kenhuangus

Use before any approved Mythos-style defensive cybersecurity run, or when reviewing a Fable/Mythos security workflow. Requires authorization, bounded targets, private evidence handling, independent validation, and a patch/report owner before the model explores vulnerabilities.

Stop Claude Fable 5 from over-planning, re-deriving settled facts, or listing options it won't pursue. Use in interactive sessions where turns feel slow on simple asks, or agentic sessions at high/xhigh effort where the model gathers more context than the task needs.

Keep Claude Fable 5 running to completion in unattended pipelines -- no stalling on a stated intent without the matching tool call, no mid-run permission questions, no self-imposed session splits from context-budget worry. Use for overnight runs, scheduled jobs, and CI agents where nobody is watching to type "continue".

Pick and re-evaluate the output_config.effort level for a Claude Fable 5 workload. Use when designing a pipeline that mixes routine and hard tasks, when a call feels slow or expensive for what it's doing, or when porting an Opus 4.7/4.8 harness that defaulted to xhigh.

Bind every progress claim Claude Fable 5 makes during a long or autonomous run to an actual tool result from that session. Use for multi-hour agents, overnight pipelines, CI runs, or any workflow where a status update has previously claimed work that wasn't actually done.

Maintain a file-based lesson memory that Claude Fable 5 reads and updates across sessions. Use for recurring agents (daily jobs, long projects, team assistants) where the same correction keeps being re-made, or when bootstrapping a new agent from past session history.

Keep Claude Fable 5's diffs mapped one-to-one to the request -- no unrequested refactors, speculative abstractions, or defensive error handling for states that can't occur. Use in code review and any coding session where the diff comes back larger than the ask, especially at high or xhigh effort.

Make Claude Fable 5's final report readable to someone who saw none of the work. Use for long agentic sessions, overnight runs, and multi-tool tasks whenever summaries come back as arrow-chain shorthand, invented abbreviations, or references to reasoning the reader never saw.

Use when Codex reviews Claude Code-authored or other-agent-authored changes and verifies ownership claims before integration.

Use when setting up or enforcing a compound engineering loop across coding, writing, research, or documentation projects with Codex, Claude Code, and other agent runtimes.

Search persistent memory for relevant notes from past sessions — prior decisions, conventions, gotchas, and how similar work was handled. Recall is cheap and fail-open, so call it rather than guess: before starting a task, editing a file, fixing a failing test, or choosing an approach — and whenever the user refers back ("what did we decide about X", "how did we do Z").

Drive the user's already-open, logged-in Google Chrome on Windows via pywinauto UIA + chrome.exe URL open + PrintWindow screenshots. Use when cursor-ide-browser MCP tabs are empty, CDP is blocked, or the agent must reuse an existing Chrome profile for Render, Devpost, dashboards, or form submit without launching an anonymous browser.

Context-engineering workflow for non-trivial coding work — building a new feature or project, fixing/finishing a PR, refactoring, debugging, or implementing any change in a real or large codebase where an agent tends to lose the plot as its context window fills. Use it to keep output quality high on big tasks: it runs Research → Plan → (gate the plan) → Implement, holds context in the 40–60% "smart zone" and compacts at the 60% "smart/dumb line", and reviews the plan before any code is written. Integrates the Glassbox tools (context HUD, Plan-Gate, autopilot tool-output curation, auto-compaction) when that repo is available; otherwise applies the same practices by hand. Trigger on "implement X", "build this", "fix this PR", "add a feature", "refactor", or any multi-step change where correctness in a sizable codebase matters.

Turn a JSON list of scenes into a narrated 1080p MP4 walkthrough. Renders polished dark slides (title, bullets, code with syntax highlighting, requirement/feature tables, architecture diagrams, and embedded screenshots), generates voiceover with Deepgram Aura-2 TTS, and montages everything with ffmpeg — one still per scene held for the length of its narration. Use when a user asks to "make a demo video", "record a walkthrough", "produce a submission/architecture video", "narrate these screenshots/slides", or turn a project into a short explainer with voiceover.

Produce a narrated demo/walkthrough video from ordered screenshots plus Deepgram TTS narration, montaged with ffmpeg. Use when the user asks to "make a demo video", "record a walkthrough", "produce a submission video", "narrate screenshots", or capture an app end-to-end (every screen/action) into an MP4 with voiceover. Covers Android (adb screencap), web (headless browser), and HTML slide frames for architecture / submission / closing sections.