ericchansen

Generate or edit images with GPT-Image-2 in Azure AI Foundry via the generate_image tool. Triggers: generate image, edit image, reference image, create image, make an image, draw, illustrate, diagram, visual, mockup, logo, concept art.

Reverse-engineer undocumented web APIs by intercepting browser network traffic via CDP. Use when you need to automate a web app that has no public API, or when the documented API doesn't work but the UI does. Captures exact payloads and request headers from live browser sessions. Triggers: "reverse engineer", "intercept", "capture the request", "sniff the API", "what does the UI actually send", "replay the request", "hack the API".

Launch Microsoft Edge with remote debugging on a dedicated debug profile, then control it via Chrome DevTools Protocol (CDP). Use when you need the signed-in work account's SSO session, managed-device auth (Conditional Access), or to extract a bearer token from an authenticated page for server-side API calls.

Run a structured closeout review on local git diffs, branches, commits, or PR changes before commit or ship. Verifies findings against the real code, fixes at the right ownership boundary, and reruns until clean. Use when the user says "git-code-review", "review my code", "code review", "closeout review", "review this branch/PR", "review before I push", or asks for a second pass after non-trivial edits.

Execute git commit with conventional commit message analysis, intelligent staging, and message generation. Use when user asks to commit changes, create a git commit, push code, create a PR, or mentions "/commit". Supports: (1) Auto-detecting type and scope from changes, (2) Generating conventional commit messages from diff, (3) Interactive commit with optional type/scope/description overrides, (4) Intelligent file staging for logical grouping, (5) Respecting repository contribution guidelines

Use for any GitHub CLI command that writes PR, issue, or comment content. Prevents shell encoding corruption by writing markdown to a body file first and enforces content quality: no machine-specific paths, generic repro steps, and structured sections. Triggers: github-writer, gh-writer, create PR, open pull request, edit PR, create issue, edit issue, comment on issue, push and create PR, gh pr, gh issue, file a bug, open a bug.

Diagnose and fix Azure deployments — investigate repos, CI/CD pipelines, Azure subscriptions, and service health. Use when user says "is my app working", "check my deployment", "why is my site down", "fix my Azure app", "check health", "investigate deployment", "what's broken", "debug my Azure deployment", "check CI/CD", "check my demo sub", or any request to diagnose, troubleshoot, or repair a deployed Azure application. Covers Container Apps, App Service, PostgreSQL, ACR, GitHub Actions, and general Azure resource health.

Post-merge git cleanup — checkout main/master, pull latest, delete merged local branches, prune remote-tracking branches, verify clean working state. Use when user says "clean up", "rebased and merged", "merged, clean up", "cleanup branches", "back to main", or any variant of post-merge housekeeping.

Use when creating or editing GitHub PRs/issues/comments with `gh`; write markdown bodies to temp files and pass `--body-file` instead of inline `--body` to avoid shell quoting and markdown corruption across Windows, WSL2, and Linux.

Use when addressing GitHub PR review feedback; gather all comments and requested changes, classify them, fix valid feedback, push back with evidence when appropriate, and reply to every thread.

Use when starting Git development work; keep main/master pristine by creating a feature worktree from the latest remote branch before editing, committing, pushing, or opening PRs.

MANDATORY before launching any q2mm batch optimization run >30 minutes (benchmark.py, multi-system convergence runs, from-QFUERZA runs, etc.). Forces success-spec write, optimizer config sanity check, FIRST-system audit gate before launching the rest, and post-batch validation. Use when user says "run benchmarks", "regenerate convergence results", "run all systems", "QFUERZA recovery", or any variant of launching a multi-system q2mm optimization batch.

MANDATORY before writing any q2mm benchmark analysis, comparison, or validation document. Forces restating the user question verbatim, listing deliverable artifacts (table headers, figure captions), mapping each artifact back to the question, and flagging mismatches BEFORE writing the doc. Use when about to create or rewrite docs/benchmarks/*.md, when about to produce a comparison page, when summarizing benchmark results, or when packaging a PR with new results.

Build graphical, web-native presentation decks from rough notes through collaborative art direction, original AI graphics, and visually inspected Playwright previews. Use for decks, slides, presentations, slide prototypes, or reskinning an existing deck.

Use GitHub Copilot CLI for substantial repository work inside one bounded native OpenClaw child.

Versioned quota-estimation methodology for Wide World Importers customers. Encodes the exact, deterministic growth-rate model the quota estimator implements so the same logic can be reused across the Copilot CLI, a Foundry agent system prompt, or a Databricks Supervisor tool.

Fix, validate, and generate visual text content — diagrams, tables, boxes, and bars. Use when the user asks to "fix diagrams", "fix tables", "fix alignment", "validate mermaid", "check ascii art", "fix emoji spacing", "realign table", "fix box drawing", "fix progress bars", or any request about broken visual alignment in text content. Also use proactively after generating markdown tables, ASCII art, or diagrams to validate they are correct.