pauli-hermes-agent

Pre-code system configuration for any AI agent. Detects model, applies token optimization, estimates input tokens, compresses context, calculates real-time savings (to the token), and tracks cost delta before coding starts. Works with all 11 models. Agent runs this BEFORE writing code—saves 75-95% tokens + cost on every project. Triggers: "before i code", "set up token optimization", "configure system", "what model am i", "estimate tokens", "how much will this save", "show me savings", "pre-code checklist", "token tracker", "cost me to code this"

Publish agent capabilities and accept Agent2Agent task requests through the local HTTP receiver.

Verify that a Vercel deployment URL displays a real page instead of a 404, auth wall, blank shell, or error page.

Verify and normalize GitHub webhook payloads for deployment automation. Use when a GitHub push, deployment_status, or workflow_run event arrives.

Iterate deployment visibility failures using only no-code-safe actions and stop when source-code changes are required.

Inventory GitHub repositories and Vercel projects, then map repos to deployments for a visibility sweep.

Link, deploy, inspect, and verify Vercel projects using Vercel CLI/API with token-based authentication.

Policy for deciding what enters memory, what stays transient, and what must be redacted.

Operating contract for Pauli agents that route tasks, preserve context, and respect scope.

Security policy for preventing credential leakage in files, prompts, logs, and exports.

Budget policy for choosing tools and limiting wasteful agent execution.

Core Pauli policy for autonomous execution with hard safety and verification guardrails.

OpenMontage™ — AI-powered end-to-end video production. Trigger when user says things like: "make a video", "create a montage", "produce a trailer", "edit this footage", "animated explainer", "podcast clip", "documentary", "cinematic trailer", "talking head video", "repurpose this content", "create a video from", "video about".

Write ML papers for NeurIPS/ICML/ICLR: design→submit.

Browser automation policy for Pauli workflows using browser-harness first, with CDP-based navigation, extraction, and dashboard control.

Coolify and Hostinger VPS deployment operations with staging-first safety gates and rollback planning.

Graph-backed codebase-to-course workflow for turning a mapped repo into a single-page interactive onboarding course.

FAL.ai provider adapter policy for dry-run generation planning, key presence checks, and cost-guarded invocation.

Flywheel and Ralphy orchestration lane for planning, bead decomposition, execution loops, and review checkpoints.

High-end UI refinement skill focused on precision, motion restraint, accessibility, and polished interaction states.

Secret inventory, secret health, and Infisical or env-source routing without exposing secret values.

Large-repo compression and context-map workflow that summarizes before deep reading while preserving source-of-truth discipline.

Reasoning harness for context normalization, expert routing, bounded iteration, and decision summaries without chain-of-thought leakage.

Open Design reference lane for high-signal UI direction, hierarchy, layout, and interaction clarity.

OpenMontage-oriented studio orchestration for briefs, scripts, shot lists, asset planning, and dry-run render workflows.

Supabase and pgvector-backed second-brain retrieval, writeback discipline, and search-first memory workflows.

Taste and visual-direction skill for deliberate, non-generic interfaces with strong hierarchy and editorial restraint.

Twilio voice status, webhook checks, transcript-oriented debugging, and allowlisted test-call procedures.

Vercel deployment triage and repair for build failures, 404s, routing issues, and environment drift.

Video ingestion and transcript workflow with yt-dlp, ffmpeg, caption extraction, and budget-aware analysis.

Zero-touch engineering execution loop for repo discovery, planning, implementation, testing, verification, and safe escalation.