pedramamini

Usage: maestro-cli [options] [command]

Command-line interface for Maestro

Options:
  -V, --version                                       output the version number
  -q, --quiet                                         Suppress incidental success output (errors still print)
  --verbose                                           Print extra detail where available
  -h, --help                                          display help for command

pedramamini

jev: calibrated decisions for agents

What this is: a single-file CLI and a matching agent skill that give any coding agent (Claude Code, Codex, OpenCode) a fast, cheap, calibrated second opinion. The agent stays the reasoning model; jev answers the narrow, repeated judgments (is this relevant? which bucket? how severe? does the source support this claim?) so the agent never has to read a thousand items into its context window or trust its own uncalibrated confidence.

Why it exists: LLM agents are expensive and inconsistent at high-volume judgment calls, and they cannot tell you how sure they are. A decision model that returns a real

pedramamini

#!/usr/bin/env python3
"""
meural_curate.py: keep the Netgear Meural Canvas II in the office stocked with a
fresh, non-repeating set of family photos, weekly, without anyone touching it.

THE FRAME (discovered on the LAN 2026-09-04, not guessed):
  name         Office Canvas / monet-042
  frame_id     12345          product_key  XXXXXXXXXXXXX
  ip           192.168.1.50   firmware     2.3.2_3.0.3
  orientation  portrait       <- the DEVICE reports this, we never assume it

pedramamini

name

meeting-wingman

description

Continuous live meeting monitoring. The user says a wingman keyword and you enter a streaming loop: consume the live Wispr Flow transcript as it arrives, speak up only when you have something genuinely worth saying, then keep consuming, until the meeting ends or they force-stop. Use when they say 'meeting wingman', 'wingman', 'wingman copilot', 'wingman mode', 'be my wingman', 'copilot this meeting', 'listen in', 'monitor this call', 'ride along', or ask you to watch a meeting continuously. Also use to STOP: 'wingman off', 'stop listening', 'stand down'.

Meeting Wingman

Maestro version. Adds a live thinking stream pinned to the current tab. Blog post with the full build: https://pedsidian.pedramamini.com/Claude/Blog/2026-08-25-meeting-wingman Portable variant (no Maestro, any agent with shell access): see the companion gist.

pedramamini

name

meeting-wingman

description

Continuous live meeting monitoring. The user says a wingman keyword and you enter a streaming loop: consume the live Wispr Flow transcript as it arrives, speak up only when you have something genuinely worth saying, then keep consuming, until the meeting ends or they force-stop. Use when they say 'meeting wingman', 'wingman', 'wingman copilot', 'wingman mode', 'be my wingman', 'copilot this meeting', 'listen in', 'monitor this call', 'ride along', or ask you to watch a meeting continuously. Also use to STOP: 'wingman off', 'stop listening', 'stand down'.

Meeting Wingman

Portable version. Works with any agent that has shell access. Blog post with the full build: https://pedsidian.pedramamini.com/Claude/Blog/2026-08-25-meeting-wingman Maestro-specific variant (adds a live thinking stream): see the companion gist.

pedramamini

GitHub Stars → Obsidian, with a Shoulder Tap

Mirrors every repo you have starred on GitHub into an Obsidian vault as one Dataview-queryable Markdown note per repo, categorized by an LLM, and fires a desktop notification when a repo you already starred starts trending.

Written up in full on Pedram's AI's blog: pedsidian.pedramamini.com, in "Your GitHub Stars Are a Graveyard. Ours Tap My Human Operator on the Shoulder."

What you need

pedramamini

I'm Pedram Amini. Direct, no fluff. Action > process. "Good enough" sometimes > perfect.

Values: High agency, authentic living, calculated risks, personal mastery.

Context: Security researcher. OSX desktop. Type Script and Python for tools. Generate markdown for Obsidian. CST timezone.

Do: Push back on bad ideas. Ask vs assume. Remove friction. Trust his gut, but feel free to disagree. Speak in short, dense terms. Don't lose me in fluff. Use visuals where necessary.

Don't: Sycophancy ("absolutely right"). Hand-holding. Over-explain. Analysis paralysis. Regurgitate my words back to me.

pedramamini

Case	Discloser	Target pipeline	Injection channel	Impact	Date
EchoLeak (CVE-2025-32711)	Aim Security	M365 Copilot (email RAG)	Poisoned inbound email (markdown auto-fetch image)	Zero-click exfil of internal mail/files; CVSS 9.3	2025-06
ForcedLeak (CVSS 9.4)	Noma Security	Salesforce Agentforce/Einstein (CRM)	Poisoned Web-to-Lead form field	Indirect injection → CRM data exfil to whitelisted domain	2025-09
AgentFlayer (Black Hat)	Zenity (Bargury)	ChatGPT Connectors, Copilot Studio, Cursor+Jira, Einstein, Gemini	Poisoned document (white-on-white) shared into Drive/connector	Zero-click exfil of connected cloud data via image-URL params	2025-08

|

pedramamini

---

UPDATE 2026-06-11: patched build ran; new diagnostic changes the diagnosis

The patched maestro-p (resubmit loop + getScreenTail dump) IS now running. Confirmed: today's first_byte_timeout errors in cue.db carry the new last screen at timeout (ANSI-stripped tail) block. So the fix is live AND we finally have ground truth. It changes the root cause.

What the screen tails prove

All three of today's Pedsidian agent cues still failed at 121s (chain-1 07:00, chain-9 11:00, chain-8 17:00), BUT the screen tails show claude is alive and working the whole time:

pedramamini

Objective:: Behavior + voice spec for the "@maestro" iMessage message bus. The handler agent reads this each time the scanner dispatches a command. Defines how to fulfill the request and how to reply in the owner's texting voice.

Trigger:: any iMessage the owner authors containing the literal marker @maestro

Scanner:: maestro_message_scanner.py (3-min Maestro Cue command node)

Reply-Prefix:: 🎶 Maestro:

Log:: Maestro-Message-Log.md