@xaccefy/pi-xpi

XPI — offensive security tools for Pi Agent. Casefile tracking, web search, library docs, exploit technique search, and todo tracking.

Packages

Package details

extensionskill

Install @xaccefy/pi-xpi from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:@xaccefy/pi-xpi
Package
@xaccefy/pi-xpi
Version
0.9.4
Published
Aug 16, 2026
Downloads
3,778/mo · 629/wk
Author
xaccefy
License
MIT
Types
extension, skill
Size
695.3 KB
Dependencies
4 dependencies · 4 peers
Pi manifest JSON
{
  "extensions": [
    "./packages/pi-casefile/src/index.ts",
    "./packages/pi-webxp/src/index.ts",
    "./packages/pi-xtodo/index.ts"
  ],
  "skills": [
    "./skills"
  ],
  "subagents": {
    "agents": [
      "./agents"
    ]
  }
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

XPI

Security tooling for the Pi agent — casefile tracking, web search, exploit-technique intelligence, code search, and todos.

npm version npm downloads License: MIT

What it is

XPI turns the Pi agent into a security researcher: a case ledger with enforced gates, real exploit-technique grounding, web lookup, fast code search, and a pipeline that keeps findings honest.

  • Casefile — hypothesis → investigating → confirmed → reported, with gates at every step
  • Machine-owned PoC gates — zero exit is necessary but never proof: direct-response findings require nonce-bound body evidence plus a DNS-pinned, conclusive target_only replay against an operator-approved control; reflection-capable requests can add a harness-generated target-only canary; only the main agent may make the semantic decision and commit phase 2
  • Exploit chains — ChainSuggest surfaces combinations the model missed
  • Coverage matrix — machine-checkable "we tested everything" claims
  • Code search — structural AST search (ast-grep) for sinks and call chains, plus built-in grep/find for text

Install

Works on Pi Agent and its fork OMP (@oh-my-pi/pi-coding-agent). One manifest serves both: OMP reads the same pi extension field, the Agent Plugins plugin.json for skills, and the task tool spawns the specialist agents.

./install.sh            # auto-detects pi or omp in PATH
./install.sh --pi       # force Pi  (installs pi-subagents + optional ast-grep structural search)
./install.sh --pi --no-subagents  # minimal Pi install; use /xp lite because swarm dispatch is unavailable
./install.sh --omp      # force OMP (copies agents/*.md to ~/.omp/agent/agents)

Or install the npm package per host:

pi install npm:@xaccefy/pi-xpi     # Pi
omp install npm:@xaccefy/pi-xpi    # OMP

Set PREVIEW_IS_API_KEY for exploit_search (see docs/guide.md).

Quick start

/xp        # toggle bounded swarm XP mode on/off
/xp lite   # explicit single-agent security workflow
/xp swarm  # bounded multi-agent pipeline

Use /xp for the default bounded swarm workflow, or /xp lite for CTFs, focused reviews, and one-target work where dispatch is unnecessary. On Pi, swarm dispatches only auditor, tracer, skeptic, and chain via pi-subagents; on OMP it uses the native task tool with the same four agent names. Validation, patching, reporting, and ConfirmFinding stay with the main agent.

Full tool reference, configuration, and pipeline docs: docs/guide.md.

Packages

Package npm
Umbrella @xaccefy/pi-xpi
Case ledger @xaccefy/pi-casefile
Web lookup + exploit search @xaccefy/pi-webxp
Todos @xaccefy/pi-xtodo

Develop

bun install
bun test --isolate
bun run typecheck