radiology-reproducibility
Audit/package computational replay with manifests, environments, provenance and evidence levels.
Install / Use
npx skills add huang-sir1/radiology-skills --skill radiology-reproducibilityInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
Development & EngineeringSupported Platforms
Our assessment of radiology-reproducibility
radiology-reproducibility scores 82/100 on our quality scale, 3360th of 4,700 Development & Engineering skills we index.
Its SKILL.md is 11 KB long, well organised into 10 sections and no code examples: a thorough specification that gives an agent plenty to work with.
With 1,785 GitHub stars, it is one of the more widely adopted skills in the catalogue.
Maintenance, license and trust
- The repository was last updated about 2 months ago, so radiology-reproducibility is actively maintained.
- Our last check on 2026-11-07 found the source still online.
- No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
- Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-10-01. Automated pattern scan on 2026-10-01. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
radiology-reproducibility compared with similar skills
All 4 of these similar skills score higher than radiology-reproducibility; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| radiology-reproducibility (this skill)by huang-sir1 | 82 | 1.8k | 56d ago | SKILL.md |
| taste-skillby Leonxlnx | 100 | 116.8k | 8d ago | CLAUDE.md |
| ai-job-searchby ai-job-search-dev | 100 | 72.1k | today | CLAUDE.md |
| claude-howtoby luongnv89 | 100 | 47.3k | 2d ago | CLAUDE.md |
| Claude-Code-Game-Studiosby Donchitos | 100 | 39.7k | 7d ago | CLAUDE.md |
Frequently asked questions
- How do I install radiology-reproducibility?
- Run
npx skills add huang-sir1/radiology-skills --skill radiology-reproducibility. The install tabs above show the steps for each supported agent. - Which AI agents does radiology-reproducibility work with?
- It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is radiology-reproducibility safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is radiology-reproducibility still maintained?
- The repository was last updated about 2 months ago, so radiology-reproducibility is actively maintained.
Skill content
View source on GitHubname: radiology-reproducibility description: "Audit/package computational replay with manifests, environments, provenance and evidence levels."
Radiology Computational Reproducibility
Use this skill to make a radiology analysis traceable, rerunnable, replayed or independently reproduced through inspectable computational receipts, and to preserve—but not adjudicate—linked external replication/transport evidence. Its unique responsibility is the executable chain from frozen inputs through code/configuration/environment to outputs. It does not own scientific design, statistical validity, data-sharing authority, software security certification or external clinical validation.
Non-negotiable boundaries
- Apply the cumulative computational axis exactly:
TRACEABLE -> RERUNNABLE -> REPLAYED -> INDEPENDENTLY_REPRODUCED. Assign only the highest level actually evidenced per target result. RecordEXTERNALLY_REPLICATED_OR_TRANSPORTEDonly on a separate scientific-validation axis owned byradiology-design/radiology-translation; it is neither the next computational level nor conditional on same-data independent replay. - File existence, a repository URL or a container tag does not prove replay. A successful replay on the same inputs does not establish independent replication, transportability or clinical validity.
- Freeze code commit and dirty state, configuration, environment/container digest, input/output manifests, run ID, seeds, model weights, data version and expected tolerance before claim-bearing replay.
- Record exact evidence states:
ASSERTED,PRESENT_UNVERIFIED,IDENTITY_VERIFIED,EXECUTED,TOLERANCE_PASSED,INDEPENDENTLY_VERIFIEDandEXTERNAL_DATA_VERIFIED. Do not upgrade a link or author statement into execution evidence. - Restricted data need not be public. Access controls, consent, DUA, licences and repository conditions override convenience; a synthetic fixture tests mechanics but does not reproduce a result obtained from unavailable real data.
- Never silently patch code, substitute data, relax tolerance or reuse hidden local state during a replay. A repair creates a new package version and a new receipt.
- Replay execution requires user authorisation, a safe workspace and available lawful inputs. An audit or package plan does not imply permission to run code, download restricted data or publish.
Route by mode
| Need | Mode | Read |
|---|---|---|
| inventory code/data/environment/provenance and assign the current level | trace-audit | replay-contract-and-levels.md |
| design a minimum replay package or repair missing artifacts | package-plan | both method references |
| perform or specify a frozen replay and issue a run receipt | replay | replay-contract-and-levels.md |
| diagnose divergence without hiding failed runs | discrepancy-diagnosis | replay-contract-and-levels.md |
| freeze licences, restricted-data access and a citable archive handoff | archive-handoff | restricted-data-archiving-and-software.md and source-registry.md |
Use reproducibility-passport.md for the evidence-level decision and minimum-replay-package-manifest.md for a persistent package. Do not load every reference for a narrow audit.
Input passport
Record:
- research object, target result/claim, package version, intended computational level and any separately supplied external-validation receipt;
- executable entrypoint, source repository, immutable commit, dirty state and code licence;
- complete configuration, command, working-directory assumptions and external services;
- runtime, OS, hardware/CPU/GPU, drivers, CUDA, libraries/lockfiles and container tag plus digest;
- input manifest with content identity/hash, format, hierarchy, data release/version, access class, preprocessing eligibility and no exposed identifiers;
- seed(s), determinism settings, stochastic repeats, trained weights/checkpoint hash and training data/version relationship;
- expected outputs, metrics, reference values, tolerance rule and whether tolerance was frozen before execution;
- operator/team, timestamp, run ID, logs, exit state, output hashes and independent-environment details;
- archive identifier, licences and restricted-data replay route.
Use AUTHOR_INPUT_NEEDED; never infer an input identity, model weight, DICOM series,
preprocessing version, environment or successful run from naming convention.
Two-axis evidence verdict
TRACEABLE: the claimed output can be linked to identifiable inputs, code/configuration and provenance, but executability has not been established.RERUNNABLE: an executable package, dependencies, instructions and lawful input route are present and identity-checked; no qualifying replay receipt yet exists.REPLAYED: the frozen package was executed on the same identified inputs and produced the prespecified outputs within the frozen tolerance. Operator independence is not required.INDEPENDENTLY_REPRODUCED: a different operator/team, without hidden author state, rebuilt or instantiated the documented environment and replayed the same input/code/method within tolerance. The four-level computational axis is cumulative only while lower-level evidence remains valid. Report a mixed state if one result is replayed and another is merely traceable; never average levels across artifacts.
On the separate scientific axis, EXTERNALLY_REPLICATED_OR_TRANSPORTED means a governed analysis on
new data, site, population, protocol or time supports a prespecified external claim. It can coexist
with any computational level and requires the appropriate design/translation owner. A new-site result
without a reproducible package may carry external evidence while remaining only TRACEABLE
computationally; conversely, independent same-data replay is not external validation.
Required workflow
- Freeze the target. Name the exact table, figure, model metric, image/segmentation derivative or decision artifact and its expected values/tolerance before replay.
- Construct the provenance graph. Link input entity -> preprocessing/activity -> intermediate -> analysis/model -> output, with responsible agent, version, time and identifiers.
- Verify identities. Hash physical files where lawful, record dataset/release and code commit, verify config/weights, and distinguish tag/name from immutable digest.
- Audit the environment. Capture OS/runtime, lockfile, system libraries, drivers/hardware, locale/timezone and external service versions. A container reduces—but does not erase—hardware, data, service and nondeterminism dependencies.
- Run only when authorised. Use a fresh controlled workspace; invoke the documented entrypoint; preserve stdout/stderr, command, resource state, start/end, exit code and every produced output. Failed and negative runs remain visible.
- Compare against frozen tolerance. Check expected files, schemas, row/subject counts, numeric tolerance/distribution and hashes as appropriate. Seeds alone do not guarantee deterministic GPU or parallel execution.
- Diagnose discrepancies by class. Use
IDENTITY,ENVIRONMENT,DEPENDENCY,NONDETERMINISM,NUMERICAL,DATA_ACCESS,EXTERNAL_SERVICE,HARDWAREorALGORITHMIC. Do not repair in place. - Issue the receipt and level. Separate observed artifact/run evidence, inference about cause and recommended repair. Archive the frozen package only after licence/privacy review.
STOP gates
STOP_INPUT_IDENTITY: input release/object identity or hierarchy is unresolved.STOP_CODE_CONFIG_IDENTITY: executable entrypoint, immutable code state, full config or model weights are missing/ambiguous.STOP_ENVIRONMENT_UNRESOLVED: required runtime/dependency/hardware state cannot be recreated or assessed.STOP_TOLERANCE_NOT_FROZEN: acceptance tolerance was absent or changed after observing replay output.STOP_DATA_AUTHORIZATION: restricted input, model derivative, credential, consent, DUA or transfer authority is unresolved.STOP_LICENSE_OR_ARCHIVE: licence conflict, embedded secret/identifier or non-versioned archive prevents lawful reusable release.STOP_REPLAY_NOT_AUTHORIZED: execution or required data access is outside current permission.
A STOP limits the replay/level claim. Return a trace inventory and smallest lawful repair where possible.
Radiology-specific provenance
- Preserve patient -> examination/time point -> series/acquisition -> reconstruction -> derived image/map -> ROI/segmentation -> feature/model/output hierarchy without embedding direct identifiers in a public manifest.
- Freeze modality, phase/sequence/tracer/view, reconstruction, resampling, registration, normalisation, harmonisation, segmentation/annotation, radiomics definition and software version.
- For deep learning preserve split/cohort version, preprocessing, architecture/config, weights, training checkpoint, random state, inference mode and hardware/runtime.
- A DICOM/NIfTI filename or series description does not verify the pixel object. A checksum verifies identity, not acquisition quality, correct series selection or authorisation.
- Harmonising or recalibrating on external data is adaptation. It cannot retain an untouched external-transport claim.
Output contract
Route and target— mode, object, package version, supplied artifacts and authority.Provenance and identity map— code/config/environment/input/activity/output chain.Replay readiness matrix— present, verified, missing and STOP states.Run receipt— command, operator, environment, timing, exit, logs and outputs when executed.Tolerance and discrepancy report— frozen rule, observed difference, class and evidence.Two-axis verdict— highest supported computational level per result, separate supplied external-validation state/owner and prohibited cross-axis upgrade.Restricted-data/licence/archive plan— access route, reusable components and persistent ID.Handoff— evidence vs causal inference vs recommendation and unique receiving owner.
Handoffs and non-ownership
- Data identity, de-identification, DMS/repository and access ->
radiology-data; consent/DUA and participant governance ->radiology-ethics. - Acquisition/reconstruction/series validity ->
radiology-acquisition-qc; segmentation ->radiology-annotation; radiomics implementation ->radiology-radiomics; model training ->radiology-deep-learning. - Statistical methods, tolerance justification and scientific inference ->
radiology-stats; method-performance comparison ->radiology-method-evaluation. - New-site/population clinical validation ->
radiology-design/radiology-translation; project-level state and cross-skill receipts ->radiology-pipeline. - This skill does not certify cybersecurity, regulatory software compliance, data-sharing permission, statistical correctness or external clinical validity.
Related Skills
taste-skill
116.8kAgent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
ai-job-search
72.1kThe job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it.
claude-howto
47.3kA visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value.
Claude-Code-Game-Studios
39.7kTurn Claude Code into a full game dev studio — 49 AI agents, 72 workflow skills, and a complete coordination system mirroring real studio hierarchy.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
