PrimeKit Agent Tools

PrimeKit is a local toolkit for preparing compact, auditable context for AI-assisted business analysis.

It turns messy source files into smaller deterministic artifacts before an LLM or human reviewer reads them. The core idea is simple: do repeatable calculations in code, then give the model only the important evidence.

Main Capabilities

• Normalize facts from CSV, JSON, Markdown, and spreadsheet-derived inputs.
• Build compact context packs.
• Compare weekly performance snapshots.
• Check margins, stock risks, commitments, stale facts, and conflicting assumptions.
• Generate structured evidence packs.
• Run local research/source snapshot workflows.
• Compose brief daily reports from explicit task logs.

Why It Matters

LLM workflows often waste context on raw files. PrimeKit reduces that waste by producing small, traceable outputs:

raw source -> adapter -> normalized facts -> deterministic tool -> compact artifact -> AI/human review

This keeps the final reasoning step focused on judgment, not file parsing.

Tech Stack

• Python standard library
• SQLite / FTS5
• JSON / JSONL / Markdown artifacts
• Unit tests with unittest

Run Tests

python -m unittest discover -s tests -v

Example Commands

python -m primekit.cli margin-guardrail --economics economics.json --output-dir runtime/demo/margin
python -m primekit.cli commitment-check --draft draft.txt --context context.json --output-dir runtime/demo/commitment
python -m primekit.cli memory-index --source docs --output-dir runtime/demo/memory

Repository Scope

This public version includes tool code, schemas, tests, and sanitized documentation.

It intentionally excludes runtime outputs, business data, local memory, private reports, credentials, and any customer/company files.

Status

Active prototype. The next step is turning the CLI into a cleaner package with example fixtures and typed interfaces.