dfrysinger-skills

Read websites that need a login. Use when a URL is behind SSO/MFA/auth that the agent can't do itself — internal repos, wikis, dashboards, vendor portals. User logs in once, agent browses after.

Guardrails for agent-initiated, externally-visible GitHub writes — filing issues, opening pull requests, posting comments, and pushing to repos the user does not own. Use when about to create or mutate anything visible to other people on GitHub (especially upstream/third-party repos), or when an offered menu option bundles such an action.

Write and review the durable work order for a systemic or critical change before code exists. Use when asked for a design or architecture document for work that alters shared state, persistence, a public contract, cross-component architecture, security, or a fail-closed boundary, or when `development-loop` triages work as more than bounded. Send bounded fixes and small features to `development-loop`.

Develop and ship one non-trivial code change through a risk-sized loop — establish the failure, triage the blast radius, prove runtime behavior with machine-validated receipts, review, and land. Use when fixing bugs, building features, refactoring, changing UI, or changing apps, services, agent workflows, pipelines, or SDKs; invoke it before claiming runtime work is ready, opening or updating a PR, or landing. When triage finds shared state, persistence, public contracts, cross-component architecture, security, or fail-closed boundaries, invoke `design-doc` before coding.

Bound dual review by evidence, scope, material risk, and a fixed round budget. Runs latest Claude Opus and latest non-mini/non-codex GPT in parallel, with finding-level verification only for disputed blockers. Use when a non-trivial diff needs review before landing, or a systemic or critical design and its guards need pre-implementation review.

Explain work to someone who steers it but never reads the code and may not know or remember how the system works. Use proactively before summarizing what you built, finished, or found; before presenting options or a recommendation; when handing off or catching someone up on a project; whenever the user asks you to explain, teach, walk through something, or show how it works in practice; and when asking the user a clarifying question.

Recover lost or displaced Git work after accidental reset, checkout, rebase, stash, or branch mistakes by using reflog, dangling commits, and targeted restore. Use when uncommitted or recently committed changes appear lost and the goal is safe recovery without rewriting unrelated history.

GitHub REST + GraphQL API gotchas that look like your bug but aren't. Use when building or debugging code that calls api.github.com or api.*.ghe.com, wiring GitHub sign-in (OAuth App, GitHub App, device-code), enumerating enterprises/orgs/repos, or hitting "Resource limits for this query exceeded", null enterprise/org nodes, or scope/SSO errors.

Relentless interview that walks every branch of a plan until it is settled.

Compile a codebase's prose architecture rules into deterministic CI guards — architecture fitness functions — so an agent cannot quietly break a global invariant while every local test stays green. Use when a fast-moving or agent-written codebase is drifting from its intended architecture, when you want the rules in AGENTS.md or a design spec actually enforced, or when reviewing whether a refactor preserved a system's global contracts.

Compact the current conversation into a structured handoff document for another agent — or for a fresh instance of yourself — to pick up where this one left off. Use when the user says "hand off to <name>", "do a handoff", "summarize this session for another agent", wants to pass work to a different Copilot CLI session, or when the `self-compact` rule calls for a self-handoff through native compaction or `/new`. If a recipient is named (e.g., "hand off to juliett"), the skill auto-delivers via the mailbox skill after writing the doc.

Screenshot or click a macOS app window sitting behind the user's foreground window, without stealing focus or moving the cursor. Use when verifying a macOS GUI end to end while sharing the user's Mac, when another skill needs a capture of an occluded native or WebView app window, or when a click must reach a background app without disrupting whatever the user is doing.

Query a local macOS Photos library with osxphotos and import screenshots or image files into a Photos album using the local import helper. Use when you need to read Photos metadata, verify an album exists, or save app-test screenshots into Photos for iPhone/iCloud review.

Hand a file or message to another named Copilot CLI, Claude Code, or Codex CLI session. Use when this session has produced a deliverable another named agent should pick up, or the user says to send something to an agent by name (e.g. "send this to juliett"). Copilot delivery uses a portable Node watcher and SDK session names; macOS tmux remains a compatibility identity and the guarded fallback for Claude and Codex.

Prototype what a screen should look like in throwaway HTML before it is built. Use when work introduces an entirely new screen or sweepingly changes existing ones, and when `scout` routes a chosen option here. Send contained or component-level UI changes to `development-loop` instead.

Rotate a long-lived Copilot CLI session into a fresh one that rebuilds context by reading the old session's plan, checkpoints, todos, and transcript off disk. Use when the user says "rotate this session" or "start fresh but keep where we are", or when a session has grown slow to load or fails to load at all.

Rubber-duck a plan, trace, or implementation past a model from a different family and get a second opinion back. Use when a bounded solution is ambiguous, a failure will not surface, the first instance of a new idiom needs confirming, repeated variations keep failing, or a reversible decision would otherwise go to the user. Reach for it whenever another skill calls for a rubber-duck pass; on Copilot CLI it defers to the built-in agent.

Scout how to achieve something in a project before any code is written. Use when the user asks how to achieve, add, or support something in a project, asks whether it is possible or how hard it would be, asks what the best or standard approach is, or asks who already ships it and how. Use when another skill needs the problem explored before the change is defined.

Keep secrets (tokens, API keys, PATs, passwords, private keys, session cookies) out of the chat transcript and out of git. Use when reading a token, configuring auth, writing a script that authenticates, asking a user to authenticate, or committing code that touches credentials.

Keep Copilot CLI working context lean while preserving decisions, active state, drafts, and session-bound resources through deliberate compaction. Use when a task finishes, work changes phase, a review round ends, tool history grows, the agent becomes stuck or repetitive, a durable work order enables a soft reset, or a long-lived session needs retirement.

Create reusable agent skills through a predictable authoring, validation, and review gate. Use when a user asks to make a new skill, turn a recurring procedure into a skill, or publish a skill without depending on a lifecycle or curation system.

Keep a long, unattended Copilot CLI run on course. Arm an `/every` charter re-brief that restores the run's operating rules after compaction, then hand off an optional autopilot objective through the session-inbox SDK extension and native command queue. Use when starting a long autopilot or `/goal` run against a plan doc, sharpening its objective, or preventing drift across context compactions.

Capture and inspect visual evidence for a current running UI candidate, rejecting blank, stale, unopened, or unfalsifiable screenshots. Use for every user-visible UI fix or feature before review, PR creation, landing, or a success claim; also use when a change note claims a screenshot or another skill needs runtime proof a human can see. Fixes may use paired screenshots, while a new user-facing UX also invokes walkthrough-video for its PR movie. This skill supplies the visual section of development-loop's machine-validated live-proof receipt; it does not replace full interaction proof.

Record, inspect, and attach a candidate-bound video walkthrough for a new user-facing UX before its pull request is created or updated. Use when a change adds a screen, control, interaction, or materially new visual journey that reviewers need to see in motion. Fixes may use visual-proof's paired screenshots instead unless the user asks for video.

The vocabulary and principles that make a skill predictable — invocation, information hierarchy, pruning, leading words, and failure modes. Use when writing, editing, reviewing, or critiquing a skill, or deciding how to structure or split one.