Browse reusable Agent Skills, each with a clear purpose and practical guidance.
Remove an agent from the system — deletes the agent file, cleans up all registry entries, removes cross-references, and updates documentation. Use when the user says "remove the X agent", "delete X-review", "retire the X role", or "we no longer need X". Handles both team agents and review agents. Always confirms before deleting.
Conventions, anti-patterns, and meta-patterns for writing skills (and the shared agent/skill philosophy). Use when creating or editing a SKILL.md file, or when reviewing the agent-vs-skill separation. For the procedural workflow that generates a new agent file, use the agent-create skill (invoked by /agent-add).
Recommend whether a plugin capability should be a markdown (LLM-interpreted) unit or a deterministic script. Use when designing a new agent/skill ("should this be markdown or a script?"), or when auditing an existing agent/skill file for a type mismatch. Accepts either a prose use-case description (forward- looking, new unit) or a path to an existing agent/skill file (retrospective).
Report on skill and agent usage data from metrics/artifact-usage.json, classifying each artifact as active, stale (>= 30 days unused), or an archive candidate (>= 90 days unused). Proposes CLAUDE.md overrides for stale artifacts and exclusions for archive candidates. Pinned skills are always exempt. Use when the user asks to "review artifact lifecycle", "find stale skills", or "/artifact-lifecycle".
Decide which agents and skills to load for a given task. Use at the start of every task to select the minimum viable context load, calculate the token budget, and stay below the 40% utilization ceiling.
Run a Claude Code skill or command headlessly in an isolated subprocess — fresh session id, clean HOME and config dir, scrubbed env, JSON result, timeout. Use for scripted one-shot invocations and benchmark-harness cases (e.g. running /code-review once per case), to run an isolated claude -p, or to avoid a nested run reusing the parent Remote session identity or tool surface. Trigger phrases include "run a skill headlessly", "isolated claude -p", "benchmark harness invocation", "run /code-review headlessly", "run it once per case", and "avoid nested session reuse".
Approval gates, intervention commands, and transparency requirements. Use to classify any agent action as autonomous/notify/approve, respond to override/pause/stop commands, or structure a plan review before the implementation phase begins.
Scaffold the eval directory structure for a plugin's review agents and advisory skills. Use after creating a plugin or adding its first review agent, or when the user says "set up evals for this plugin", "init plugin evals", "scaffold eval fixtures", or "create the eval harness for <plugin>". Creates the fixtures/ and expected/ dirs plus a README describing the grading contract.
Assess Power BI semantic models for Copilot, Fabric data agent, and natural-language Q&A readiness. Use when reviewing whether a model has clear business terminology, unambiguous metrics, usable date defaults, focused field exposure, descriptions, AI instructions, AI data schema recommendations, verified-answer candidates, or natural-language validation tests.
Prepare and recover Power BI SemanticOps MCP sessions. Use when starting a session, connecting to a model, checking current connection, applying preferences, selecting the right tool family, troubleshooting empty results or stale metadata, or fixing malformed tool arguments and bulk payloads.