Cinematic Upgrade
Plan premium motion and UI polish upgrades without mutating code prematurely.
Collection
First-party orchestration, safety, evaluation, and workflow packs built to push beyond generic prompt libraries.
20 skills in this lane
Plan premium motion and UI polish upgrades without mutating code prematurely.
Design hard prompt boundaries, tool gating, and context sanitization so indirect prompt injection has fewer places to land.
Shape approvals, status surfaces, and human handoff moments so advanced agent workflows stay legible and trustworthy.
Allocate work across local, cloud, and premium models so teams maximize capability coverage per dollar and per latency budget.
Compare two API or schema revisions and surface behavioral breakage, migration steps, and compatibility risk.
Reshape sprawling repositories and briefs into stable context lanes, memory checkpoints, and retrieval boundaries for long-horizon agent work.
Tune provider routing policy for quality, cost ceilings, and fallback behavior across multiple model subscriptions.
Detect when implemented UI patterns drift away from the intended design language, tokens, or motion rules.
Keep README, docs, examples, and CLI behavior aligned so public repos do not drift out of date.
Investigate a failed delegation by replaying context, isolating likely root causes, and proposing the smallest reliable recovery path.
Build eval cases that expose fabricated citations, brittle reasoning chains, and ungrounded tool usage before they hit real workflows.
Review an MCP server for prompt-exfiltration, shell abuse, overbroad tool scope, and unsafe logging.
Tune timeout, retry, and concurrency budgets across multi-model routes so orchestration stays fast without silent quality collapse.
Turn a product or engineering spec into bounded sub-tasks, ownership lanes, and integration checkpoints for multiple agents.
Turn advanced agent workflows into reusable skill and prompt packs with trigger rules, output contracts, and maintainable docs.
Audit a repo for secrets, personal paths, client-specific references, and OSS-readiness gaps before publishing.
Improve chunking, metadata, and ranking design so agent answers stay grounded under larger repositories and longer tasks.
Derive a high-signal regression matrix from changed code, user risk, and likely failure surfaces.
Build a practical threat model for agent workflows, MCP tools, provider routing, and persisted run data.
Normalize tool input and output schemas so multi-agent tool calling becomes more reliable, typed, and retry-safe.