methodsmith

Methodsmith

Methodsmith is a plugin for Claude Code and Codex that brings the habits of a careful methods team to your coding agent. It plans the analysis before it runs, reviews it before anyone reads the results, and records what each study taught you, so the next one starts ahead. Clinical statistics, Python and ML papers, and meta-analysis; R or Python.

It enhances the scientist; it does not replace one. You frame the question, approve the plan and judge the findings. Methodsmith does the drafting, running and checking that make the work reproducible and grounded: pre-specified SAPs, reviewed code, reporting checklists and numbers traced to the analysis. 70 skills, 55 specialist agents, and 49 reporting standards.

# Install in Claude Code (2.1.292 or later), then set up a project
claude plugin install methodsmith --marketplace sajor2000/methodsmith

Then open your study folder and run /methodsmith:method-setup. Other agents and offline installs: see every install route or the setup guide.

Research use only; not a medical device for clinical care. Methodsmith supports research workflows and is not intended to diagnose, treat, cure, mitigate, or prevent disease or to guide patient-specific care. It is not validated for clinical use and does not replace qualified clinical, statistical, ethics, or institutional review.

Start here

If you want to	Go to
See what it does in 90 seconds	See it work
Understand the plan -> work -> review -> compound loop	You decide. Methodsmith does the legwork.
Find the skills for your study type	Built for three kinds of study
Install it	Install
See worked examples, command by command	Worked examples
Know how your data is protected	Sensitive-data boundary
Check a specific claim or limitation	FAQ and Troubleshooting
Update it safely, without losing your own skills	Keep Methodsmith updated
Install on a locked-down or offline laptop	Locked-down or demo laptop

What Methodsmith can do

Stage	What it helps you do
Frame and plan	Turn a research question into a study design, estimands, an analysis plan, power or sample-size calculations, and preregistration packages.
Build analyses	Check data and cohorts, then build reproducible R/Python analyses for clinical studies, machine learning, meta-analysis, CLIF, and omics.
Review the work	Review methods, statistics, and code for leakage, calibration, fairness, and reproducibility issues before results are reported.
Prepare results	Create tables, figures, model cards, manuscript drafts, and submission packages; trace registered numbers to SAP outputs, code commits, and source-file hashes.
Carry learning forward	Record reviewed decisions and validated methods so the next study can start from them.

You choose the research question, approve the analysis plan, resolve review findings, and sign off the conclusions. Methodsmith drafts, runs, and checks the work; it does not make the scientific decisions for you. See the worked examples and the sensitive-data boundary and its limits.

Install

Your situation	Do this
Claude Code, 2.1.292 or later	claude plugin install methodsmith --marketplace sajor2000/methodsmith
Claude Code, inside a running session (2.1.275+)	/plugin install methodsmith --marketplace sajor2000/methodsmith
Codex	git clone https://github.com/sajor2000/methodsmith.git ~/methodsmith then bash ~/methodsmith/install.sh codex
Pi	Pi, generated install
Gemini CLI, OpenCode, Kiro, Qwen Code	Also works with Codex, Pi, Gemini, and more
Locked-down or offline laptop	Locked-down or demo laptop
Source checkout, with bare /method-* aliases	Claude Code, easiest path
Windows PowerShell	Claude Code, easiest path

Installation makes Methodsmith available to your agent; then open your study folder and run /methodsmith:method-setup (in Codex, ask it to use the Methodsmith method-setup skill). Full platform details are in the setup guide and Get started below.

See it work

Watch the 90-second tour on the Methodsmith site, or play the MP4 here on GitHub: guideline routing, literature search, question framing, the analysis plan, data checks in R or Python, figures, and a first Results draft. Silent, scripted illustration on synthetic data; screens are mock-ups of the formats the skills define, not a recording of a live run.

Scripted replay on synthetic data, abridged from the output formats the skills define; it is not a recording of a live run. Source: docs/brand/scripts/demo_transcript.py.

You decide. Methodsmith does the legwork.

Every study runs the same loop: plan, work, review, compound. The plugin drafts, runs and checks; every scientific choice passes through you. Nothing is analysed until you approve the plan, and nothing is reported until you have judged the review. The loop comes from Every's Compound Engineering; Methodsmith adapts each step for research methods.

Not sure where you are? /method-workflow reads the project and recommends the next safe step.

Built for three kinds of study

Clinical statistics	Python and ML papers	Meta-analysis
/method-question PICO, FINER, hypothesis	/method-model-choice justify the model after the estimand	/method-lit-search PubMed, full text or authoritative sources
/method-power sample size and power, with a script	/method-ml-track data hashes, splits and metrics	/method-evidence-map a source-backed evidence map
/method-sap claims linked to estimands and decision evidence	/method-verify leakage, effect direction and PHI gate	/method-prereg PROSPERO, OSF or ClinicalTrials.gov
/method-cohort reproducible cohort and waterfall counts	/method-model-card calibration, fairness and limits	/method-meta-analysis random effects, REML with Hartung-Knapp
/method-table1 traceable Table 1	/method-marimo reactive, reproducible notebooks	/method-checklist PRISMA 2020 and its extensions
/method-numbers every reported number traced to the analysis	/method-figure calibration, ROC and forest plots	/method-write evidence-bound manuscript drafts

Why you can check its work

• 49 reporting standards, covering reporting guidelines, appraisal tools, and templates, encoded with their official item IDs (CONSORT, STROBE, PRISMA, TRIPOD+AI, TARGET and more). Only the newest version of each guideline can load, and a test enforces it. Sources are listed in Guideline sources.
• A PHI guard. A pre-tool hook keeps data values out of the model in schema-only projects and checks identifiers in supported web, MCP, and shell egress calls in every mode. Outputs are aggregate, with small cells suppressed. Details in Sensitive-data boundary.
• 55 specialist reviewers: methods, causal inference, multiplicity, leakage, calibration, fairness, reproducibility, PHI and more, run in parallel and deduplicated before you see them.
• Numbers locked to raw data. /method-numbers has manuscripts, tables, captions and abstracts cite numbers from a registry written by the analysis code, and a fail-closed gate traces each one to its SAP output, script commit and source hash.

How it compares

From each project's public documentation, checked 2026-10-19.

	Methodsmith	Compound Engineering	Superpowers	Scientific Agent Skills	MedSci Skills
Built for	Biostatistics, clinical and ML research	Software engineers	Software engineers	Scientists across fields	Physician-researchers
Plan -> review -> compound loop	Yes	Yes	Plans and TDD	No	Router plus defined chains; no compound step
SAP with estimands	Yes	No	No	Power only	Analysis plan, no estimands
Reporting checklists	49, item IDs, newest version only	No	No	Writing aids	49 stated (guidelines and risk-of-bias tools)
PHI guard hook	Yes	No	No	No	De-identification skill
Hosts	Claude Code, Codex, Pi, more	14 hosts	16 hosts	Many	Claude Code, Codex, Cursor, Copilot

The reporting-checklist row is not a count race: MedSci Skills states a similar number, and the difference is how each is encoded and maintained. See the competitive audit for the verified comparison and the gaps we have not closed.

Try it

Use Methodsmith to set up this project, then help me plan a study of
30-day mortality among ICU patients with sepsis. Show me the decisions I need
to make before we analyze any data.

How the package works

---

Moving from ce-datascience

Methodsmith is the same plugin as ce-datascience, renamed in 5.0. Every command and agent now starts with method- instead of ce-. A few names are also shorter, for example /ce-statistical-analysis-plan is now /method-sap.

What stays the same: your own skills and the plan -> work -> review -> compound loop.

Move each project's state folder: the per-project folder is now .methodsmith/. In every project that has a .ce-datascience/ folder, run mv .ce-datascience .methodsmith (/method-update prints this command for the current project). The folder holds the PHI-boundary settings in config.local.yaml, so until it is moved the boundary guard treats the project as schema-only and says why. It never takes the mode from the old folder; it reads that folder only for settings that tighten the boundary (data root, identifier index, run outputs).

What the switch does: the first Methodsmith install moves the old plugin's files to a backups/ folder next to the new install. Nothing is deleted, and files you created are not touched.

To see what is installed on your machine, run /method-update after installing Methodsmith, or run the same read-only checker from a checkout. It prints exactly the commands below for each harness it finds:

bash plugins/methodsmith/skills/method-update/scripts/detect-pre-rename-install.sh

Harness	Switch to Methodsmith
Claude Code	claude plugin uninstall ce-datascience@ce-datascience-plugin, then claude plugin marketplace remove ce-datascience-plugin, claude plugin marketplace add sajor2000/methodsmith, claude plugin install methodsmith@methodsmith. If you used --aliases, re-run bash install.sh claude --aliases.
Codex	codex plugin marketplace remove ce-datascience-plugin, then codex plugin marketplace add sajor2000/methodsmith and bunx github:sajor2000/methodsmith install methodsmith --to codex. Open /plugins and install Methodsmith.
OpenCode	bunx github:sajor2000/methodsmith install methodsmith --to opencode
Pi	bunx github:sajor2000/methodsmith install methodsmith --to pi --pi-home ~/.pi/agent
Gemini CLI, Kiro	bunx github:sajor2000/methodsmith install methodsmith --to gemini --output . (or --to kiro) in each workspace
Qwen Code	qwen extensions uninstall ce-datascience, then qwen extensions install sajor2000/methodsmith:methodsmith

The old /ce-* commands no longer work; use the new names. The deprecated /ce-* aliases that 5.x wrote with --aliases are removed the next time you run bash install.sh claude --aliases.

Before 5.0	Methodsmith
/ce-adversarial-review	/method-adversarial-review
/ce-bioinfo-qc	/method-omics-qc
/ce-brainstorm	/method-brainstorm
/ce-checklist-match	/method-checklist
/ce-clean-gone-branches	/method-clean-branches
/ce-clif	/method-clif
/ce-clif-audit	/method-clif-audit
/ce-clif-meta-analysis	/method-clif-meta-analysis
/ce-clif-project-template	/method-clif-template
/ce-clifpy	/method-clifpy
/ce-clinical-research-voice	/method-voice
/ce-code-review	/method-review
/ce-cohort-build	/method-cohort
/ce-commit	/method-commit
/ce-commit-push-pr	/method-pr
/ce-compound	/method-compound
/ce-compound-refresh	/method-compound-refresh
/ce-data-qa	/method-data-qa
/ce-debug	/method-debug
/ce-doc-review	/method-doc-review
/ce-effect-size	/method-meta-analysis
/ce-evidence-map	/method-evidence-map
/ce-fabric	/method-fabric
/ce-figure	/method-figure
/ce-genome-build	/method-genome-build
/ce-grill	/method-grill
/ce-handoff	/method-handoff
/ce-ideate	/method-ideate
/ce-journal-select	/method-journal-select
/ce-language-detect	/method-language-detect
/ce-literature-search	/method-lit-search
/ce-manuscript-citations	/method-citations
/ce-manuscript-package	/method-submission-package
/ce-manuscript-section-discipline	/method-section-check
/ce-marimo	/method-marimo
/ce-method-extract	/method-extract
/ce-ml-experiment-track	/method-ml-track
/ce-model-card	/method-model-card
/ce-model-strategy	/method-model-choice
/ce-notebook-edit	/method-notebook-edit
/ce-notebook-standards	/method-notebook-standards
/ce-numbers	/method-numbers
/ce-optimize	/method-optimize
/ce-phenotype-validate	/method-phenotype
/ce-plan	/method-plan
/ce-power	/method-power
/ce-pre-submission-audit	/method-presubmit
/ce-prereg	/method-prereg
/ce-pubmed	/method-pubmed
/ce-release-notes	/method-release-notes
/ce-report-bug	/method-report-bug
/ce-research-question	/method-question
/ce-resolve-pr-feedback	/method-resolve-feedback
/ce-review-pack	/method-signoff
/ce-rstats	/method-r
/ce-sap-tabular	/method-sap-workbook
/ce-sas-stata-assess	/method-sas-stata
/ce-scientific-anti-slop	/method-anti-slop
/ce-scientific-writing	/method-write
/ce-sessions	/method-sessions
/ce-setup	/method-setup
/ce-sprint	/method-sprint
/ce-statistical-analysis-plan	/method-sap
/ce-table1	/method-table1
/ce-trial-design	/method-trial-design
/ce-update	/method-update
/ce-verify	/method-verify
/ce-work	/method-work
/ce-workflow	/method-workflow
/ce-worktree	/method-worktree

Agents keep the rest of their name: ce-methods-reviewer is now method-methods-reviewer.

Get started

The normal path is intentionally short: install the plugin, start your agent, then run setup. For complete platform details, see the public setup guide.

Choose your install route

From a source checkout, run bash install.sh doctor to see which local tools are available and the exact standard or locked-down route. Use the standard route on a personal or managed laptop that can run Git and the agent CLI. On a corporate or locked-down laptop, ask IT for the approved offline artifact and use the artifact-specific path below; Bun, Git, GitHub CLI, and Quarto are not required for basic use.

Claude Code, easiest path

No checkout needed (terminal command, Claude Code 2.1.292 or later):

claude plugin install methodsmith --marketplace sajor2000/methodsmith
claude

Or, inside a running Claude Code session (2.1.275 or later):

/plugin install methodsmith --marketplace sajor2000/methodsmith

No terminal at all: the in-session install works on any Claude Code surface (desktop app or terminal). Paste the /plugin install ... line above in the prompt, run /methodsmith:method-setup, and enable auto-update under /plugin -> Marketplaces. Details: docs/setup.md.

From a source checkout, with the optional bare /method-* aliases, on macOS, Linux, WSL, or Git Bash:

git clone https://github.com/sajor2000/methodsmith.git ~/methodsmith
cd ~/methodsmith
bash install.sh claude --aliases
claude

Windows PowerShell:

git clone https://github.com/sajor2000/methodsmith.git "$HOME\methodsmith"
cd "$HOME\methodsmith"
.\install.ps1 claude -Aliases
claude

Then open the target project and run the native command:

/methodsmith:method-setup

The --aliases flag is optional; it installs safe local aliases so the demo-friendly bare /method-* commands work. The namespaced command above is the reliable default for every native Claude plugin install.

Claude Code via GitHub CLI (skills only)

GitHub CLI 2.90+ (verified with 2.100.0) can install the skills without the plugin machinery. Install every skill, pinned to the latest release tag, or one skill by name (add --pin methodsmith-v5.1.1 to hold a fixed version):

gh skill install sajor2000/methodsmith --all --agent claude-code --scope user
gh skill install sajor2000/methodsmith method-citations --agent claude-code --scope user

This route copies skills only — no reviewer agents, no commands, no hooks, and no auto-update (update by re-running the command after a new release) — writes them flat into your skills folder, and --all also picks up the sample skills from the repository's tests/fixtures/; prefer the single-skill form, or delete the extras afterwards. See the setup guide for details.

Codex, easiest path

macOS, Linux, WSL, or Git Bash:

git clone https://github.com/sajor2000/methodsmith.git ~/methodsmith
cd ~/methodsmith
bash install.sh codex
codex

Windows PowerShell:

git clone https://github.com/sajor2000/methodsmith.git "$HOME\methodsmith"
cd "$HOME\methodsmith"
.\install.ps1 codex
codex

Inside Codex, open /plugins, install Methodsmith, restart Codex, then start a new thread and ask Codex to use Methodsmith for setup. The helper also installs the generated agent bridge when Bun is available.

Pi, generated install

Install Pi's two workflow extensions once, then generate the Methodsmith files into the Pi agent root:

pi install npm:pi-subagents
pi install npm:pi-ask-user
cd ~/methodsmith
bun run src/index.ts install ./plugins/methodsmith --to pi --pi-home "$HOME/.pi/agent"

Restart Pi, open the project, then invoke method-setup with Pi's normal skill interface. pi-subagents is required for Methodsmith review/workflow delegation; pi-ask-user enables the short setup confirmation UI.

Use the plugin after installation

Installation makes Methodsmith available to the agent; setup configures it for each individual code or research project. After installing:

1. Restart Claude Code, Codex, or Pi so it loads the new plugin files.
2. Open the project or study directory you want Methodsmith to work on.
3. Configure that project with the setup skill.
4. Run the workflow skill to see the ordered research or engineering lifecycle and the next recommended skill.

In Claude Code, native plugin commands are namespaced:

/methodsmith:method-setup
/methodsmith:method-workflow

When --aliases or -Aliases was used, the shorter /method-setup and /method-workflow forms work too. In Codex, start a new task and say:

Use the Methodsmith method-setup skill for this project.
Then use the Methodsmith method-workflow skill and recommend the next step.

After that, request a specific skill in plain language or use its displayed name, for example method-question, method-data-qa, method-plan, or method-review. The plugin works in the currently open project; repeat setup when moving to a different project with a different stack or data layer.

Setup starts with one detected-profile summary instead of a full questionnaire. It shows inferred language, environment, reporting format, storage, and data domain with evidence, then offers Continue, Adjust, or Full survey. It asks only for values that are ambiguous or needed by the current workflow.

Locked-down or demo laptop

Use the offline release artifacts when package managers, GitHub CLI, Git, Bun, or Quarto are blocked by IT policy. Basic Claude Code and Codex use does not require Bun, GitHub CLI, or Quarto.

Claude Code can load the approved plugin folder or ZIP directly:

claude --plugin-dir /approved/path/methodsmith
claude --plugin-dir /approved/path/methodsmith.zip

Native Claude plugin skills are namespaced:

/methodsmith:method-setup --locked-down
/methodsmith:method-workflow

Bare /method-* commands are an optional local alias layer, not a native plugin guarantee. Install them only when you want demo-friendly command names:

bash scripts/install/install-claude-aliases.sh --plugin-dir /approved/path/methodsmith --scope user

For Codex without Bun, unpack methodsmith-codex-local.zip and run:

bash install-codex-offline.sh --source /approved/path/methodsmith-codex-local --codex-home "${CODEX_HOME:-$HOME/.codex}"

On Windows PowerShell, use the package's native installer:

.\install.ps1 codex -Source C:\approved\methodsmith-codex-local -CodexHome "$HOME\.codex"

Restart Codex, open /plugins, install Methodsmith from the local marketplace, then restart again. The installer writes the personal marketplace file under .agents/plugins/marketplace.json and points it at ./.codex/plugins/methodsmith relative to that marketplace root.

Recommended research add-ons

Methodsmith's bundled /method-pubmed workflow works on its own. For agent-native PubMed, Europe PMC, MeSH, citation, and related-article tools, optionally add cyanheads/pubmed-mcp-server. For full-text synthesis, claim verification, figures, trials, regulatory documents, preprints, and biological databases, optionally add the Paperclip CLI and its official Paperclip skill, or its MCP server.

Start with PubMed MCP for lightweight biomedical discovery; add Paperclip when research planning needs deeper full-text or cross-source evidence. Neither is required or installed automatically. Review institutional privacy and network policy before connecting a hosted research service. See the optional research add-ons setup. The Paperclip skill is fetched and maintained by Paperclip, not bundled with Methodsmith. Methodsmith workflows currently auto-detect the Paperclip CLI only; external MCP or provider-skill results are direct agent capabilities, not automatic Methodsmith artifact handoffs.

Source checkout for contributors

Bun and Git are contributor/release tooling. Install them only when you need to build, validate, or convert the plugin from source:

curl -fsSL https://bun.sh/install | bash
export CE_DS_REPO="$HOME/methodsmith"
git clone https://github.com/sajor2000/methodsmith.git "$CE_DS_REPO"
cd "$CE_DS_REPO"
bun install

Launch Claude Code from the checkout:

claude --plugin-dir "$CE_DS_REPO/plugins/methodsmith"

Pro tip — save yourself typing forever:

printf "\nalias claude-ds='claude --plugin-dir %s/plugins/methodsmith'\n" "$CE_DS_REPO" >> ~/.zshrc
source ~/.zshrc

Now just type claude-ds in any project.

Configure your stack

/methodsmith:method-setup

Picks up your language (R or Python), IDE, libraries, and data layer automatically. It checks project evidence first and only asks questions that change generated config or workflow routing. If you installed optional aliases, /method-setup works too.

Sensitive-data boundary

Before setup looks at any data, method-setup asks one question and saves the answer as phi_environment.mode in the gitignored, machine-local .methodsmith/config.local.yaml:

Answer	Saved as	What happens
Approved endpoint: full access (HIPAA-compliant with a BAA, or institution-approved)	mode: approved	No restrictions on the model for the rest of the session. The agent never raises the boundary again. You are responsible for that approval.
Not approved or unsure: schema-only (also used when no answer is saved, or the saved value cannot be read)	mode: schema-only	The agent sees column headings, types, row counts, and % missing, never data values such as names, MRNs, or claim numbers. It writes scripts that print only aggregates, with counts under 11 suppressed. It cannot inspect example rows, debug value-level parsing or join mismatches by example, read free text, or spot individual entry errors.
No sensitive data: public or synthetic	mode: no-sensitive-data	No restrictions, never raised again.

Older configs keep working: compliant: true reads as approved, compliant: false as schema-only, and sensitive_data: false as no-sensitive-data. If the keys disagree, any schema-only signal wins.

What enforces it. In Claude Code, a plugin PreToolUse hook (hooks/data-boundary-guard.py) runs only inside projects that have a .methodsmith/ directory, in the working tree or the main checkout of a linked worktree. It is silent unless it blocks a call. In schema-only mode it denies:

• Read, NotebookRead, and content-mode Grep on data files. Data files are .csv, .tsv, .parquet, .xlsx, .sas7bdat, .dta, .rds, .sqlite, .jsonl, and similar, including compressed or backup copies (.csv.gz, .csv.bak), plus anything under the configured data_root and the safe runner's logs. A notebook with saved outputs also counts; clear its outputs first.
• Any shell command (Bash, Monitor, PowerShell) whose text names a data path, an extension glob, or data_root, whatever the program, unless it only lists metadata (ls, wc, du, find, git status/log) or runs the schema helper. Inline code (python -c, Rscript -e, heredocs) that opens data or globs a data directory, writing a script and running it in the same command, copying a data file to a non-data name, and recursive grep/rg without an --include/-g/-t limited to code are denied too.
• Running a script file whose code names a data path or data_root, or calls a reader (read_csv, read_parquet, read_excel, read_sas, read_stata, fread, readRDS, arrow::, DuckDB, SQLite) on a path held in a variable, unless the command runs through the safe runner. This covers every common way to start it: python, Rscript, bash/sh, quarto render, and jupyter nbconvert --execute; run wrappers such as uv run and pdm run; a direct path with a script extension or interpreter shebang; a script fed on stdin; source; python -m for a project module; Rscript -e "source(...)"; and code inside sh -c, bash -c, or eval. A script that runs another file the hook cannot resolve before it runs (source(f), eval(parse(f)), source passed to lapply, runpy, exec/eval of a computed string, importlib.import_module(name)) is denied too; literal paths, including file.path(...) and here(...) over literals, are followed and checked. The hook reads the script itself, up to 1 MB. Scripts that do not touch data, or that only read aggregate tables that pass the check below, run as before.
• Workflow runners (make, snakemake, nextflow, dvc repro, pytest, tox, nox, just, papermill, dbt run, package scripts such as npm run, targets::tar_make(), rmarkdown::render(), and similar), unless they run through the safe runner, because the hook cannot see what their targets read. find -exec and xargs over files the hook cannot rule out as data are denied too.

A command, Read, or Grep that reaches into a project from outside it (a relative path, cd, git -C, make -C, a symlink, or a path inside inline code) gets that project's boundary, and when several projects are named the strictest wins. Unquoted shell wildcards are expanded (up to 256 matches) and checked like the files they name; in schema-only mode a wildcard that matches more than that is treated as data.

Database MCP tools (Postgres, DuckDB, SQLite, Snowflake, BigQuery, MongoDB, Neo4j, and similar, recognized by a database word in the tool name or by SQL or Cypher statement text in the arguments) return query results straight into the context, so in schema-only mode they are denied unless the call only inspects the schema: a tool that lists or describes tables, or a single DESCRIBE, SHOW TABLES/COLUMNS, PRAGMA table_info, or plain select from information_schema.tables/columns. Run real queries in a script through the safe runner, write an aggregate table, and read it with method-aggregate-read. Literature and search connectors are not affected.

Safe runner. scripts/method-safe-run -- <command> runs the command and writes its full stdout and stderr to .methodsmith/runs/<UTC-timestamp>-<pid>.log (gitignored). It prints only the exit code, the duration, the files created or modified under the project (path, size, and row count for tabular files, including in projects too large to snapshot and files copied with their old timestamps), and a stderr summary that keeps only traceback file and line frames and exception type names, plus a count of withheld lines; identifier values in file names are masked. It never prints stdout or other stderr text: open the log yourself. It refuses to run when no .methodsmith/ or .git is found above the working directory. The command gets its own temp directory (TMPDIR, TMP, TEMP) outside the project, which counts as data and is removed when the run finishes. While a run, or anything it started, is still going, project files changed since it began count as data; a run that was killed before recording its outputs keeps that rule in force until the next run or method-safe-run --reconcile records what it wrote. Processes a command leaves running are stopped about 10 seconds after it exits. The hook's deny message gives the runner's full path. It accepts only the plugin's own runner (or a byte-identical copy) and still checks uploads inside it.

Aggregate reads. Read, cat, head, and tail of .csv, .tsv, and .md files under the aggregate directories are allowed in schema-only mode when the table passes the aggregate check: no identifier-like column headings, no cell in the identifier index or shaped like an identifier, no integer count from 1 to 10 (the small-cell rule; 0 is fine), and at most 500 rows. Otherwise the deny reason names the rule that failed, never the value. The directories default to output, results, analysis/publication, and analysis/sap-tables; set phi_environment.aggregate_dirs to change them. data_root always wins: nothing under it is readable.

Locale packs. Projects outside the US add national identifier shapes with locale: uk (or another pack) under phi_environment in config.local.yaml. A pack is data — hooks/locales/<locale>.json holds identifier regexes, optional named checksum validators (a match counts as an identifier only when its checksum validates, so checksum-invalid synthetic test values pass), and column-name patterns — so new countries need no guard-code changes. An absent key changes nothing; a malformed or unknown pack fails closed. The schema helper masks locale-flagged headings the same way and refuses to print on a malformed or unknown pack.

The schema helper scripts/schema-peek.py prints headings, types, counts, and % missing, never a cell value. It prints col_1..col_n when a file has no header row, and col_<k> [masked header] for a heading that is itself an identifier (an identifier shape, or a value in the identifier index).

What external tools receive

PubMed, Paperclip, and every other MCP tool receive search terms, MeSH terms, PMIDs, DOIs, and paper titles. They never receive data values, identifiers, row text, or data files. This holds in every mode, including after you approve the endpoint: approving the model endpoint does not cover third-party services such as NCBI or Paperclip.

Claude Code WebSearch queries, WebFetch URLs and prompts, MCP tool arguments, and detected shell uploads are checked for common identifier patterns and values in the project identifier index. Research search terms, MeSH terms, PMIDs, DOIs, and paper titles pass when they do not match an identifier.

The same hook enforces this for every mcp__* call, for shell uploads, and for shell calls to NCBI E-utilities (curl or wget to eutils.ncbi.nlm.nih.gov, and the method-pubmed search script). It denies arguments that contain an SSN, an email address, a US phone number, a keyword-anchored record number (MRN, medical record, patient ID, member ID, claim number), a keyword-anchored date of birth, or a value in the identifier index. It also denies Paperclip upload, import, or share calls that name a local data file, and curl, wget, scp, or rsync uploads of data files. PMIDs, DOIs, PMCIDs, author surnames, and ordinary research queries pass silently. The skills carry the same rule.

Identifier index. Whenever data_root is set, in every mode, method-setup builds and method-data-qa refreshes the identifier index with scripts/method-identifier-index. It reads identifier-like columns under data_root (names; MRN, patient, member, subscriber, and record IDs; claim numbers; SSN; date of birth; address; phone; email) and stores only salted hashes in .methodsmith/identifier-index.local.json: local, gitignored, with a random per-project salt and no plaintext. It prints counts only. Each full value is stored, plus first+last name pairs, plus single tokens that contain a digit for ID columns only, so a lone surname or common word never matches. The hook hashes every one- to three-word run of each argument, denies on a match without echoing it, and is silent when no index exists. With a 200,000-entry index the lookup adds a few milliseconds.

Limits. The hook is defense in depth, not a sandbox. Enforcement is Claude Code only: Codex and OpenCode installs carry the hook entry but it exits immediately there, and Gemini, Kiro, Copilot, and Pi skip hooks, so those harnesses rely on the skill instructions alone. The hook requires python3. If it is missing, the launcher prints a warning and exits successfully, so the guard is inactive and calls are not blocked. The script check is heuristic static analysis: a script that builds a data path at run time, or reads data through a function the hook does not recognize, can still print values, and programs it cannot parse (deliberate obfuscation) are out of scope. Files a safe-runner command writes to hard-coded paths outside the project and its temp directory (/tmp/x.csv, Java's java.io.tmpdir, Spark's local directory), and processes that leave the run's process group with setsid, are not tracked. A database server whose tools have no database word in their names and take no SQL text is not recognized, and MCP resources are not checked. Editing a file while a safe-runner command is still running makes that file data until the run ends. Values that are not in identifier columns, such as free-text notes, are not indexed, and the identifier patterns catch common shapes, not every one. A patient who shares a full name with an author you search for blocks that name. A config problem resolves to schema-only. An unexpected internal error blocks only calls that mention a data file or an upload; anything else is allowed with a one-line warning, so a hook bug never blocks unrelated work.

Changing the answer. Re-run method-setup, or edit or delete the phi_environment block in .methodsmith/config.local.yaml.

See your workflow

/methodsmith:method-workflow

Shows every step for your project type and tells you what to do next. Each slash skill now starts with a Skill Value block that states the problem, expected output, question boundary, and non-goal so new users can pick the right command without guessing.

---

Worked examples

The examples below use bare /method-* commands for readability. In native Claude plugin installs, use /methodsmith:method-* unless local aliases are installed.

Run an observational study

/method-grill "sepsis bundles and 30-day mortality in ICU, target Critical Care Medicine"
/method-question "sepsis bundles and 30-day mortality in ICU"
/method-pubmed
/method-evidence-map
/method-extract
/method-checklist
/method-meta-analysis
/method-power
/method-cohort
/method-data-qa
/method-model-choice
/method-plan
/method-sap-workbook
/method-sprint
/method-work
/method-verify
/method-review
/method-compound

Keep analysis decisions explicit

The lifecycle now makes the most consequential analytical assumptions visible before code is written or trusted:

• /method-data-qa reconciles row counts and joins, checks keys and type stability, records missing-data handling, and returns GO, WARN, or NO-GO rather than silently substituting fallback data.
• /method-model-choice resolves nontrivial model, estimand, dependence, and diagnostic choices after QA and before SAP finalization.
• /method-plan owns the canonical versioned SAP. Add /method-sap only when a complex clinical or observational design needs a claim-to-dataset schema, explicit estimands, and decision evidence; use /method-sap-workbook for the workbook rather than creating a competing plan.
• /method-work fails loudly at missing or corrupt inputs. /method-review checks integrity and causal timing, and requires censoring-aware, decision-aligned time-dependent AUC for dynamic survival models.

For an administrative claims study, use the claims-oriented SAP workflow when the canonical SAP needs linked claims, datasets, variables, analyses, diagnostics, outputs, and decision evidence:

/method-sap

Statistical correctness

The statistical methods behind the skills and reviewers were audited against primary sources. What that means in practice:

• Estimands first. SAPs from /method-plan and /method-sap state ICH E9(R1) estimands (population, treatment, endpoint, intercurrent events, summary measure), and method-sap-drift-detector flags estimand, missing-data, and inference drift in code.
• Sample size and pooling. /method-power covers two-proportion, non-inferiority (one-sided 0.025), survival, cluster, and Riley-criteria pmsampsize designs. /method-meta-analysis pools with REML, a Hartung-Knapp CI, and a k-1 df prediction interval. /method-trial-design computes type I error under non-binding futility without the futility bound.
• Tables and registration. /method-table1 --design rct|observational|descriptive adds an SMD column for observational comparisons and never shows baseline p-values for randomized trials. /method-prereg blocks a draft SAP, unblinded data, and late PROSPERO registration.
• Reporting routes. All prediction models route to TRIPOD+AI (it supersedes TRIPOD 2015), target trial emulation to the standalone 21-item TARGET checklist, and protocols to SPIRIT 2025 or PRISMA-P.
• Reviewers. Calibration, causal-inference, multiplicity, fairness, data-leakage, and methods reviewers follow current guidance (for example, the calibration hierarchy, M-bias, co-primary endpoints needing no alpha correction, and random splits not being leakage by themselves).
• Data and validation. /method-phenotype reweights sensitivity and specificity for verification bias under algorithm-stratified chart review; /method-data-qa drops mean imputation and LOCF advice; /method-pubmed keeps unindexed and non-English records by default.

Prepare a manuscript for handoff

/method-journal-select
/method-submission-package
/method-citations
/method-write
/method-presubmit
/method-signoff

Pick the target journal with /method-journal-select before packaging: its per-journal next-step list (article type, word limits, required checklist) sets what the package must satisfy.

Use /method-voice, /method-section-check, or /method-anti-slop directly only for a focused prose pass. The combined pre-submission audit is the final editing gate; required Methods rationale is preserved, while retrospective interpretation remains in Discussion.

Scientific input and disclosure contracts

Power-effect pooling requires a selected compatible estimand, outcome/horizon, adjustment set, scale/unit and independent contributions. Its contract= artifact travels with the pooled signal; legacy tables/signals need reconciliation before automatic power use. Formal study/performance synthesis uses method-clif-meta-analysis, not the power-anchor parser. Verification separates inferential MI from prediction validation/deployment and event targets from recruitment N. Equivalence alpha follows the study framework rather than a universal NI convention. Phenotype validation rejects duplicate patient keys. Tracking preserves the chosen validation design and keeps full sensitive logs restricted; only reviewed, disclosure-safe aggregate summaries enter Git.

Numbers locked to raw data

Clinical journals, reviewers, and data-sharing audits ask where each number in a paper came from. /method-numbers makes that answer checkable: a rigor signal, not a style rule. Analysis code registers aggregates in analysis/numbers/registry.json (R or Python; counts from 1 to 10 are stored suppressed). The manuscript cites them as {{num:<id>}}. A gate fails the submission when a number can't be traced. Each number follows one chain: SAP output -> script@commit -> source file hash -> registry entry -> sentence.

Worked example (synthetic):

# analysis/primary_cox.R, committed before it runs
source("analysis/numbers/method_numbers.R")
register_number("primary.hr", 0.781, kind = "estimate", ci = c(0.664, 0.918),
                sources = "data/analytic.parquet", sap_output = "A3_cox_primary.csv")

manuscript/results.qmd:  The hazard ratio was {{num:primary.hr}}.
rendered:                The hazard ratio was 0.78 (95% CI, 0.66-0.92).
gate on a typed number:  manuscript/abstract.md:7: untraced number: 0.81
gate after a data pull:  registry:primary.hr: stale: source data/analytic.parquet changed

The gate also fails on unknown ids, unsuppressed small cells, entries without a SAP output or committed script, and when it found nothing to check. It allows years, citation markers, figure, table, and item numbers, dates, quoted guideline text, and the doses or units you declare. Table 1, figure captions, model cards, and scientific writing cite the registry; method-verify, method-presubmit, and method-submission-package run the gate and fail closed. Word manuscripts keep each number as a registry-owned field (ADDIN CE_NUM <id>, beside Zotero's citation fields); the gate fails a field whose text was hand-edited, and a number typed into a Word table. Rendered text copies keep an adjacent .numbers.json provenance map; typed matching numbers and quoted results fail the gate. Word headers, footers and notes are checked alongside the body.

Numeric-prefix compounds such as 7-patient and 2.4-fold require registry provenance. Fixed design phrases such as 28-day need an explicit declaration; hyphenation is not an exemption. It does not read numbers inside images.

Build a prediction model

/method-question
/method-checklist
/method-cohort
/method-plan
/method-ml-track
/method-optimize
/method-work
/method-model-card
/method-review

Work with CLIF consortium data

# Anchored to clif-icu.com; setup can infer a declared matching family.
# Use an explicit call when choosing the family for this task:
/method-clif --version 2.1.0  # CLIF 2.1 + mCIDE 2.1
# or
/method-clif --version 3.0.0  # CLIF 3.0 + mCIDE 3.0
/method-clif-meta-analysis  # Site-to-lead or literature synthesis in R/Python
/method-workflow
/method-work

If a direct call conflicts with an explicitly declared project pair, Methodsmith asks which source is intended before it validates categories or generates filters.

Analyze omics data

/method-omics-qc
/method-genome-build
/method-plan
/method-work
/method-review

Ship a software fix

/method-brainstorm
/method-plan
/method-work
/method-review

Maintain the plugin professionally

/method-sessions
/method-resolve-feedback
/method-release-notes
/method-report-bug

These workflow utilities are adapted from the original compound-engineering plugin so the public plugin can support release questions, bug reports, cross-session debugging, and statistical-methodology PR feedback without pulling in unrelated Rails, frontend, Xcode, Slack, product-pulse, dogfood, LFG, or agent-native workflows.

Repository maintenance is reported weekly through Weekly Branch Maintenance. It identifies remote candidates only; inspect local worktrees and confirm any deletion with /method-clean-branches.

---

Works with your stack

Stack	IDE	Output	Libraries
Python + Jupyter	JupyterLab, VS Code	.ipynb	pandas, polars, scipy, statsmodels, scikit-learn
Python + Marimo	Marimo, VS Code	reactive .py	pandas, polars, scipy, statsmodels, scikit-learn
R	RStudio, VS Code	Quarto .qmd, .Rmd	tidyverse, data.table, survival, lme4, gt, tidymodels

Knows your data layer

Data layer	How it activates	What it does
OMOP CDM	SQL with cdm_source, concept, person	OMOP SQL + concept sets, vocabulary pinning
CLIF	CLIF_CLAUDE.md, clif-icu remote, or CLIF handoff	Anchors to clif-icu.com; explicit CLIF/mCIDE 2.1 or 3.0 family, Parquet-only, version-correct mCIDE vocab, POC sign-off
Admin claims	Medicare/Medicaid/MarketScan in code	Enrollment gaps, NDC-to-RxNorm, claims reviewer
Custom EHR	Default	PHI scanning, generic cohort building
Bioinformatics	.fastq, .bam, Snakefile	FastQC/MultiQC, genome build, batch-effect screen

This fork tracks useful infrastructure and workflow improvements from the original compound-engineering plugin while keeping the product scoped to health data science.

Area	What changed for data scientists
Planning and brainstorming	/method-plan keeps SAP/implementation dual mode and now starts summaries with an implementation-independent objective, supports upstream output modes, preserves format on resume, and strengthens synthesis and research routing. /method-brainstorm keeps study-design framing while adding grouped requirements, output-mode handling, and visual communication behavior.
Safe execution and optimization	/method-work verifies that parallel workers can start from a complete tracked snapshot and keeps uncommitted dependencies inline. /method-optimize gives parallel Codex experiments separate directories and requires measurable baselines and outcomes.
Debugging	/method-debug covers correctness and performance investigations, records a baseline before tuning, redacts secrets from captured evidence, and repeats the original measurement after a fix.
PR and review workflow	/method-resolve-feedback carries upstream GraphQL pagination and split-reference handling, then restores statistical methodology and SAP-aware response language. /method-review applies path-scoped CODING_STANDARDS.md criteria without replacing binding repo instructions. /method-commit and /method-pr use a temporary Git index so named-file commits do not absorb unrelated staged or later working-tree changes.
Setup health	/method-setup detects obsolete Codex tool-map blocks and offers bounded cleanup guidance instead of silently rewriting user configuration.
Session history	/method-sessions uses upstream cross-platform discovery improvements for Claude Code, Codex, and Cursor sessions, with repo-root pre-resolution and structured extraction scripts.
Distribution and installation	The converter supports current method-*.md agent source files while still parsing legacy *.agent.md, respects CODEX_HOME, writes Codex roots correctly, and manages .codex/hooks.json without clobbering manual hooks.
Public support	/method-release-notes and /method-report-bug are included as curated support skills for a professional public plugin surface.

Core-only Compound Engineering skills such as Proof review, demo-reel capture, frontend/Rails/iOS helpers, LFG, simplify-code, strategy, promote, and polish remain external. When the core Compound Engineering plugin is also installed, methodsmith handoffs can use those skills with a visible fallback; they are not packaged as methodsmith slash skills.

Reviews against 49 reporting checklists

The checklist registry records publication evidence, current versus legacy versions, and methodological roles. Reporting guidelines, appraisal tools, planning templates, and unresolved candidates are not treated as interchangeable routes; unresolved entries remain explicitly labeled rather than being presented as authoritative.

Guideline sources lists each checklist's primary citation, PMID, DOI, verification status, and superseded versions. It is generated from the registry.

Study type	Primary	Extensions
Observational cohort	STROBE (STROBE-MR stand-alone for Mendelian randomization)	RECORD, RECORD-PE, STREGA
Randomized trial	CONSORT (SPIRIT 2025 for protocols)	CONSORT-AI, SPIRIT-AI, Cluster, Adaptive, N-of-1, Outcomes 2022 (report and protocol)
Prediction model	TRIPOD+AI (supersedes TRIPOD 2015)	TRIPOD-LLM, MI-CLAIM
Diagnostic accuracy	STARD (STARD-AI stand-alone for AI index tests)	CLAIM
Systematic review	PRISMA (PRISMA-P for protocols; stand-alone DTA, NMA, IPD, ScR by review type)	QUADAS-3 (DTA appraisal; supersedes QUADAS-2), CHARMS (prediction-model reviews)
Target trial emulation	TARGET (standalone)
Early live clinical evaluation of AI	DECIDE-AI
Other	SQUIRE, GRAMMS, STaRT-RWE, ARRIVE, CARE, CHART, CHEERS, COREQ, DEAL (LLM), PDSQI-9 (LLM summaries), REFORMS

---

Also works with Codex, Pi, Gemini, and more

Run generated installs from the repo root (cd "$CE_DS_REPO"). Use ./plugins/methodsmith with the leading ./ for local installs.

Platform	Easiest command
Claude Code, no checkout	claude plugin install methodsmith --marketplace sajor2000/methodsmith then /methodsmith:method-setup
Claude Code via GitHub CLI 2.90+	gh skill install sajor2000/methodsmith --all --agent claude-code --scope user (skills only; caveats in the setup guide)
Claude Code, one-command setup	bash install.sh claude --aliases then /method-setup
Claude Code, Windows PowerShell	.\install.ps1 claude -Aliases then /method-setup
Claude Code	claude --plugin-dir "$CE_DS_REPO/plugins/methodsmith" then /methodsmith:method-setup
Claude Code, offline	claude --plugin-dir /approved/path/methodsmith.zip
Claude bare aliases	bash scripts/install/install-claude-aliases.sh --plugin-dir "$CE_DS_REPO/plugins/methodsmith" --scope user
Codex, one-command setup	bash install.sh codex then install Methodsmith from /plugins
Codex, Windows PowerShell	.\install.ps1 codex then install Methodsmith from /plugins
Codex native + agent bridge	bun run src/index.ts install ./plugins/methodsmith --to codex --codex-home "$CODEX_HOME"
Codex offline local marketplace	bash install-codex-offline.sh --source /approved/path/methodsmith-codex-local --codex-home "$CODEX_HOME"
Codex standalone	bun run src/index.ts install ./plugins/methodsmith --to codex --codex-home "$CODEX_HOME" --include-skills
Pi	bun run src/index.ts install ./plugins/methodsmith --to pi --pi-home "$HOME/.pi/agent"
Gemini CLI	bun run src/index.ts install ./plugins/methodsmith --to gemini --output /path/to/gemini-workspace
OpenCode	bun run src/index.ts install ./plugins/methodsmith --to opencode --output /path/to/workspace
Kiro	bun run src/index.ts install ./plugins/methodsmith --to kiro --output /path/to/kiro-workspace
Qwen Code	qwen extensions install sajor2000/methodsmith:methodsmith
All generated targets	bun run src/index.ts install ./plugins/methodsmith --to all

Pi also needs pi install npm:pi-subagents first. Qwen Code uses its native extension installer; it is not a generated --to qwen converter target. Cursor, Copilot, and Droid are also native plugin ecosystems, not --to targets. --to all only writes generated targets detected on the machine and skips native-only plugin ecosystems.

For a non-default Codex profile, point both Codex and the installer at the same root:

export CODEX_HOME="$HOME/.codex/profiles/research"
codex plugin marketplace add "$CE_DS_REPO"
cd "$CE_DS_REPO"
bun run src/index.ts install ./plugins/methodsmith --to codex --codex-home "$CODEX_HOME"
CODEX_HOME="$CODEX_HOME" codex

Inside Codex, run /plugins, select this local marketplace, install methodsmith, then restart. Codex's native plugin install provides the skills; the Bun command above adds generated agents until Codex supports plugin-defined agents natively.

Codex installs have two supported modes:

• Native plugin + agent bridge (recommended): install the plugin inside Codex with /plugins, then run install --to codex --codex-home "$CODEX_HOME" to add generated agents to the same profile.
• Standalone generated install: run bun run src/index.ts install ./plugins/methodsmith --to codex --codex-home "$CODEX_HOME" --include-skills when native plugin install is unavailable. This writes generated skills and agents. Managed .codex/hooks.json entries and MCP configuration are written only for plugins that actually declare hooks or servers; Methodsmith currently declares neither.

Managed Codex hooks are tagged with plugin metadata so upgrades can replace this plugin's hook entries without deleting manual hooks or hooks owned by another plugin. If an existing hooks.json is malformed, the installer backs it up before writing a managed replacement.

See Codex profile and hook installation for the profile, standalone, and recovery details. See the setup guide for exact Claude, Codex, OpenCode, Gemini, Kiro, Pi, and Qwen walkthroughs.

---

Keep Methodsmith updated

How a new version reaches you depends on the agent.

Agent	Automatic?	Update by hand
Claude Code	Yes, once auto-update is on for the methodsmith marketplace. bash install.sh claude and .\install.ps1 claude turn it on (add --no-auto-update or -NoAutoUpdate to skip). After a one-line claude plugin install ... --marketplace sajor2000/methodsmith, turn it on yourself: run /plugin, open Marketplaces, select methodsmith, and choose Enable auto-update.	claude plugin update methodsmith@methodsmith, then /reload-plugins
GitHub CLI skills-only install	No.	Re-run your gh skill install command after a new release.
Codex	No. Codex has no documented automatic plugin update.	codex plugin marketplace upgrade methodsmith, then restart Codex. Refresh the generated agents with bunx github:sajor2000/methodsmith install methodsmith --to codex
OpenCode, Pi, Gemini CLI, Kiro	No. These use generated files.	bunx github:sajor2000/methodsmith install methodsmith --to opencode (or --to pi, --to gemini, --to kiro; add --output <workspace> for a project install)
Qwen Code	No.	qwen extensions update methodsmith
Local checkout or offline package	No.	git pull in the checkout (or unpack the new package), then re-run the installer

With auto-update on, Claude Code checks in the background a few minutes after your first message and shows Plugin updated ... Run /reload-plugins to apply. Your current session keeps the version it started with. A marketplace added from a local folder never auto-updates, because Claude Code loads that folder in place; that is why the installers now add the GitHub marketplace unless you pass --source. Setting DISABLE_AUTOUPDATER=1, DISABLE_UPDATES=1, or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 turns the background update off unless you also set FORCE_AUTOUPDATE_PLUGINS=1.

Not sure which command applies? Run the method-update skill (/method-update in Claude Code). It prints the exact command for the agent you are in.

Your own skills are safe

Updates never delete or overwrite skills, agents, prompts, or commands you created.

• Generated installs (Codex agents, OpenCode, Pi, Gemini CLI, Kiro) record every file they write, with a content hash, in <agent folder>/methodsmith/owned-files.json. An update deletes only unchanged files on that list.
• Files from an install made before ownership tracking are not on the list. The installer treats one as the plugin's only when its install record or its exact description shows methodsmith wrote it, and then moves it to <agent folder>/methodsmith/backups/ or legacy-backup/ instead of deleting it. This includes Codex's retired ~/.codex/methodsmith/skills folder.
• If you edited a file the plugin installed, the update copies your version to <agent folder>/methodsmith/backups/<time>/ before replacing it.
• A file the plugin did not write is left alone, even when it has the name of an old methodsmith skill such as setup or skill-creator. The installer prints one line saying it kept the file.
• Claude Code keeps the plugin in its own plugin cache under ~/.claude/plugins/, replaces that copy on every update, and deletes old versions later, so never keep your own files there.

Keep your own skills here:

Agent	For you, in every project	For one project
Claude Code	~/.claude/skills/<name>/SKILL.md	.claude/skills/<name>/SKILL.md
Codex	~/.agents/skills/<name>/SKILL.md	.agents/skills/<name>/SKILL.md
OpenCode	~/.config/opencode/skills/<name>/SKILL.md	.opencode/skills/<name>/SKILL.md
Pi	~/.pi/agent/skills/<name>/SKILL.md	.pi/skills/<name>/SKILL.md
Gemini CLI	~/.gemini/skills/<name>/SKILL.md	.gemini/skills/<name>/SKILL.md
Kiro	Not covered here	.kiro/skills/<name>/SKILL.md
Qwen Code	~/.qwen/skills/<name>/SKILL.md	.qwen/skills/<name>/SKILL.md

Do not start your own skill names with method-. That prefix belongs to the plugin; if a name is already taken by your own skill, the installer skips the plugin's version and tells you.

Update a source checkout

cd ~/methodsmith && git pull && bun install

Then restart your coding agent.

To build corporate/offline ZIPs from a source checkout:

bun run package:corporate

---

Troubleshooting

"Unknown command" on /method-setup: Native Claude plugin commands are namespaced. Use /methodsmith:method-setup, or install the optional local aliases into .claude/commands.

bun install fails: Run bun --version. If missing: curl -fsSL https://bun.sh/install | bash

Corporate laptop blocks Bun, GitHub CLI, Git, or Quarto: Use the offline Claude folder/ZIP or Codex local marketplace package. Bun and Git are for source builds; GitHub CLI is only for GitHub helpers; Quarto is only required for Quarto render/manuscript workflows.

Plugin seems outdated: cd ~/methodsmith && git pull && bun install, then restart.

Codex installed into the wrong profile: Set CODEX_HOME on both the codex command and the Bun installer. The installer defaults to $CODEX_HOME when set, otherwise ~/.codex.

Codex hooks stopped loading after a broken edit: Re-run the standalone install with --include-skills. The installer backs up malformed .codex/hooks.json before writing managed hook entries.

CLIF activating on a non-CLIF project: /method-clif --off disables it for the session.

---

Full inventory

	Count
Skills	67
Agents	55
Reporting checklists	49

See every skill and agent

---

FAQ

Is this an AI statistician? No. It is a set of workflows and reviewers that make an analysis explicit and checkable. The researcher chooses the methods, reads the review, and is accountable for every result, as ICMJE and the ASA expect.

Does my data leave my machine? Methodsmith has no server of its own; it runs inside your coding agent. Literature skills query public services such as PubMed when you ask them to, and your agent's model provider sees what the agent reads. That is why the PHI guard keeps data values out of the model in schema-only projects and keeps identifiers out of MCP tools. See Sensitive-data boundary and What external tools receive.

R or Python? Either. Setup detects the language, notebook format and data layer from the project and only asks about what it cannot infer.

Which agents does it run on? Claude Code and Codex natively, Pi through a generated install, and others through the converter CLI. See Also works with Codex, Pi, Gemini, and more.

How is it related to Compound Engineering? It is an MIT-licensed fork. The loop, the review architecture and much of the engineering workflow come from Every's plugin; Methodsmith adds the research-methods skills, reviewers, checklists and guards.

Built on

Methodsmith started as a fork of Every's Compound Engineering plugin by Kieran Klaassen. The plan -> work -> review -> compound loop, the review personas, and the cross-platform converter come from that work. Also influenced by BMAD Method and Superpowers. Brand assets and their sources are in docs/brand.

License

MIT — Copyright (c) 2026 Juan Carlos Rojas. Original compound-engineering Copyright (c) 2025 Every.