AgentPM™

@zack/agentpm-docs

Prepared vector-mode retrieval corpus built from the public AgentPM v0.1 docs set.

Install
agentpm install @zack/agentpm-docs@0.1.0
Inspect / query
agentpm knowledge inspect @zack/agentpm-docs@0.1.0
Weekly installs
0
0%
Last publish
Today
v0.1.0
OverviewReadmeSecurity

agentpm-docs

agentpm-docs is a publishable AgentPM Knowledge package built from the public AgentPM documentation corpus.

It is intentionally the larger, pipeline-driven Knowledge example in this repo. The point of this package is to show how real content can be copied, normalized, chunked, sourced, embedded, and then packaged into the AgentPM mode: "vector" shape.

This package includes:

  • copied source docs under knowledge/source-docs/v0.1/
  • generated knowledge/chunks.jsonl
  • generated knowledge/sources.jsonl
  • generated knowledge/provenance/sources-manifest.json
  • a canonical raw float32 vectors file at knowledge/embeddings/default.f32
  • build-generated local index metadata after agentpm knowledge build

Corpus prep workflow

The scripts in scripts/ are the reproducible corpus-prep pipeline for this package:

  • scripts/build_corpus.py
    • copies agentpm-api/docs/v0.1/**/*.mdx
    • strips frontmatter for chunk text
    • uses langchain-text-splitters for header-aware Markdown splitting plus recursive chunking
    • preserves source traceability
    • emits AgentPM-compatible chunks.jsonl, sources.jsonl, and provenance metadata
  • scripts/openai_embeddings.py
    • generates text-embedding-3-small vectors for the chunk corpus
    • can also generate query JSON or act as an --embedding-command adapter for agentpm knowledge query

Local development

From this directory:

uv pip install langchain-text-splitters
python3 scripts/build_corpus.py
export OPENAI_API_KEY=...
python3 scripts/openai_embeddings.py chunks-to-f32 \
  --chunks knowledge/chunks.jsonl \
  --output knowledge/embeddings/default.f32
agentpm knowledge build
agentpm knowledge inspect .
agentpm publish --dry-run

If you are not using uv, the equivalent install step is:

python3 -m pip install langchain-text-splitters

This keeps the final package in AgentPM's expected JSONL/vector shape, while showing a corpus-prep workflow that is closer to what many teams already use in retrieval pipelines.

Query demo

After build, you can generate a matching query vector:

export OPENAI_API_KEY=...
  python3 scripts/openai_embeddings.py query-to-json \
    --text "How do I scaffold and test an AgentPM template package?" \
    --output knowledge/query.json

agentpm knowledge query . --vector-json knowledge/query.json --top-k 5 --include-text

Or use the adapter path:

agentpm knowledge query . "How do I scaffold and test an AgentPM template package?" \
  --embedding-command "python3 scripts/openai_embeddings.py adapter" \
  --top-k 5 \
  --include-text

The source for this Knowledge package can be found here.