Security Scan orchestrator for AI Agents.

Parsentry analyzes repository structure, enumerates attack surfaces, and generates per-surface analysis prompts. Pipe them to any CLI agent for parallel security analysis.

| 日本語

Agent Skill

npx skills add HikaruEgashira/parsentry

Once installed, just ask your agent to scan a repository — the skill dispatches the scan process automatically.

Usage

parsentry model owner/repo | claude -p
parsentry scan owner/repo | claude -p

model produces a threat model, scan generates per-surface prompts and outputs an orchestrator prompt to stdout. Pipe it to any CLI agent and it dispatches parallel subagents automatically.

See HACKING.md for usage with other CLI agents.

How it works

parsentry model   →  repo metadata  →  threat model prompt (stdout)
parsentry scan    →  threat model + source code  →  orchestrator prompt (stdout)
  └─ orchestrator dispatches subagents per surface  →  SARIF files
parsentry log     →  monitor scan progress in real-time

1. Model — Collect repo metadata (languages, manifests, entry points), generate a threat model prompt. Pipe to an agent to enumerate attack surfaces.
2. Scan — Load the threat model, read source code per surface, generate focused .prompt.md files. Outputs an orchestrator prompt that launches parallel subagents — each reads its own prompt file and writes SARIF results.
3. Log — Stream scan progress like docker compose logs -f.

Install

# Via mise
mise use -g github:HikaruEgashira/parsentry

# Via cargo
cargo install parsentry

Example Reports

Repository	Surfaces
langgenius/dify	25
OWASP/NodeGoat	19
HikaruEgashira/hikae-vulnerable-python	25

License

AGPL 3.0