llmctl-baseline

Plugin llmctl-baseline from marketplace llmctl-marketplace in siegenthalerroger/.llmctl-marketplace. Commands read from the repo's manifest — 12 days ago.

Skills (9)

• authoring-mermaid-diagrams

  Authors, repairs and validates Mermaid diagrams: picks the diagram type from what the reader must learn, writes the frontmatter config and accessibility header, and checks the source before it ships. ALWAYS invoke when asked to draw, diagram, chart or visualise something, to document an architecture

• batch-task-execution

  Guidelines for planning and executing batches of tasks from todo lists, backlogs, or multi-item requests. ALWAYS invoke when asked to work through a list of tasks, start multiple sub-agents in parallel, or tackle several items at once. Do not fan out sub-agents or start a multi-item batch without th

• complex-research

  Guidelines for delegating complex research tasks that need multi-source investigation, synthesis, and comprehensive analysis. ALWAYS invoke when research needs extensive web searches, documentation queries, repository exploration, or cross-referencing multiple sources — delegate to a researcher agen

• grill-me

  A relentless interview to sharpen a plan or design.

• grilling

  Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.

• humanizer

  Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyse

• meta-harness

  Authoring guide for the config an agent harness executes or installs: lifecycle hooks (*.hook.json, PreToolUse/PostToolUse/SessionStart matchers), MCP servers (apm.yml dependencies.mcp, stdio/http/sse transports, ${VAR} secrets) and plugin bundles (plugin.json, marketplace, apm pack). ALWAYS load be

• meta-steering

  Authoring guide for every steering file an agent harness reads: skills (SKILL.md), subagents (*.agent.md), instructions (*.instructions.md, applyTo/paths) and prompts (*.prompt.md) — frontmatter, description shape, discovery budgets, provenance. ALWAYS load before creating or editing one of those fi

• troubleshooting

  Behavioral rules for diagnosis and root-cause analysis: failures, tool quirks, command hangs, version-dependent behavior, and Docker container debugging (terminal state, script execution, entrypoint overrides, busybox differences). ALWAYS invoke when a command fails, hangs, or behaves unexpectedly,

Slash commands (2)

• /llmctl-baseline:reflect

  Reflects on the current conversation and folds the learnings back into the steering files that produced them, tracing a file to its real source when it arrived from a plugin marketplace rather than this repo. ALWAYS invoke when asked to self-improve, capture a lesson, or stop a mistake from recurrin

• /llmctl-baseline:setup-mcp

  Converts MCP server definitions into an apm.yml dependencies.mcp block with secrets externalized to ${VAR}. ALWAYS invoke when asked to add or wire an MCP server into this repo from pasted config or an mcp.json path. Do not generate the per-tool files (.vscode/mcp.json, .mcp.json, .codex/config.toml

Subagents (5)

• Executor (Broad)

  Mid-tier execution agent for well-specified work spanning multiple components, systems, or files — code, configuration, IaC, documentation, or specs. ALWAYS invoke for cross-cutting refactors, migrations, bulk rewrites, and multi-file changes that need sustained reasoning across a large surface. Do

• Executor (Focused)

  Fast, cost-efficient execution agent for well-specified, tightly-scoped work within a single component — code, configuration, IaC, documentation, or specs. ALWAYS invoke when the plan is clear, the task touches ≤5 files, and no cross-component reasoning is needed. Do not use for cross-cutting or mul

• Explore

  Fast read-only codebase exploration and Q&A subagent. ALWAYS invoke instead of manually chaining multiple search and file-reading operations — it keeps the main conversation uncluttered and is safe to call in parallel. Do not use to edit files; it is read-only. Specify thoroughness: quick, medium, o

• Plan

  Researches a goal, clarifies scope with the user, and produces a detailed multi-step implementation plan for executor agents. Selected by the user before implementing anything non-trivial or under-specified; do not delegate to it automatically — it pairs with the user rather than running unattended.

• Researcher (Advanced)

  Specialized agent for complex, multi-source research across web, documentation, and repositories, synthesized into comprehensive, validated reports. ALWAYS invoke when research spans 3+ sources or requires comparative analysis and synthesis for a technical decision. Do not use for simple lookups or

MCP servers (2)

• context7

  connects to: https://mcp.context7.com/mcp

• github

  runs: gh mcp

Similar plugins