Emanuele Ballarin's personal Claude Code plugin marketplace.
From any Claude Code session:
/plugin marketplace add emaballarin/ccplugins
Then install individual plugins with /plugin install <name>@ccplugins.
| Name | Description | Docs |
|---|---|---|
mf |
mindfunnel — project-agnostic session management. Skills that funnel session state into auto-memory and back (/mf:setup, /mf:prime, /mf:dump, /mf:spinup), carry the session scratchpads across a reboot (/mf:park, /mf:unpark), plus /mf:author — the reference for writing the documents an agent reads. |
plugins/mindfunnel |
ccsci |
ccscience — research & scientific-computing skills adapted from Claude Science: literature-review, bib-audit, pdf-explore, the figure-style / figure-composer / paper-narrative trilogy, paper-review, plus the computational-scientist and deep-researcher subagents. |
plugins/ccscience |
ar |
autoresearch — an autonomous experiment loop for any numeric objective. Five skills (/ar:start, /ar:resume, /ar:status, /ar:report, /ar:stop) that propose one change, measure it, and keep it only if it beats the measured noise floor. |
plugins/autoresearch |
tml |
tuneml — the scientific method for tuning and the speed↔quality frontier, in one place. Five skills (/tml:audit, /tml:plan, /tml:round, /tml:analyze, /tml:review) that read a pipeline, fix an operating point and a step budget, design experiments with scientific/nuisance/fixed hyperparameters, and return variance-aware adopt verdicts. Replaces the former parml plugin. |
plugins/tuneml |
ws |
whetstone — sharpen the thinking before the work, and the work before it lands. /ws:grill interviews a plan until nothing is silently assumed; /ws:test-gate holds every new test to one value bar; /ws:test-audit prunes and consolidates an existing suite, evidence first; /ws:deslop cleans AI slop from a diff before review; /ws:review-gate puts a change in front of a fresh-context reviewer before it is acted on. |
plugins/whetstone |
ccbar |
ccbar — a quiet status band above the prompt, drawn natively by function hooks: model and effort, repository (optionally linked) and diff counter, a context bar in the /context colours, new session tokens (or all metered ones, when set) counted once per response, and the session and weekly rate limits with reset countdowns. No skills; four hooks. |
plugins/ccbar |
/plugin install mf@ccplugins
/plugin install ccsci@ccplugins
/plugin install ar@ccplugins
/plugin install tml@ccplugins
/plugin install ws@ccplugins
/plugin install ccbar@ccplugins
ccplugins/
├── .claude-plugin/
│ └── marketplace.json
├── .github/workflows/validate.yml # CI: Tier-1 static validation
├── tests/ # pytest Tier-1 suite (+ requirements.txt)
├── README.md # this file
├── LICENSE # MIT
└── plugins/
├── mindfunnel/ # plugin name: mf
│ ├── .claude-plugin/plugin.json
│ ├── README.md CHANGELOG.md NOTICE
│ ├── skills/{setup,prime,dump,spinup,park,unpark,author}/SKILL.md
│ ├── references/{ledger,skill-mechanics}.md
│ ├── scripts/{park,unpark}.sh
│ └── templates/{AGENTS,project-AGENTS,PROJECT,SOUL,USER}.md
├── ccscience/ # plugin name: ccsci
│ ├── .claude-plugin/plugin.json
│ ├── README.md CHANGELOG.md LICENSE # Apache-2.0
│ ├── agents/{computational-scientist,deep-researcher}.md
│ └── skills/{literature-review,pdf-explore,figure-style,
│ figure-composer,paper-narrative,paper-review,
│ bib-audit}/
├── autoresearch/ # plugin name: ar
│ ├── .claude-plugin/plugin.json
│ ├── README.md CHANGELOG.md LICENSE NOTICE # MIT
│ ├── skills/{start,resume,status,report,stop}/SKILL.md
│ ├── references/{protocol,statistics,state-schema,resume-loop,completion-status}.md
│ └── templates/{ar.config.json,benchmark.sh,checks.sh,
│ evaluator.py,ar-loop.sh}
├── tuneml/ # plugin name: tml
│ ├── .claude-plugin/plugin.json
│ ├── README.md CHANGELOG.md LICENSE NOTICE # MIT (+ CC BY 4.0 material)
│ ├── skills/{audit,plan,round,analyze,review}/SKILL.md
│ ├── references/{evidence-grades,literature,optimisers,regime,
│ │ hyperparameter-roles,study-design,diagnostics,
│ │ instability,step-budget,tier-a-algorithmic,
│ │ tier-b-systems,tier-c-protocol,tier-d-architecture,
│ │ modality-map,hardware-notes,pitfalls}.md
│ └── templates/{findings,frontier,study-spec}.md + results-example.jsonl
├── whetstone/ # plugin name: ws
│ ├── .claude-plugin/plugin.json
│ ├── README.md CHANGELOG.md LICENSE NOTICE # MIT
│ ├── skills/{grill,test-gate,test-audit,deslop,review-gate}/SKILL.md
│ │ └── test-audit/references/campaign.md
│ └── references/test-value.md
└── ccbar/ # plugin name: ccbar
├── .claude-plugin/plugin.json
├── README.md CHANGELOG.md tsconfig.json # MIT (repository LICENSE)
├── hooks/{hooks.json,register.tsx,band.ts,lib.ts}
├── types/index.d.ts
└── tests/{lib,layout}.test.ts + band.test.tsx
New plugins go under plugins/<name>/ and get an entry in .claude-plugin/marketplace.json.
A dependency-light pytest suite keeps the marketplace internally consistent —
frontmatter validity, plugin.json ↔ CHANGELOG.md version parity, marketplace
integrity, README ↔ disk skill-table sync, referenced bundled-path existence, the
ccsci kernel ↔ SKILL.md entrypoint contract, and the tml grading contract
(every tier-catalogue item carries an evidence grade, every change item also
declares a quality exposure, and both ladders match the reference). Beside these
static checks, a behavioural tier runs the pure ccsci kernel helpers and mf's
park/unpark scripts (the latter skip without bash, GNU tar, sha256sum and git). It
needs nothing beyond pytest + PyYAML.
pip install -r tests/requirements.txt
python -m pytest tests/ -q
CI runs the same on every push and pull request
(.github/workflows/validate.yml), less one
local-only check: the shipped mf AGENTS.md template against the live
~/.mindfunnel/AGENTS.md, skipped where that file is absent. Details:
tests/README.md.
ccbar is TypeScript run by Claude Code itself, so its checks need Claude Code
and stay outside CI. Run them before a ccbar release:
claude plugin validate plugins/ccbar
claude plugin test plugins/ccbar
tsc -p plugins/ccbar
The marketplace and the mf and ccbar plugins are MIT — see LICENSE. The ccsci
plugin is Apache-2.0 with its own plugins/ccscience/LICENSE. The
ar plugin is MIT with its own LICENSE and a
NOTICE crediting the three MIT upstreams whose
protocol behaviour it re-implements. The tml plugin is MIT with its own
LICENSE and a
NOTICE that additionally records its adaptation of the
CC BY 4.0 Deep Learning Tuning Playbook, with the changes made. The ws
plugin is MIT with its own LICENSE and a
NOTICE crediting the two MIT upstreams its adapted skills
come from (mattpocock/skills and openclaw/openclaw), with the changes
made.
Emanuele Ballarin — emanuele@ballarin.cc