muthuishere

Give an agent a persistent, evidence-first "brain" — a git repo it records experiences into and recalls grounded, cite-or-abstain answers from, with a deterministic constraint shield that vetoes "profitable but ruinous" decisions. Backed by the standalone `brain` CLI (Go, no network, no model endpoint); the agent itself is the reasoning LLM. Trigger on: "remember this", "what have I learned about X", "should I do this / check this decision", "record the outcome", "what do I believe about X", "consolidate the brain", "set the objective", or when wiring an autonomous loop (a desk, a project) to durable, auditable memory. Supports MULTIPLE brains, each its own git repo.

Give an agent a persistent, evidence-first "brain" — a git repo it records experiences into and recalls grounded, cite-or-abstain answers from, with a deterministic constraint shield that vetoes "profitable but ruinous" decisions. Backed by the standalone `brain` CLI (Go, no network, no model endpoint); the agent itself is the reasoning LLM. Trigger on: "remember this", "what have I learned about X", "should I do this / check this decision", "record the outcome", "what do I believe about X", "consolidate the brain", "set the objective", or when wiring an autonomous loop (a desk, a project) to durable, auditable memory. Supports MULTIPLE brains, each its own git repo.

Defender / fix proposer for AgentProxy. Consumes the latest red-team findings JSON, looks up regex coverage in community catalogs (Gitleaks, TruffleHog, secretlint — see sources.md), proposes patches to config/patterns.yaml or config/pii_patterns.yaml with unit tests, and runs the test suite to prove the fix lands cleanly. Use when the user says "fix masking gaps", "apply blueteam fixes", "patch from findings", or after running redteam-mask.

Adversarial tester for AgentProxy's regex-based masking. Crafts payloads designed to slip past current patterns in config/patterns.yaml and config/pii_patterns.yaml, drives them through the running proxy, then reads /api/traffic to score which leaked. Outputs findings JSON for the blueteam-mask skill to consume. Use when the user says "redteam patterns", "find masking gaps", "break the proxy", "probe coverage", or before any launch / release cut.

Create hand-drawn Excalidraw diagrams, flows, explainers, wireframes, and page mockups. Default to monochrome sketch output; allow restrained color only for page mockups when the user explicitly wants webpage-like fidelity.

A tiny example skill. Use when you want to see how skill loading (progressive disclosure) works end to end.

Read-only SRE / diagnostics for Kubernetes clusters on Talos Linux. Trigger when the user wants to review cluster health, run a reliability review, scan for security drift, check certificate expiry / predict a cert outage, review vulnerabilities (CVE / image scan / Talos+Kubernetes version currency / supply chain), produce a platform maturity report (scored), or remediate a finding via GitOps (open a PR that ArgoCD syncs). Phrases: "check the cluster", "is dev healthy", "health of staging", "reliability review", "security review", "any privileged pods", "are any certs expiring", "any CVEs / scan the images", "are we behind on Talos", "maturity report", "score the platform", "fix the missing probes / PDB", "remediate this", "open a PR for the drift". Always asks which cluster first (dev / staging / prod); defaults to dev and never the current kube context. Refuses any mutating command — read-only except remediation, which changes git (a PR), never the cluster.

Drive a running VS Code (this fork) from an agent — create files, run commands, type into the editor, go to a line, ctrl+click / go-to-definition, run terminal commands — by dropping action files into a spool directory. Use when the user says "drive vscode", "sendkeys to vscode", "have the agent control vscode", "run these actions in the editor", "open this file and jump to line N", or wires a session to control the IDE. Producer CLI is `vsc-sendkeys.cjs`; the listener is the bundled `vscode-sendkeys` extension. Full design in SENDKEYS_SPEC.md.

Build this Ghostty fork (debug or release), launch an instance with the GHOSTTY_SENDKEYS_DIR automation-testing watcher enabled, and inject synthetic keystrokes into it via a spool directory or the sendkeys.js CLI. Use for automation testing of this repo's fork, not upstream Ghostty.

Restore prior work context for the current folder using the crossmem CLI — it pulls the last real session from local Claude Code, Codex, Devin, Copilot (VS Code and CLI), and OpenCode histories. Use this FIRST, before listing or reading files, whenever the user says any of: pick up where I left off, resume, continue where I was, what was I doing here, where did I leave off, catch me up on this folder, load context, load my last session, resume from another agent/tool, or import Claude/Codex/Devin/Copilot/OpenCode memory.

Manage desktop windows and monitors on macOS, Windows, and Linux. List open windows, list monitors (with which one is ACTIVE / has the FOCUSED window), move a window into a predefined zone (1A, 1B, 2A..2D), an N:M split, or absolute / monitor-relative coordinates, resize a window in place, focus a window, **bulk-apply a layout of many windows in one call (`windowctl batch`)**, and request the macOS Accessibility permission needed for move / resize / focus / batch. Trigger on: list windows, show open windows, what windows do I have, list monitors, show displays, which monitor is active, where's my cursor, which monitor has focus, move chrome to the left half, snap terminal to the right half, put slack on monitor 2, place editor in the top-right quarter, three-way / N-way split, send to external display, resize chrome to 900x700, make this window smaller, focus jira, raise window, bring app to front, apply my layout, save my arrangement, restore work mode, batch place windows, request / check accessibility perm