By name: claude-effective-usage description: | Use this skill whenever the operator could work with Claude more effectively — by picking the right surface, the right tool tier, or the right session hygiene. Three families: (1) RIGHT SURFACE — choosing among Claude Code / Chat / Cowork / Dispatch. Triggers like "by EOD", "have this ready when I get back", "investigate X and report", "go gather all the Y and summarize" (async multi-step → Dispatch); "brainstorm with me", "what do you think of", "help me think through", "think out loud" (open exploration → Chat); "draft a proposal", "write the brief", "compose an email", "I need a spec doc" (document drafting → Cowork). (2) TOOL-TIER EFFICIENCY — about to use computer-use / screenshots when a dedicated app MCP exists; about to poll or sleep-loop for a condition that could be event-driven; about to make several independent tool calls sequentially when they could run in parallel in one message; a bounded task that maps to a specialized agent or a codebase-wide search. (3) SESSION HYGIENE — "save this" / "remember this" / "that's an important point" / a decision or insight worth keeping (pin a durable conclusion); "you keep forgetting" / "we already did this" / response quality drifting (clear or compact stale context); a long resolved thread that's starting to drift (summarize + archive). Consult the rules/ folder for the specific pattern and apply its corrective. Soft teaching response — names the better path and yields. Never forces a surface switch or runs /clear autonomously (those are operator actions). Do NOT use when the operator is mid-build on the correct surface, in solo-run mode, or has explicitly chosen the current surface/tool. license: MIT — see plugin LICENSE metadata: priority: medium promptSignals: phrases: - "by EOD" - "have this ready when I get back" - "investigate" - "research this and" - "gather all the" - "brainstorm with me" - "what do you think of" - "help me think through" - "draft a proposal" - "write the brief" - "compose an email" - "I need a spec doc" - "just screenshot" - "let me look at the screen" - "keep checking" - "poll until" - "wait for it to finish" - "read these files" - "save this" - "remember this" - "that's an important point" - "you keep forgetting" - "we already did this" - "you just told me that" - "we basically solved this" - "let me start a new chat" - "persistent doc" - "bounded task" - "search across the whole codebase" - "sleep loop" - "poll the deploy" - "read this file, then"
Claude effective usage
Patterns for getting more out of Claude — right surface, right tool tier, right session hygiene. Each is a small rule file with its own trigger, corrective, and override. When the operator's situation matches a rule's trigger, consult that rule and apply its specific response. One nudge per turn — don't stack.
Rules
| Family | Pattern | Trigger | Rule |
|---|---|---|---|
| Surface | Async multi-step work | "by EOD", "have this ready", "investigate X and report" | dispatch-this |
| Surface | Open-ended brainstorming | "let me think about", "brainstorm with me", "what do you think of" | switch-to-chat-instead |
| Surface | Long-form document drafting | "draft a proposal", "write the brief", "compose an email" | use-cowork-not-code |
| Tier | Bounded task → specialized agent | bounded task that maps to an agent, or a codebase-wide search | invoke-subagent |
| Tier | Polling when event-driven is possible | about to sleep-loop / retry-poll for a condition | monitor-instead-of-poll |
| Tier | Generic tool when a specific MCP exists | about to screenshot / computer-use when an app MCP exists | prefer-dedicated-mcp |
| Tier | Serial calls that could be parallel | about to Read→Read→Read or grep→grep sequentially | single-message-multi-tool |
| Hygiene | Pin-worthy conclusion | "save this", "remember this", a decision/insight surfaced | pin-this |
| Hygiene | Context staleness | "you keep forgetting", repeating, drift back to old assumptions | clear-context-when-stale |
| Hygiene | Long resolved thread | 30+ turns, core question resolved, drift starting | summarize-and-archive |
Dispatch logic
- Match the situation to the closest rule trigger. If unclear, name what you see and let the operator pick.
- Open that rule file. Apply its corrective shape.
- Be soft. Name the better path; offer the handoff; yield.
- Don't stack. One nudge per turn unless the operator engages a second.
Suppression rules (across all sub-rules)
- The operator is mid-build on the correct surface (don't suggest switching surfaces mid-flow)
solo-runmode is active — suppress surface/efficiency nudges- The operator explicitly chose the current surface or tool ("stay here", "use [tool]")
- The corrective is destructive of conversation state (
/clear,/compact) — suggest, never run autonomously
What this skill never does
- Force a surface switch (Code ↔ Chat ↔ Cowork ↔ Dispatch are operator actions)
- Run
/clearor/compactautonomously - Spin up a Dispatch task or send a draft to Cowork on the operator's behalf
- Add narrative noise when the right move is just to silently batch tool calls
- Repeat a nudge the operator already declined
Cross-references
build-disciplineumbrella — sibling; right-tool-for-the-job at the build layerdecision-disciplineumbrella — companion;pin-thischains into decision capturesession-flowumbrella — companion; session-shape and handoff patterns
Why this umbrella earns its keep
The most common quiet losses in a Claude session: doing chat-shaped work in a heavy Code session, pixel-clicking when an MCP exists, serial tool calls that could be parallel, letting durable conclusions evaporate, and running a session past the point its context stays sharp. Each is cheap to catch and compounds across every session. Bundling them under one umbrella means the operator doesn't track ten slugs — the umbrella fires on a broad signal set and each rule stays small.