Garrett Manley

Garrett Manley's Claude Code plugin marketplace. Capability-bundled plugins for dev discipline, evidence-grounded research, local LLM orchestration, Windows tooling, autonomous stewardship, AI app development, and the Aether Engine TTRPG framework.

Generic dev hygiene: TODO+issue enforcement, frontmatter lint, plan validation (issue+value+retrospective), spec-code drift checker, finish-and-push workflow, and pitfalls-pointer routing, plus a fact-forcing edit gate, git-state checkpoint/snapshot tooling, and council + session-handoff skills, and a compact-plan skill for plan-preserving context compaction.

Evidence-discipline workflow: citation-seeker, truth-seeker, HMAC override token framework, scope-binding hook scaffold, secret-scan PreToolUse hook. Targets security research and any project where claims must be substantiated.

Local LLM tier handoff: Reasoning + Tooling tier model defs (llama-server), local-orchestrator skill (when to hand off to local), horizon-scanning skill (track SOTA <=14B models), hardware-profile awareness via ~/.claude/context/hardware-profile.md.

Windows + PowerShell helpers: copy-pasteable allowlist patterns for .claude/settings.local.json, PS5.1 vs PS7 routing decision tree, common Windows-isms (CRLF, BOM, $env:, path separators).

Autonomous maintenance toolkit: drift-check runner against context-file verification commands, morning-briefing template, Windows Task Scheduler registration helper, auto-memory housekeeping for ~/.claude/projects/*/memory/.

Opinionated meta-overlay for AI app development. Wires existing claude-api / agent-sdk-dev / mcp-server-dev plugins into a coherent starting point with opinionated defaults: latest model IDs, prompt-caching on by default, eval discipline, repo structure conventions.

Aether Engine TTRPG/narrative framework support: domain-specific hooks (cd_core_guard, ledger truncation, classifier/harness/Rust-rebuild reminders), 6 skills (doc-cluster, edit-checklist, plan-writer, ledger-doctor, eval-run, value-justify), classifier-regression-checker agent. Hooks no-op outside Aether-shaped repos; safe to enable globally.

Documentation craft: tech-writing (universal prose rules from Google Tech Writing), mermaid-diagram (Mermaid authoring with a configurable brand palette and renderer-agnostic fencing), design-document (structured design-doc creation with a pluggable template).

Multi-lens review: the reviewer-personas skill maps an artifact (design docs, PRs, work items, wiki pages, or skill files) to a set of 16 archetype agents and dispatches them as sub-agents (security-auditor, observability-champion, data-architect, incident-commander, ecosystem-context-reviewer, new-engineer, skill-craft-reviewer, architect, contract-surface, performance-scalability, test-strategy, migration-safety, compliance, api-contract, error-handling, accessibility); each agent definition is the persona. SessionStart hook nags about review-triggering artifacts that completed without a review token. Includes persona-stub template for adopters to add their own archetype agents.

Plan retrospective discipline: a self-contained cycle of marker-drop (exit plan mode), session-start nag, and a retro-authoring skill. Forces a pause after every approved plan to capture what worked, what surprised you, and concrete improvements.

Git workflow: Conventional Commits formatting that resists the common CI-guard failure modes, and PR creation that detects the remote host (GitHub / Azure DevOps Repos / GitLab) instead of hardcoding one. Pairs with a project-local writing-style fork for tone.

Continuous-learning toolkit: atomic instinct storage plus project-scoped observation hooks (default-off) and an /analyze-observations review report. Adapted from affaan-m/everything-claude-code @ 4774946d.

End-to-end value-delivery lifecycle: a /deliver command + skill that drives one body of work through plan -> adversarial plan review -> subagent execution -> completion gate -> adversarial code review -> land -> retrospective, composing superpowers + docs + retrospective skills. Project-specific plan/doc/checklist steps bind per-repo via .claude/delivery.local.md; no project-internal names are committed.