@chicamhao/pi-phalanx
Phalanx multi-agent architecture for Pi — shared memory via agora, role dispatch with chain-of-command, shield_wall retry, and consult-the-oracle escalation
Package details
Install @chicamhao/pi-phalanx from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:@chicamhao/pi-phalanx- Package
@chicamhao/pi-phalanx- Version
0.1.1- Published
- Sep 6, 2026
- Downloads
- 288/mo · 288/wk
- Author
- chicamhao
- License
- MIT
- Types
- extension, skill
- Size
- 83.7 KB
- Dependencies
- 0 dependencies · 3 peers
Pi manifest JSON
{
"extensions": [
".pi/extensions/phalanx"
],
"skills": [
".pi/skills"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
Phalanx
Command the phalanx — multi-agent dispatch for Pi with shared memory (agora), chain-of-command, shield_wall retry, and consult-the-oracle escalation.
pi install npm:@chicamhao/pi-phalanx
Rules
| Rule | Meaning |
|---|---|
chain_of_command |
A lochagos escalates to the strategos, never sideways |
scout_first |
Probe with psiloi when the target is unknown; skip it when already known |
shield_wall |
Retry once at narrowest scope, once more on an escalation model if configured, then escalate |
consult_the_oracle |
If ambiguous or retries exhausted, ask the oracle |
single_state |
No private state; all reads/writes go through agora |
concise_output |
Extremely concise output — no preamble or narration |
Commands
/phalanx-muster— one-line status: agora key/log count, token cost & usage since the last start, and elapsed time/phalanx-reform— reset agora runtime state (keys, messages, log, attempts)/phalanx-arrange— readphalanx-architecture.yamland auto-create missing agent (.pi/agents/*.md) files
How to use
- (Optional) update
phalanx-architecture.yamlto define new custom roles, rules, deployment scope. - (Optional) Create
conventions.yamlin your project root — code style (see this repo's copy as a template) - Copy
.pi/into your project root - Start a Pi session ->
/phalanx-arrange->/reload. The strategos loads automatically and the phalanx handles the rest
How it works
The extension provides the infrastructure — the agora, phalanx_dispatch,
and phalanx_status tools, plus /phalanx commands.
Skills are not loaded automatically. They are referenced by name in the strategos prompt and read on demand when the task matches their description. Each skill file teaches the agent how to handle a specific job:
| Skill | When to use |
|---|---|
phalanx-strategos |
Planning an objective and reporting outcomes — the strategos's default mode |
phalanx-psiloi |
Fast codebase reconnaissance when the target location is unknown |
phalanx-lochagos |
Getting work done — one generalist pass or a large multi-domain split |
phalanx-agora |
Sharing state across dispatches via the memory bus |
phalanx-oracle |
Escalating to the user when stuck or ambiguous |
The strategos (the main session) loads these skill files as needed and applies their instructions. You never install or enable skills — they are just markdown files that describe how to use the extension's tools.
Use in your own project
Install the package in your project dir:
pi install npm:@chicamhao/pi-phalanx(Optional) update
phalanx-architecture.yamlto define new custom roles, rules, deployment scope.(Optional) Create
conventions.yamlin your project root — code style (see this repo's copy as a template)Start a Pi session →
/phalanx-arrange→/reload. The strategos loads automatically and the phalanx handles the rest
File layout
├── package.json # Pi package manifest
├── conventions.yaml # Code style, git, file conventions (loaded by lochagos-work/build)
├── phalanx-architecture.yaml # Roles, tiers, rules, extend templates
├── .pi/
│ ├── agent/AGENTS.md # Strategos system prompt (overrides global)
│ ├── agents/ # Subagent system prompts (psiloi, lochagos-*)
│ ├── extensions/phalanx/ # Extension source code (TypeScript)
│ ├── skills/phalanx-*/SKILL.md # Skill instructions (loaded on demand)
│ └── phalanx/agora.json # Runtime shared memory (gitignored)
⚠️ The content and scripts in this project are AI-generated and may contain errors or inaccuracies.