haowjy/meridian-base

knowledge-layers

Use when deciding where knowledge goes or reading/writing durable docs: AGENTS.md, .context/, KB, docs/, and work directories.

qi-layer

Use when writing or maintaining AGENTS.md, .context/CONTEXT.md, or CLAUDE.md mirrors: keep intent docs minimal and load-bearing.

qi-maintenance

Always load on agents that make durable source changes: keep colocated agent-facing knowledge synchronized with the source.

handoff

Produce a transition brief and launch command for the user's next interactive primary.

structured-artifact

Load when building a static HTML artifact: single page or multi-page site to make structured information explorable through layout and navigation.

work-artifacts

Anchor work in a work item: artifacts on disk, not in context.

grill-with-docs

Use when grilling should be grounded primarily in project artifacts. Loads `/grill`, checks project vocabulary and prior decisions first, and spawns bounded explorers when the answer should come from docs.

information-hierarchy

Load for large human-facing outputs: reports, docs, multi-section explanations, artifacts. Choose the right form for each beat (prose, diagram, table, mockup) and put the answer first, depth behind it. If the medium allows, use progressive disclosure.

explore-and-engage

Investigate and engage concurrently: don't block on evidence before surfacing your read.

session-mining

Use when recovering decisions or context from conversation history: session log navigation, transcript delegation.

shared-dao

Use when establishing shared vocabulary: resolve terminology early, one name per concept, vocab files as record.

shared-workspace

Use when a workspace has changes from other agents: safe orientation, preserving unfamiliar files.

zoom-out

Use when orienting in an unfamiliar area: fans out parallel explorers across code, knowledge, and history, then synthesizes one orientation map.

md-validation

Use when validating markdown links or Mermaid diagrams.

meridian-privilege-escalation

Use when a spawn fails due to sandbox or permission restrictions.

llm-writing

Load before writing or revising human-facing text. Choose words deliberately, ground the piece in the reader's context, and remove default LLM phrasing before the final draft.

goal-writing

Load when turning intent into an executable goal for an agent, spawn, work item, experiment, or handoff. Use to define outcome, evidence, constraints, anti-goals, stop conditions, and reporting shape before work begins.

grill-with-evidence

Use when grilling should be grounded in evidence. Loads `/grill`, then checks the plan, terms, and assumptions against project artifacts first and primary sources when needed.

grill

Use in conversation to stress-test a plan or direction. Interviews the human branch by branch, sharpens terminology, challenges assumptions, and asks the next high-leverage question with a recommended answer.

intent-modeling

Use before acting on human instructions: separate what they said from what they meant.

reflect

User-invoked pause before reporting to check intent vs literal completion, surface adjacent wins, and route knowledge capture.