taspac

taspac is a small CLI-first Agent-Ready Task Packet Generator.

Give it:

• a GitHub repository (or local repo)
• a plain-English engineering task

and it generates a structured markdown packet that gives AI coding agents useful repo-specific context before implementation starts.

taspac generate --repo https://github.com/example/repo --task "add JWT auth to upload endpoint"

For slower networks or larger repositories, increase the clone timeout:

taspac generate \
  --repo https://github.com/example/repo \
  --task "add JWT auth to upload endpoint" \
  --clone-timeout 300

Why This Exists

Normally, when you give an AI coding agent a task, it starts with almost no context.

You end up repeatedly explaining:

• which files matter
• what parts of the repo are related
• what commands to run
• existing patterns in the codebase
• what could break if changed carelessly

taspac automates that briefing process.

It scans a repository and generates a structured task packet containing:

• relevant files
• nearby code snippets
• useful commands
• recent commits
• risks and constraints
• a focused implementation prompt

The goal is to reduce the amount of repo exploration and context reconstruction an AI agent has to do before it can start making useful changes.

taspac stays deliberately small: paste the packet into Claude Code, Codex, Cursor, Aider, or similar—no extra chat UI or multi-agent stack required.

What It Produces

The generated packet includes:

• Task
• Relevant Files
• Relevant Snippets
• Suggested Commands
• Recent Commits
• Risks / Constraints
• Agent Prompt

Install

From PyPI:

pip install taspac

From source (editable):

python -m pip install -e .

You can also run without installing:

python -m taspac generate --repo ./examples/demo-repo --task "add JWT auth to upload endpoint"

Demo Workflow

python -m taspac generate \
  --repo ./examples/demo-repo \
  --task "add JWT auth to upload endpoint" \
  --output examples/generated-packet.md

Open examples/generated-packet.md to see a sample task packet.

Design Decisions

taspac intentionally keeps the implementation direct and local-first.

It:

• scores files using lightweight heuristics instead of embeddings infrastructure
• extracts nearby snippets around task-relevant matches
• infers likely commands from project files like package.json, Makefile, and pyproject.toml
• reads recent commits touching matched files using git log
• generates simple risk/invariant hints from task keywords and matched paths

The project deliberately avoids:

• authentication
• dashboards
• vector databases
• hosted infrastructure
• multi-agent orchestration
• long-running background services

The goal was to build the smallest genuinely useful version possible within a few hours.