiable acceptance and authority. |
| axiom-program-management | 1 | Use when coordinating delivery across workstreams with real dependency agreements, risks, capacity limits and evidence-based forecasts. |
| axiom-pyo3-interop | 1 | Use when a PyO3 extension needs safe ownership/lifetimes, array exchange, thread attachment, async cancellation, interpreter teardown, wheel compatibility or measured FFI performance. |
| axiom-python-engineering | 1 | Use for Python-specific compatibility, type/lint repair, cancellation/resource handling, profiling, packaging or Textual lifecycle issues that need focused checks against a real project. |
| axiom-rust-engineering | 1 | Use for Rust-specific ownership/API, async cancellation, unsafe/FFI soundness, feature/toolchain compatibility, diagnostics, packaging or profiling problems in a real crate. |
| axiom-rust-workspaces | 1 | Use when multi-crate Cargo composition, feature unification, workspace policy inheritance, internal/public APIs, publishing order or target-specific test subsets cause concrete problems. |
| axiom-sdlc-engineering | 5 | Use when an organization explicitly selects a software governance profile requiring lifecycle evidence, traceability, risk decisions or measurement. |
| axiom-solution-architect | 1 | Use when making or reviewing system design choices against measurable requirements, alternatives, migration constraints and source evidence. |
| axiom-static-analysis-engineering | 1 | Use when building analyzers or rules whose traversal, abstract domain, unknown-call handling, termination, incremental cache or findings output must have explicit guarantees. |
| axiom-system-archaeologist | 1 | Use when reconstructing an existing codebase: source-linked architecture, dependency and risk findings, coverage limits and incremental refresh. |
| axiom-tensor-compiler-engineering | 1 | Use when tensor lowering/transforms, torch.fx/compile capture, numerical tolerances, artifact identity or cache behavior need reference-versus-compiled conformance checks. |
| axiom-web-backend | 1 | Use when implementing or reviewing production API behavior: authorization, compatibility, transaction boundaries, retries, bounded results and contract tests. |
| bravos-game-design | 1 | Use when inventing, diagnosing, repairing, balancing, or playtesting games across digital, tabletop, live, physical, educational, asynchronous, or hybrid media. |
| bravos-simulation-tactics | 1 | Use when game simulation needs a fidelity/LOD decision, engine implementation, frame-budget evidence or desync diagnosis. |
| bravos-systems-as-experience | 1 | Use when gameplay depends on interacting mechanics, player experimentation, optimization, emergent stories or a modding ecosystem. |
| lyra-creative-writing | 1 | Use when drafting, revising, critiquing, or planning prose fiction or creative nonfiction, especially when author voice, continuity, genre expectations, or truth claims need deliberate treatment. |
| lyra-site-designer | 1 | Use when building or maintaining static developer-tool, open-source, or documentation sites, including navigation, versioned URLs, code examples, theming, and build or deployment pipelines. |
| lyra-tui-designer | 1 | Use when building or reviewing interactive terminal applications, especially capability detection, rendering, input/focus, terminal restoration, accessibility, and cross-environment behavior. |
| lyra-ux-designer | 1 | Use when designing or reviewing application interfaces, audience needs, interaction flows, accessibility, or AI surfaces across web, mobile, desktop, and games. |
| meta-skillpack-maintenance | 1 | Use when reviewing, simplifying, extending, or validating this marketplace’s plugins, skills, commands, agents, references, hooks, or packaging. |
| muna-document-designer | 1 | Use when producing or validating Pandoc/Typst documents, reusable templates, print-ready PDFs, multilingual layouts, or accessible publication artifacts. |
| muna-panel-review | 1 | Use for optional staged synthetic reader reviews: precommitted expectations, controlled exposure and evidence-linked hypotheses requiring human validation. |
| muna-technical-writer | 2 | Use when documentation accuracy, executable procedures, institutional register, sensitive information, or source-to-document traceability needs deliberate review. |
| muna-wiki-management | 1 | Use when related documents need explicit source lineage, audience-specific derivatives, claim consistency, reading paths, or controlled change propagation. |
| ordis-quality-engineering | 1 | Use when test reliability, isolation, coverage gaps, performance experiments, resilience, or release evidence needs investigation or a concrete verification strategy. |
| ordis-security-architect | 1 | Use when system trust boundaries, threat models, control verification, AI/tool privileges, supply-chain assurance, classified information flows, or authorization evidence need design or review. |
| yzmir-counterfactual-statistics | 1 | Use when paired or branched ML experiments need uncertainty estimates, grouped splits, selection-bias controls, power analysis, or abstention calibration. |
| yzmir-deep-rl | 1 | Use when reinforcement-learning environment, reward, data-regime, algorithm or evaluation decisions need implementation checks. |
| yzmir-dynamic-architectures | 1 | Use when a neural network grows, prunes, composes modules or adapts across tasks and needs lifecycle, gradient-isolation or state-integrity checks. |
| yzmir-llm-specialist | 1 | Use when an LLM application needs model/configuration evaluation, retrieval, context/cost controls, fine-tuning, tool reliability or safety checks. |
| yzmir-ml-production | 1 | Use when a model or ML dataset must be released, served, monitored or recovered with operational evidence. |
| yzmir-morphogenetic-rl | 1 | Use when an RL controller decides when or how to change neural-network topology and needs governor, rollback, replay or fair-evaluation checks. |
| yzmir-neural-architectures | 1 | Use when neural architecture families or components must be compared against data, quality, compute, latency or adaptation constraints. |
| yzmir-pytorch-engineering | 1 | Use when PyTorch code has tensor/gradient, checkpoint, AMP, compile, distributed, memory or profiling obligations. |
| yzmir-simulation-foundations | 1 | Use when a simulation needs numerical-model, integration, stability, control, stochastic or determinism checks. |
| yzmir-structure-synthesis | 1 | Use when a generator emits typed graph/program candidates and needs legality, canonical identity, diversity or search-space checks. |
| yzmir-systems-thinking | 1 | Use when recurring interventions, feedback, delays or accumulation need a causal model and evidence-backed comparison. |
| yzmir-training-optimization | 1 | Use when a training run needs evidence about convergence, objective, gradients, regularization, search budget, batch or precision strategy. |
Marketplace 4.0.0 reduces 52 packs/61 entrypoints to 48 packs/54 entrypoints. Specialist domains remain; broad tutorials and compulsory orchestration were reduced. This is a content refocus, not a measured claim of improved model performance.
| Retired entrypoint | Retained owner |
|---|---|
yzmir-ai-engineering-expert |
AI/ML specialist catalog; install the relevant specialist directly |
axiom-engineering-foundations |
Solution-architecture debugging/refactoring checks and existing delivery specialists |
axiom-system-architect |
Solution-architecture evidence, impact and debt assessment |
meta-sme-protocol |
Shared evidence contract and inline agent rules |
| SDLC design-and-build, quality-assurance, platform-integration | Retained SDLC governance/requirements; optional solution architecture, quality and DevOps owners |
The consolidation record covers every original pack, migrations and validation limits. Historical review files remain labeled as historical. Retired plugins are removed from this marketplace; uninstall/update any separately cached old installation in your client.
See CONTRIBUTING.md for packaging, evidence and versioning policy. Run the repository integrity checks before committing. Technical examples must be verified against the project's actual toolchain before use. Current standards/compliance claims need named authoritative sources; local checks do not establish certification.
Licensed under CC BY-SA 4.0, with the license addendum. Code of conduct.