agentic-analyzer

agentic-analyzer

A Claude Code plugin marketplace hosting plugins that implement or support the agentic-analyzer pattern — LLM-judged, schema-validated, replayable code analyzers.

An analyzer, in this marketplace, is a Claude Code skill that walks a target repository with LLM judgment (non-deterministic) and emits a structured inventory (deterministic) of a specific kind of thing — caches, log call-sites, feature flags, session state, auth boundaries, ... — classified against a versioned ruleset, with reviewer overrides that replay across runs.

The pattern's twelve invariants are documented at docs/pattern.md. A one-page author reference lives at docs/PATTERN-CARD.md.

Plugins in this marketplace

Plugin	Role	Status
agentic-analyzer	Authoring kit. Scaffolds new analyzer skills via an interactive /new-analyzer interview (with fixture auto-gen + post-stamp quality gate) and a companion /expected-entities command for feeding dev-team oracle lists into the discovery backstop. Ships five specialist subagents and the generic _core/ runtime.	0.3.0

Pre-built analyzers (caches, logging, feature flags, session state, ...) are planned. Each new plugin lands as a plugins/<name>/ directory with its own .claude-plugin/plugin.json; the marketplace manifest is updated to list it.

Install

As a marketplace (recommended)

claude plugin marketplace add <this-repo-url>
claude plugin install agentic-analyzer@agentic-analyzer

Then, to pull new versions or newly-added plugins later:

claude plugin marketplace update agentic-analyzer

As a single plugin

If you only want the authoring kit and don't care about marketplace semantics:

claude plugin install <this-repo-url>/plugins/agentic-analyzer

Local / vendored

Clone the repo, register the plugin path in your Claude Code settings.json — full options in docs/INSTALL.md.

Prerequisite: the plugin's _core/ runtime needs Ajv:

npm --prefix plugins/agentic-analyzer/_core install
npm --prefix plugins/agentic-analyzer/_core test    # node --test, all green

Start here

1. Read docs/pattern.md — what you're buying into (twelve invariants, forty-five minutes).
2. Skim docs/PATTERN-CARD.md — the cheat-sheet to keep open while authoring.
3. Read the authoring-kit plugin's README: plugins/agentic-analyzer/README.md. The interactive /new-analyzer workflow and next steps are there.

Worked example

• examples/analyze-logging/ — a freshly-scaffolded analyzer for a PII-regulated logging domain. Shows what /agentic-analyzer:new-analyzer produces before the author fills in rules and prompts.

Layout

.
├── .claude-plugin/
│   └── marketplace.json                 catalogue of plugins
├── plugins/
│   └── agentic-analyzer/                the authoring-kit plugin
│       ├── .claude-plugin/plugin.json
│       ├── _core/                       shared runtime (bin/, templates/)
│       ├── agents/                      four specialist subagents
│       └── commands/                    /agentic-analyzer:new-analyzer
├── docs/                                pattern spec, card, install guide, iteration log
├── examples/                            scaffolder output for the logging domain
├── CHANGELOG.md  LICENSE  README.md

Status

Pre-1.0. Schemas and CLI contracts may change before 1.0. Semver after 1.0.

One load-bearing design choice to surface up front: an analyzer's overrides.json is a plain repo-local file. Concurrent edits across reviewers are the user's problem — the pattern does not attempt distributed locking. Teams that need that should stand up a small review app around the file rather than editing it by hand. See docs/pattern.md for the full rationale.

Contributing

Fork, branch, commit, PR. Run the plugin's test suite before submitting:

npm --prefix plugins/agentic-analyzer/_core test

License

MIT — see LICENSE.