Silex-Research
OrganizationKeep AI coding agents honest. DontPanic is a vendor-neutral meta-harness that makes agents work to a locked plan, has a second model check the work, and asks a human before risky steps — with circuit breakers, a local operator console, and a full evidence trail.
Categories
Indexed Skills (24)
changelog
Render a markdown changelog for a git revision range, grouped by Conventional Commits prefix
cost-guard
Compare observed app-level run-rate to per-app and per-model budgets, emit INBOX warnings on breach
cost-model
Project month-end and next-month spend per app and per LLM provider from existing cost state
eval-harness
Define and run evaluation criteria against code changes. Supports code-based (deterministic), model-based (LLM-as-judge), and human (flag for review) graders. Pairs with autoresearch for metric-driven optimization.
printing-press-adapter
Wrap an external OpenAPI-shaped service as a DontPanic-governed CLI + MCP adapter using CLI Printing Press
revenue-check
Pull product revenue per app from Firestore (or fixtures) and produce a cash-flow report
github-pr
Fetch, preview, merge, and test GitHub PRs locally. Great for trying upstream PRs before they're merged.
agent-browser
Specialized browser automation for agent workflows with structured data extraction, multi-step workflows, error recovery, and state persistence.
autoresearch
Autonomous experiment loop inspired by Karpathy's autoresearch. Iteratively modifies code, runs evaluation, measures a metric, and keeps or discards changes using git. Use when optimizing code against a measurable target (test pass rate, performance, bundle size, model quality, etc).
browser-use
Browser automation for scraping, form submission, screenshots, and web interaction via CDP or Playwright.
git-essentials
Essential Git commands and workflows for version control, branching, and collaboration.
kronos-agent
Financial time-series forecasting using the Kronos foundation model (MIT, NeoQuasar). Takes OHLC candles, returns predicted future candles with configurable horizon. Infrastructure skill — called by trader agents or scheduled ingestion, not directly invoked by users or the model.
migration-planner
Plan and track cross-platform or cross-service migrations (Supabase to Firebase, single to multi-tenant, schema migrations, etc). Generates rollback-safe migration plans with state tracking.
pr-reviewer
Automated GitHub PR code review with diff analysis, lint integration, and structured reports. Use when reviewing pull requests, checking for security issues, error handling gaps, test coverage, or code style problems. Supports Go, Python, and JavaScript/TypeScript. Requires `gh` CLI authenticated with repo access.
product-health-agent
Managed agent that analyzes production app engagement and operational health. Reads Firestore aggregates, Cloud Logging, and Cloud Monitoring; mounts source repos read-only; writes structured insights to a workspace Firestore collection. Designed for daily digests and on-demand deep dives on the Glam and SpinDineSwift production apps. Read-only by design.
prompt-optimizer
Optimize LLM prompt templates against an eval harness. Works with autoresearch for overnight prompt optimization. For projects that call LLMs (try-on prompts, chatProxy, MCP tools).
remotion-video-toolkit
Complete toolkit for programmatic video creation with Remotion + React. Covers animations, timing, rendering (CLI/Node.js/Lambda/Cloud Run), captions, 3D, charts, text effects, transitions, and media handling. Use when writing Remotion code, building video generation pipelines, or creating data-driven video templates.
skill-creator
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
yahoo-finance
Get stock prices, quotes, fundamentals, earnings, options, dividends, and analyst ratings using Yahoo Finance. Uses yfinance library - no API key required.
brainstorm-gate
Design-first gate that prevents premature implementation. Use when the task is ambiguous, has multiple valid approaches, or the user hasn't specified how something should work. Forces clarification before any code is written.
plan-artifacts
Structured artifact trail for plans, brainstorms, and solutions. Triggers when creating plans or documenting learnings. New plans default to directory format with schema-validated plan.md + features.json (agent-conventions v1.0). Single-file v0 plans remain readable.
subagent-driven-dev
Structured pattern for dispatching focused subagents per task with two-stage review. Use when implementing a plan with 3+ independent steps, or when parallelizable work spans multiple files/domains.
token-budget-tracker
Track AI model usage, token costs, and optimize spend across agents
worktree-isolation
Git worktree-based task isolation. Use when making changes that should not affect the main working tree until verified, or when running parallel development tasks that might conflict.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.