Skip to content
emaballarinPublic

About

✨ Personal plugins for Claude Code (& friends)... maybe worth sharing!

Topics

Resources

Stars

4 stars

Watchers

0 watching

Forks

Repository files navigation

ccplugins

Emanuele Ballarin's personal Claude Code plugin marketplace.

Install the marketplace

From any Claude Code session:

/plugin marketplace add emaballarin/ccplugins

Then install individual plugins with /plugin install <name>@ccplugins.

Plugins

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
dm42 dm42 — write, review and debug HP-42S-family RPN programs, the SwissMicros DM42n first. /dm42:rpn is the workflow and the verified reference, with a round-trip check for the lines the text parser drops; /dm42:setup builds dm42sim, SwissMicros' own Free42 core compiled as for the device and driven headless by action scripts that return JSON. plugins/dm42dev
hup cchupipe — the built-in "You should know" heads-ups, piped to Claude mid-turn as fallible notes it checks, with a lookup tool. By default each then leaves your screen and Claude tells you about it, leaving out only its own simply fixed slips or passing mishaps, never your work or choices, a choice of its own that changes what you get, or a decision of yours. Function hooks only. plugins/cchupipe

Install a plugin

/plugin install mf@ccplugins
/plugin install ccsci@ccplugins
/plugin install ar@ccplugins
/plugin install tml@ccplugins
/plugin install ws@ccplugins
/plugin install ccbar@ccplugins
/plugin install dm42@ccplugins
/plugin install hup@ccplugins

Layout

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
    ├── dm42dev/                             # plugin name: dm42
    │   ├── .claude-plugin/plugin.json
    │   ├── README.md  CHANGELOG.md  LICENSE  NOTICE  # MIT
    │   ├── skills/{rpn,setup}/SKILL.md
    │   ├── references/{rpn-semantics,dm42n,dm42sim}.md
    │   └── scripts/{build.sh,dm42sim.sh,roundtrip.py,smoke.py,smoke.42}
    │       └── sim/{dm42sim.cc,io.cc,io.h,core_display.patch}
    └── cchupipe/                            # plugin name: hup
        ├── .claude-plugin/plugin.json
        ├── README.md  CHANGELOG.md  tsconfig.json  # MIT (repository LICENSE)
        ├── hooks/{hooks.json,register.ts,core.ts}
        ├── types/index.d.ts
        └── tests/core.test.ts

New plugins go under plugins/<name>/ and get an entry in .claude-plugin/marketplace.json.

Validation

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, 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), and the structural Markdown validity of mf's shipped templates. 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 and hup are TypeScript run by Claude Code itself, so their checks need Claude Code and stay outside CI. Run them before a release of either (shown for ccbar; hup is plugins/cchupipe):

claude plugin validate plugins/ccbar
claude plugin test plugins/ccbar
tsc -p plugins/ccbar

License

The marketplace and the mf, ccbar and hup 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. The dm42 plugin is MIT with its own LICENSE and a NOTICE recording the third-party code its build downloads (SwissMicros' Free42 fork, GPL-2.0; the Intel decimal library, BSD-3-Clause) and the DM42PGM front-end behaviour (BSD-3-Clause) its emulator adapts.

Contact

Emanuele Ballarin — emanuele@ballarin.cc

About

✨ Personal plugins for Claude Code (& friends)... maybe worth sharing!

Topics

Resources

Stars

4 stars

Watchers

0 watching

Forks

Used by

Contributors

Languages