vector-db-ingest（矢量数据库摄取）

Dependencies

This skill requires Python 3.8+ and standard library only. No external packages needed.

To install this skill's dependencies:

pip-compile ./requirements.in
pip install -r ./requirements.txt

See ./requirements.txt for the dependency lockfile (currently empty — standard library only).

VDB Ingest Agent

Role

You ingest (index) repository files into the ChromaDB vector store so they can be semantically searched. You build or update the parent-child chunk structure that query.py searches against.

This is a write operation. The vector store is the backing index for Phase 2 search.

Prerequisites

1. First-time setup

If chromadb is not installed or vector_profiles.json is missing, run the init skill first:

python3 ./scripts/init.py

2. Verify server is running

Use the vector-db-launch skill if the server is not already up:

# Check heartbeat
curl -sf http://127.0.0.1:8110/api/v1/heartbeat

# If not running, start it:
chroma run --host 127.0.0.1 --port 8110 --path .vector_data &

See SKILL.md for full launch instructions.

Execution Protocol

Full ingest (first time or full rebuild)

python3 ./scripts/ingest.py \
  --profile knowledge --full

Incremental ingest (only new/changed files since N hours)

python3 ./scripts/ingest.py \
  --profile knowledge --since 24

Code files (uses AST parsing shim)

python3 ./scripts/ingest.py \
  --profile knowledge --full --code

ingest_code_shim.py is invoked automatically for .py and .js files to extract functions and classes as discrete chunks rather than raw text blocks.

After Ingesting

Run a quick smoke test to confirm the new content is retrievable:

python3 ./scripts/query.py \
  "describe what was just ingested" --profile knowledge --limit 3

Rules

• Never write to .vector_data/ directly -- always use ingest.py.
• Never read .sqlite3 files with cat or sqlite3 -- will corrupt context.
• Source Transparency Declaration: state which profile was ingested, how many files, and any errors.