cobusgreyling

Read loop-constraints.md at the start of every run and enforce every rule. This skill runs BEFORE triage or any action skill. Constraints are binding.

Scan merged PRs and commits since a given reference, extract titles, labels, types, and signals. Produces structured input for release notes drafting.

Classify CI failures — distinguish clear regressions from infra flakes and security-test failures. Produces structured failure reports.

Scan package manifests and lockfiles for outdated and vulnerable dependencies. Classify by severity and update type.

Scan open issues and discussions, deduplicate, prioritize, and propose labels. Provides a clean actionable queue.

Scan recent merges to main for tech debt, TODOs, debug code, and small cleanup opportunities. Produces a prioritized fix list.

Watch open PRs, check CI status, review staleness, merge conflicts, and unanswered review comments. Produces a prioritized watchlist.

Triage recent changes, CI failures, issues, and conversations. Produces a concise, actionable findings report suitable for a loop to consume. Writes structured output to a state file or issue tracker.

Manage what enters and stays in the context window — pruning, compaction, summary fidelity, ordering — so the agent stays coherent on long runs without inflating cost. Use when the user is hitting context limits, running long agentic loops, paying for full-history replays, or asks "how do I keep context manageable?" / "the agent forgets things after N turns".

Design where, when, and how a human gates, reviews, or rescues an LLM agent — without turning the agent into a button labelled "approve". Use when the user is building an agent that takes irreversible actions or operates in regulated workflows and mentions human-in-the-loop, HITL, approval gate, escalation, review queue, oversight, or asks "when should a human approve this?" / "how do I add review without killing the agent's speed?".

Design and validate LLM-as-judge scoring — pairwise vs pointwise, bias correction, anchor calibration, and the cases where a judge is the wrong tool. Use when the user is building an eval, scoring open-ended outputs, or comparing model versions and mentions LLM-as-judge, model grader, pairwise comparison, position bias, length bias, judge calibration, meta-eval, or asks "how do I score open-ended responses?" / "is my LLM-judge biased?".

Design memory for an LLM agent — what to keep, where to keep it, and when memory hurts more than it helps. Use when the user is adding memory to an agent and mentions short-term memory, long-term memory, episodic, semantic, conversation history, summary memory, vector memory, memory store, mem0, Letta, MemGPT, or asks "should this agent remember?" / "why is the agent recalling the wrong thing?".

Pick the right model per call, not per project — route Opus/Sonnet/Haiku, GPT-5/4o/mini, Gemini Pro/Flash by task, and cut cost without losing quality. Use when the user is choosing model tiers, building a router, or debating Opus-only vs mixed-tier deployments and mentions model selection, model router, cascade, fallback, cheap-first, draft-then-verify, or asks "which model should I use?" / "do I need Opus for this?".

Decide when to split work across multiple agents vs one agent with tools, and design the handoffs when you do. Use when the user is sketching a multi-agent system or debugging one, and mentions handoff, delegation, supervisor, swarm, crew, sub-agent, agent-to-agent, A2A, manager-worker, team of agents, or asks "should I split this into multiple agents?" / "why do my agents talk forever and never finish?".

Defend an LLM agent against prompt injection — direct, indirect, tool-result, and document-borne. Use when the user is building an agent that reads untrusted content (web pages, emails, documents, tool outputs) or exposes user-provided text to a downstream agent, and mentions prompt injection, indirect injection, jailbreak via document, tool-result injection, untrusted input, instruction override, or asks "how do I stop the agent from following injected instructions?" / "is RAG safe from injection?".

Get reliable structured output (JSON, typed objects) out of an LLM without regex repair, retry loops, or silent corruption. Use when the user is parsing model output, fighting malformed JSON, comparing JSON mode vs function calling vs structured outputs, or asks "why does the model keep breaking my schema?" / "how do I force valid JSON?".

Perform focused architecture and modularity reviews. Identify boundary violations, god objects, tangled dependencies, and propose the smallest reversible improvements. Use the repo-graph MCP when available. Trigger with "architecture review", "check the boundaries", "is this modular enough?".

Enforce conventional commits, clean branch strategy, small focused diffs, and PR-ready history. Use before any commit or when preparing work for review. Triggers: "commit this cleanly", "prepare for PR", "fix the git history", "use git-discipline".

Discover available MCP servers, evaluate their fit, install and configure them safely, and then use them effectively within sessions. Use when the user needs new capabilities (browser control, advanced git, observability, etc.) or when existing MCPs are under-utilized. Triggers: "set up MCPs", "install the right tools", "add an MCP for...", "what MCPs do we need?"

For any non-trivial task, start in Plan Mode, produce a clear step-by-step plan with risks and verification method, and wait for explicit approval before any edits. Use when the user says "plan this", "use plan mode", "don't edit yet", or describes a complex feature/refactor/migration.

Perform focused security reviews on code, diffs, hooks, and MCP servers. Look for auth issues, injection, secret leakage, unsafe tool/MCP usage, and hook risks. Use for any change involving external input, credentials, or agent capabilities. Triggers: "security review", "audit this diff", "check for vulnerabilities".

Run parallel independent subagents on the same investigation or design question, collect structured outputs, then synthesize the best elements or score approaches. Use for complex research, architecture decisions, or when multiple valid strategies exist. Trigger with "run an arena", "parallel subagents", "compare approaches", "investigate from multiple angles".

Apply rigorous test-first development. Write failing tests (or characterization tests) that capture desired behavior before implementing. Maintain high test coverage on changed code. Use for new features, refactors, and bug fixes where regression protection matters. Triggers: "TDD this", "test first", "write the test then implement", "add characterization tests".

Example skill that demonstrates good frontmatter. Used in tests.

Perform a thorough Grok Build readiness audit on a project directory. Detects presence and quality of AGENTS.md, .grok/skills, test setup, CI configuration, git hygiene, and ignore files. Produces a 0-100 score plus actionable recommendations. Use when asked to "audit", "gbs audit", "check grok readiness", "is this project agent-friendly", "grok build readiness", or when building/using the gbs audit command.

Calculate and explain a "Grok Readiness Score" (0-100) for a project based on concrete signals: quality of AGENTS.md / project rules, quantity and validity of skills, test/CI presence, git hygiene, ignore configuration, and verification culture. Use when asked to "score", "readiness score", "grok score", "how agent friendly is this", or as part of an audit workflow or the gbs audit implementation.

Create high-quality, production-ready Grok Build skills (SKILL.md files). Generates correct YAML frontmatter (name, description with specific triggers, when-to-use, argument-hint), clear numbered steps, tool references, and principles. Use when the user says "create a skill", "write a skill for...", "gbs scaffold skill", "/create-skill", "/skillify", or when implementing the scaffold command for skills.