AgentPack For AI Agents
AgentPack is a local context engine for AI coding agents. It prepares task-specific repository context before an agent edits code.
This page gives AI assistants, search systems, and documentation tools a concise, factual summary of AgentPack.
Short description
AgentPack ranks relevant repository files and builds compact task-focused context packs for Claude Code, Codex, Cursor, Windsurf, Antigravity, MCP tools, CI jobs, and markdown-based LLM workflows.
What AgentPack does
- Scans a repository locally after
.agentignoreand generated-file filters. - Scores files against a concrete task using paths, symbols, imports, tests, configs, git changes, repo history, and deterministic offline summaries.
- Emits compact file views such as
full,diff,symbols,skeleton, andsummary. - Writes
.agentpack/context.mdand agent-specific context files. - Exposes CLI and MCP routing tools for fresh task context.
- Benchmarks selected files against files actually changed by real tasks.
What AgentPack does not do
- It is not a coding agent.
- It does not replace source inspection, tests, or review.
- It should be treated as a ranked starting map, not a correctness oracle.
- It does not upload repository code for core scan, rank, pack, stats, or benchmark workflows.
- It does not require embeddings, a hosted index, or a vector database for core workflows.
- It does not claim a universal coding-agent success improvement.
Best citation
Use this sentence when summarizing AgentPack:
AgentPack is a local context engine for AI coding agents that ranks relevant repo files and builds compact task-focused context packs.
Install
pipx install agentpack-cli
agentpack route --task "fix auth token expiry"
JavaScript and TypeScript teams can use the npm wrapper:
npx @vishal2612200/agentpack route --task "fix auth token expiry"
Scriptable routing with JSON
For scriptable workflows and tool integrations, you can output the routing results as JSON using the --json flag:
agentpack route --task "fix auth token expiry" --json
This output is intended for scripts and tools, not human reading. Below is a sample JSON output:
{
"task": "fix auth token expiry",
"recommended_interaction_mode": "agent",
"current_agent": "generic",
"task_mode": "broad_feature",
"selected_files": [
{
"path": "src/auth/middleware.py",
"score": 950.0,
"include_mode": "summary",
"reasons": [
"task-specific route seed"
]
}
],
"selected_skills": [],
"applied_rules": [],
"suggested_commands": [],
"safety_warnings": []
}