skills

Orchestrate iterative, role-guided, sub-agent-heavy work on software, documents, contracts, plans, workflows, policies, designs, images, or other directed targets from a supplied work source or whole-target Roast queue through create-review-verify-repair loops, Evidence Gate checks, remediation, risk review, cleanup, and readiness reporting. Use for development and release hardening or any substantial work that should converge through independent critique and evidence; requires explicit sub-agent approval or explicit degraded local-only confirmation before proceeding.

Locate prior coding agent conversations by topic, project, date, file path, command, error text, or remembered fragment; verify the strongest candidate across authorized history sources; and return its title, session id or resume handle, evidence, coverage, and honest alternatives.

Deeply review available coding-agent collaboration history, then produce an evidence-bound teaching report that helps the caller communicate better with coding agents through prompting, context-setting, boundaries, corrections, verification requests, handoffs, durable agent instructions, and reusable context/tooling infrastructure. Use for full-history retrospectives at any user skill level without inventing unsupported history or substituting current-conversation advice.

Repair commentary-over-background audio or video, including single-program audio, into a listener-first REMIX-VOICEOVER output using deterministic measurement, macro-regime repair, verifier-derived component gates, source-container preservation, copied video, and untouched source media.

Accessibility audit of a codebase, interface, app, document workflow, or UI surface. Use when the caller wants project-specific accessibility research, standards baseline, color/contrast, typography, scaling, mobile/responsive behavior, keyboard/focus access, semantics, forms, motion, dynamic content, user-impact priorities, per-category scores, and a practical final grade with next steps.

Agent-readiness audit for websites, apps, APIs, tools, documentation, or codebases. Use when the caller wants research-first assessment of whether solution or coding agents can discover, understand, retrieve, act on, and verify authoritative information through web, docs, API, feed, MCP/OpenAPI, or tool surfaces, while keeping generic crawler SEO, rankings, and search-growth work separate.

Classic SEO readiness audit for websites, landing pages, documentation, local-business sites, ecommerce pages, or content libraries. Use when the caller wants evidence-backed assessment of crawl/index foundations, search intent, on-page metadata, content quality, internal linking, structured data, local/entity signals, performance/page experience, analytics evidence, competitive visibility, prioritized fixes, and clear separation from agent-readiness or ranking guarantees.

Toggleable eldritch narration mode for normal coding-agent work. Use only when the caller explicitly asks to activate or use eldritch, haunted, cosmic-horror, Lovecraftian, King in Yellow, Yellow Sign, madness, Zalgo, or corrupted narration; bare literary or topic mentions do not activate it. Keep applying it in-session after valid activation until the caller asks to turn it off, use plain text, disable glyphs, or return to technical-only writing.

Capture restart-safe handoff notes for unfinished work in a reachable HANDOFF.md file by default. Use when pausing, transferring, or resuming after context loss, token pressure, machine restart, session change, or any stop that needs goal, status, blockers, artifacts, and next steps preserved.

Recover interrupted or transferred coding work by locating accessible conversation history, task state, workspace evidence, handoffs, logs, and scratch artifacts; reconstruct past progress, current state, intended future plan, and a safe next action; and produce or execute a semantic continuation after crashes, usage or quota exhaustion, context loss, session changes, agent handoffs, or work continued from another coding agent.

Strictly review any directed target, including a codebase, PR, module, release candidate, document, contract, plan, workflow, policy, design, image, or other artifact, through caller-specified or context-appropriate lenses. Use for blunt, serious, evidence-backed findings about security, correctness, architecture, tests, release risk, quality, coherence, usability, maintainability, clarity, or fitness for purpose, with severity, fix directions, and a grade.