alexmskills — Claude Code Skills Marketplace
alexmskills is a curated Claude Code plugin marketplace of reusable,
self-improving skills and agents. Each skill is packaged as an independently versioned plugin, so
you install and update exactly what you need.
The emphasis is on self-learning tooling — skills that get better at your repository over time.
A CLAUDE.md that prunes and graduates its own decisions. A delivery crew that re-tiers its roles from
run history. A screenshot audit whose checklist tightens every time a defect slips past it.
/plugin marketplace add alexmond/alexmskillsThen install what you want — Getting Started has the full list.
Self-learning
Skills that accumulate knowledge about the repo they run in. State lives in the consuming repo’s
.claude/, never in the plugin, so what they learn survives an update.
| Plugin | What it does |
|---|---|
Turns CLAUDE.md into a living Decisions & Learnings log that prunes, graduates, and archives itself via three hooks. |
|
Keeps agent-written memory files from rotting — re-verifies version pins, sequence facts, and cited paths against the tree; flags what became wrong, never deletes. |
|
Reads every prompt you send and rewrites it toward better prompting habits — 42 rules across 6 tiers, mastery earned by demonstration, adaptive precision-gating, and a local web dashboard. |
|
The shared substrate: a per-repo registry of evolving personas ( |
|
Composes a task-fit roster and runs it as a gated delivery relay, each role a subagent on its own model tier. |
|
Assembles a task-fit panel of role-specialized agents, picks a coordination style, and runs a generate-critique-refine loop with a unanimity guard. |
|
Reads the whole screenshot, not just the bug you came for. Its defect checklist tightens each time something slips past it. |
|
Audits |
|
Auto-saves a durable learning to project memory whenever a task takes more than one fix cycle. |
Workflow
Guided procedures for a specific job, start to finish.
| Plugin | What it does |
|---|---|
Sketch ideas on a canvas and compile the map into an organized prompt. |
|
Drives a GitHub issue from branch → implement → verify → PR with a guided workflow. |
|
Discovers the aspects worth showing and captures a captioned, narrative-ordered screenshot deck of the current product. |
|
One visible channel for every long-running process — learned per-job ETAs, pre-start forecasts, and stall/orphan detection derived from each job’s own history. |
|
Backlog-drain loop: rank the open tickets, dispatch the startable ones in parallel (worktree-isolated, role-briefed, honest width), verify, merge, and re-triage on every completion. |
|
Format, static-analysis, coverage, and pre-commit skills for Maven/Java projects. |
|
Scans a codebase for OWASP-style vulnerabilities and reports them with remediation. |
Agents & research
| Plugin | What it does |
|---|---|
Fans out independent research agents across distinct angles, then synthesizes and adversarially verifies the findings. |
|
Read-only specialist subagents for code review, test running, and dependency/CVE auditing on lighter model tiers. |
Beta plugins carry a -beta suffix and live in the same marketplace — currently
tune-repo-beta and systemic-fix-beta.
See Marketplace & Beta Plugins.
Four of these are one system
roles, dev-crew, brainstorm-panel and research-sweep share a substrate of evolving personas.
One skeptic behaves the same whether it runs solo, sits on a panel, or verifies a research finding —
and accumulates what it learns along the way. The three orchestrators chain: research discovers,
the panel decides, the crew delivers.
Full model: The Role System.
Next steps
-
Getting Started — add the marketplace and install a plugin.
-
Versioning & Maintenance — how plugins are versioned, validated, and released.
The source lives at github.com/alexmond/alexmskills.