TianGong LCA Documentation
Integration & ExtensionAgent Skills guide

Skill catalogue and prerequisites

Choose by task and distinguish standalone search packages, full-repository workflows, and Data Foundry requirements.

Choose by the input you have and the result you need, then check prerequisites and write risks. These 21 entries come from the reviewed Skills revision. This is not a promise that installing everything makes every workflow ready.

Independently installable: three search skills

Each includes its launcher, example, and references, with no Skills/CLI checkout or Data Foundry requirement. Follow installation and replace only the name after --skill. All need the specified Node.js/pnpm versions and human-completed CLI OAuth sign-in.

SkillInput → resultPrerequisite or boundary
flow-hybrid-searchFlow description and optional filters → candidate flowsRemote read-only
process-hybrid-searchProcess description and optional filters → candidate processesRemote read-only
lifecyclemodel-hybrid-searchModel description and optional filters → candidate modelsRemote read-only

Full Skills repository: build, review, and operations

These wrappers still use the repository-level shared launcher, and some read other skills’ rules or assets. Copying a single directory is insufficient. Have the workflow maintainer prepare the complete layout, matching CLI, and task inputs. These are not the beginner standalone installation path.

SkillInput → resultPrerequisite or boundary
process-automated-builderReference flow, evidence, and plan → local process build and publishing handoffBuilding and publishing are separate; publishing needs approval
lifecyclemodel-automated-builderExisting process builds → life-cycle model and connectionsRequires upstream build directories
lifecyclemodel-resulting-process-builderExisting model and process data → aggregated process and relationsMay read remote data; publishing needs separate authorization
lifecyclemodel-recursive-orchestratorPrepared candidates and build request → multi-layer models, graph, and lineageDepends on other builders and the complete runtime layout
lifecycleinventory-reviewLocal builds or frozen remote snapshots → inventory quality reportReview output is not publishing approval
flow-governance-reviewFlow snapshots and repair plan → QA, repair, and publishing preparationSome subcommands write; approve each step
process-dedup-reviewGrouped duplicate snapshots → keep/delete recommendations and evidenceA recommendation is not deletion permission
process-scope-statisticsAuthorized process scope → snapshot, coverage statistics, and reportsRemote reads; retain scope and count evidence
current-account-dataset-reviewCurrent-account snapshots → validation, reference repair, drafts, and auditReview and repair permissions are distinct
tiangong-lca-remote-opsSpecific process maintenance input → reference refresh and post-write checksMay write remotely; not an unrestricted admin interface
lca-publish-executorCompleted publishing handoff → preflight, publish, and verification reportsActual writes need explicit approval
embedding-ftOperational input → embedding-job investigation or processingOperations only, not part of the introduction

Instruction-led skills

These organize existing CLI capabilities. Installing instructions does not prepare inputs, dependencies, or permissions. Use them within an explicit task scope; advanced maintenance is not ordinary search.

SkillInput → resultPrerequisite or boundary
tidas-contract-contextDataset type and task → schema, methodology, and rules contextNeeds a working CLI; prepares authoring context
tidas-bilingual-transcreationExtracted translation units and evidence → reviewed Chinese/English fieldsExplicit post-import language completion only, not an import/publish gate
dataset-rls-maintenanceCurrent-account maintenance scope → plan, approval, and verificationMay delete or modify data; high risk

Data production that depends on Data Foundry

These tasks require Foundry workspaces, manifests, or scripts. Do not assume one Skill installation is enough. For local conversion only, start with the independent tidas tool.

SkillInput → resultPrerequisite or boundary
external-dataset-curated-importLegitimately obtained package and Foundry manifest → curated candidatesNeeds Foundry, import configuration, and downstream skills
source-evidence-dataset-developmentSource documents and Foundry seed → evidence dossier and candidatesMay need additional research/document tools
tidas-data-importPackage or source files → conversion/authoring and import preparationNeeds Foundry; local conversion can use tidas directly

Complete one read-only task before introducing production workflows. Before real writes, read Organize tasks safely; for sign-in or script issues, use troubleshooting.

On this page