L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Développement · SKILL.md

ingest-l0

ingest-l0

L0 bootstrap of the abap_wiki knowledge base: creates stubs for all custom objects from the TADIR. Use this skill to initialize the repo from scratch or to ingest a new TADIR export. Deterministic operations (no LLM sub-agent), idempotent.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: ingest-l0 description: "L0 bootstrap of the abap_wiki knowledge base: creates stubs for all custom objects from the TADIR. Use this skill to initialize the repo from scratch or to ingest a new TADIR export. Deterministic operations (no LLM sub-agent), idempotent."

Ingest L0 - stub bootstrap from TADIR

Creates (or updates) the L0 stub for every custom object in the TADIR. Everything is deterministic: no sub-agents, parallelism via devclass partitioning. State lives in state/abap_wiki.db (see core/docs/01-pipeline-l0-l1.md).

When to use it

  • First-time initialization of the repo (empty DB).
  • Arrival of a new export raw/tadir/TADIR_Z_<YYYYMMDD>.XLSX.

Procedure

Shortcut: python core/src/tools/pipeline.py l0-run runs steps 1-6 as a single deterministic command (newest TADIR in raw/tadir/ or --file). The step-by-step procedure below remains the reference for diagnostics.

All commands must be run from the repo root with the venv active.

  1. Initialize the schema (idempotent):
    python core/src/tools/pipeline.py init-db
    
  2. Import the TADIR (dtype=str, excludes objects already removed from the L1 queue):
    python core/src/tools/pipeline.py import-tadir --file raw/tadir/<most recent TADIR>.XLSX
    
  3. Resolve sources (unique in-memory index, deterministic hash):
    python core/src/tools/pipeline.py resolve-sources
    
  4. Create L0 stubs (all, or by devclass partition in parallel across multiple sessions):
    python core/src/tools/pipeline.py ingest-l0            # all
    python core/src/tools/pipeline.py ingest-l0 --partition ZPACKAGE   # one package
    
    Idempotent: re-running on an already-processed package is a no-op.
  5. Enqueue L1 (only analyzable types with available source):
    python core/src/tools/pipeline.py enqueue-l1
    
  6. Verify and commit:
    python core/src/tools/pipeline.py progress
    python core/src/tools/pipeline.py export-excel
    python core/src/tools/pipeline.py git-commit --message "ingest L0: <export>"
    

Rules

  • Never modify raw/ (only the user stages it when updating the export).
  • Indexes (abap_wiki/_packages/, abap_wiki/index.md) are regenerated from queries: do not edit them manually.
  • If unknown TADIR types appear they are mapped to tadir-<x> and listed in output/reports/unknown-tadir-types.md (non-blocking).
Skills enregistrées