From sentinel
Structured exploration before implementation. Activates when building new features, making design decisions, or facing multiple valid approaches. Prevents jumping to code before understanding intent. Explores context, asks clarifying questions, proposes approaches, and produces a spec.
How this skill is triggered — by the user, by Claude, or both
Slash command
/sentinel:brainstormThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This skill enforces a structured exploration phase before any implementation begins. The core problem it solves: Claude jumps straight to writing code when the user says "build X," but the user's mental model of X and Claude's interpretation of X are often different. The result is technically correct code that misses the intent.
This skill enforces a structured exploration phase before any implementation begins. The core problem it solves: Claude jumps straight to writing code when the user says "build X," but the user's mental model of X and Claude's interpretation of X are often different. The result is technically correct code that misses the intent.
Before proposing anything, understand what exists:
Do NOT propose solutions during this phase. Just gather context.
Ask the user focused questions to resolve ambiguity. Rules:
Present 2-3 distinct approaches with trade-offs:
Approach A: [Name]
How: [1-2 sentence description]
Pros: [what's good about it]
Cons: [what's risky or costly]
Files: [which files would change]
Approach B: [Name]
How: [1-2 sentence description]
Pros: [what's good about it]
Cons: [what's risky or costly]
Files: [which files would change]
Recommendation: [A or B] because [reason]
Rules for proposals:
Once the user picks an approach, write a brief spec before coding:
## Spec: [Feature Name]
### What
[1-2 sentences describing what will be built]
### Approach
[The chosen approach from Phase 3]
### Changes
- [ ] [File 1] — [what changes]
- [ ] [File 2] — [what changes]
- [ ] [File 3] — [what changes]
### Not Doing
- [Explicitly list what's out of scope]
### Verification
- [How to verify this works — specific test or command]
The spec serves two purposes:
After the user approves the spec:
Wrong: User says "add a search bar" → Claude immediately creates SearchBar.tsx Right: User says "add a search bar" → Claude asks "Should this search the current page content or make an API call? The existing list components use client-side filtering."
Wrong: "Here's how we'll implement this" → one approach with no alternatives Right: "Two ways to approach this: A uses the existing filter pattern, B adds a dedicated search endpoint. I'd recommend A because..."
Wrong: "Before I start, can you answer these 8 questions: 1) ... 2) ... 3) ..." Right: "Should this filter results in real-time as the user types, or only on submit? The current pattern uses debounced real-time filtering."
Wrong: "Option A: Good approach. Option B: Same approach but worse." Right: "Option A: Client-side filtering (fast, works offline, limited to loaded data). Option B: Server-side search (handles large datasets, requires API endpoint, adds latency)."
Wrong: "Let me just quickly code this up and you can see if it's right" Right: "Let me understand the requirements first, then propose an approach for your approval"
After brainstorming completes and the spec is approved:
sentinel-writing-plans to decompose the spec into bite-sized 2-5 minute tasks. The plan file becomes the input to either sentinel-subagent-driven-development (preferred when subagents are available) or sentinel-executing-plans.vault/workflows/feature-improvement.md (step 2: Define)vault/workflows/refactor.md (step 1: Understand)vault/workflows/research-spike.mdThe brainstorm output is a spec, not a plan. The spec answers "what should we build and why." The plan answers "what's the exact sequence of file edits to build it." Don't conflate the two — see sentinel-writing-plans for the plan-vs-spec distinction.
npx claudepluginhub digistrique-solutions/strique-marketplace --plugin sentinelProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.