SShadowS

Compile and deploy AL code to a BC environment. Handles single-app and multi-app deploys with topological dependency ordering. Use when (1) AL code was changed and needs deploying, (2) the user asks to compile and publish, (3) a test fix needs deploying before re-running tests, or (4) a fresh environment needs all apps deployed. Invoke continia-env-setup first if no envId is available.

Install external dependencies on a BC environment and download symbol packages for AL compilation. Use when (1) compilation fails with missing symbol or reference errors, (2) a fresh environment needs base apps installed before deploying, (3) the user asks to install or update dependencies, or (4) .alpackages is empty or outdated.

Find or start a running BC developer environment and configure launch.json. Use when (1) you need a running environment for deploy, test, or dependency operations, (2) a session was just started and no environment is configured, (3) a command fails because the environment is stopped, or (4) the user asks to set up or connect to an environment.

Run AL tests on a BC environment and interpret results. Starts a test job via DemoPortal, waits for completion, and parses XML results with pass/fail status, error messages, and call stacks. Use when (1) the user asks to run tests, (2) a fix was deployed and needs verification, (3) regression testing is needed after code changes, or (4) a bug report references a test codeunit.

Turn an approved Azure DevOps pull request into hard CentralGauge benchmark trap-tasks. Use when authoring new AL tasks from real merged PRs / work items, when a task should probe BC runtime semantics (not just syntax), or when the existing suite is too easy. Mines the non-obvious, comment-justified decisions in a PR (the traps a syntax-only model gets wrong), distills each into a self-contained or base-app-faithful task, and ships it ONLY after a discrimination probe proves a correct solution passes and a naive one fails on a real container.

Refresh the CentralGauge task taxonomy — the 9 groups + ~72 facet tags that power the site's task discoverability filter and per-category model analysis. Use when tasks are added/changed/removed in tasks/, when the taxonomy feels stale or under-covers a workflow, or on a periodic cadence. The taxonomy is UI/analysis-only metadata, fully decoupled from the task_set hash, so refreshing it NEVER triggers a re-bench. Produces site/catalog/task-categories.yml and pushes it to prod via `sync-taxonomy --apply`.

Style guide and conventions for writing CentralGauge blog posts. Use when creating or editing blog posts, devlogs, announcements, or any public-facing written content about CentralGauge benchmark results, updates, or findings. Ensures consistent voice, tone, and formatting across all posts.

Use when asked to find recurring Bash patterns in agent telemetry (pr-reviewer OR pipeline stages — coder, planner, analyzer, etc.) that should become helper scripts (to cut turns + tokens), OR to measure uptake of a recently-deployed helper (before/after comparison).

Runbook for bumping the Claude Agent SDK and the container's Claude Code CLI in DevOpsWorker, then reconciling every AL-LSP hack, verifying agents still see the LSP tool, and rebuilding images. Use this whenever upgrading @anthropic-ai/claude-agent-sdk or @anthropic-ai/claude-code, when the user says "uplift", "bump the SDK", "update Claude Code", "pull the newest dependency", or mentions a new Claude Code / SDK build fixing an LSP bug. Claude Code ships frequently, so reach for this skill on any routine dependency uplift of these packages — even when the user doesn't name every step.

Interactively debug AL code on a Business Central server with the bc-dev-mcp tools — attach, breakpoints, break-on-error, step, inspect variables and watches during a test run. Use when asked to debug AL code, find why an AL test fails, or inspect live state on a BC server.

Run AL tests against a Business Central dev endpoint with the bc-dev-mcp tools — discover test codeunits, run them with structured per-method results, optionally collect code coverage. Use when asked to run, re-run, or triage AL tests on a BC server.

Release `business-central-mcp` to npm — bumps version, builds, tests, dry-run-inspects the tarball, publishes, and pushes the tag. Use this whenever the user wants to ship a new version, publish to npm, cut a release, tag a version, or asks "how do we release this". Also use when the user has finished a body of work and asks what's next, since release is often the natural next step after merging features. Walks through preflight checks, semver decision, and post-publish smoke. Does NOT actually run `npm publish` without explicit confirmation — releases are irreversible.