Install
$ agentstack add skill-vdustr-vp-claude-code-marketplace-guided-focus ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Guided Focus
Structured, one-at-a-time questioning to converge on requirements and decisions before planning or execution. Each question presents options with descriptions, trade-off differences, and a recommended choice with reasoning.
Quick Start
> /guided-focus
> /guided-focus 5
> Help me clarify before we plan
> Ask me questions first, then plan
When to Use
- Complex tasks with multiple valid approaches that need alignment before planning
- Ambiguous requirements where scope, goals, or constraints are unclear
- High-impact decisions (architecture, tech stack, migration strategy) that benefit from structured exploration
- User explicitly asks to clarify or align before starting work
When NOT to use: Single-file edits, bug fixes with clear reproduction steps, tasks where the user provides explicit requirements, or when the user has already described a clear plan.
Workflow
Phase 0: Activation
- Manual: User invokes
/guided-focusor asks to clarify first - Proactive: When detecting a complex/ambiguous task, ask: "This task has multiple valid approaches. Want to run guided focus first to align on direction?" — proceed only if user confirms
- Do NOT trigger proactively for: single-file edits, bug fixes with clear reproduction steps, tasks with explicit requirements, or when the user has already described a clear plan
Phase 1: Configuration
Defaults (user can override via arguments or during the session):
| Parameter | Default | Override Example | |-----------|---------|------------------| | Question count | 10 | /guided-focus 5 or /guided-focus --questions 5 | | Min options per question | 3 | /guided-focus --min-options 4 |
Phase 2: Questioning Loop
Ask questions one at a time. Wait for user response before proceeding.
Question format (use bullet list, not table — safe for narrow terminals):
### Question [current]/[total] — [Topic]
[Context sentence explaining why this question matters]
**A. [Option name]**
- Description: [what this option means]
- Difference: [how it differs from other options]
- Recommendation: Weak / Moderate / Strong
**B. [Option name]**
- Description: ...
- Difference: ...
- Recommendation: ...
**C. [Option name]**
- Description: ...
- Difference: ...
- Recommendation: ...
**Recommended: [letter] — [option name]**: [1-2 sentence reasoning]
Recommendation scale: Weak = viable but not ideal, Moderate = good fit, Strong = best fit for this context.
Question behavior rules:
- Dependency awareness: Adapt subsequent questions based on previous answers. Skip questions that become irrelevant, add questions that become necessary. When adjusting, tell the user: "Based on your answer to Q[N], I'm [skipping Q about X / adding a question about Y] because [reason]."
- Progress indicator: Always show
[current]/[total]and update total if it changes - Dynamic adjustment: May increase or decrease question count based on task complexity, but never exceed 15 questions total. Always inform the user: "I'm [adding/reducing] questions because [reason]. New total: [N]."
User response flexibility:
- Select an option (A/B/C)
- Combine options (A+B) — if combined options conflict, explain the conflict and ask the user to choose one or describe a reconciliation
- Provide a free-form answer
- Ask a counter-question for clarification
- Skip and defer: "skip" or "later" — mark as deferred, addressed in Phase 3
- Express confidence level: "B, but not sure" → marked as uncertain in summary
- End early: "enough", "done", or "wrap up" → immediately proceed to Phase 3
Phase 3: Deferred Question Resolution
If any questions were skipped/deferred during Phase 2:
- Present all deferred questions in a single list
- Ask once: "Want to revisit these, or proceed with them marked as undecided?"
- If revisit → ask each deferred question with context from other answers
- If proceed → mark all as undecided in summary
Phase 4: Summary
Generate a structured summary of all decisions:
## Focus Summary
**1. [Topic]**: [Decision] [confidence]
**2. [Topic]**: [Decision] [confidence]
...
### Uncertain Decisions
- **[Topic]**: [Decision marked as uncertain] — may need revisiting during implementation
### Undecided
- **[Topic]**: Deferred — [reason or context]
Confidence markers: [confirmed] | [uncertain] | [undecided]
Phase 5: Review & Handoff
After presenting the summary:
- Ask: "Want to modify any decisions, or add more questions?"
- If yes → allow edits or additional questions, regenerate summary
- If no → ask: "Ready to proceed? Options: plan (enter plan mode), execute (start directly)"
Guidelines
DO
- One question at a time — never batch multiple questions
- Adapt questions dynamically — use dependency awareness, not a fixed script
- Explain adjustments — always tell the user why questions are added/removed/reordered
- Respect all response types — selections, combinations, free-form, counter-questions, skips
- Show progress — always display
[current]/[total] - Surface uncertainty — mark low-confidence decisions visibly in the summary
- Wrap up early — if requirements are clear before reaching the question count, propose to end and proceed to summary
DON'T
- Start without confirmation when triggered proactively — always ask first
- Ask obvious questions — skip questions where the answer is clearly implied by context
- Force linear order — if a user's answer makes a later question urgent, reorder
- Over-question — if requirements are clear, wrap up. The number of questions is dynamic but must not exceed the hard cap of 15
Notes
- Questions should progress from high-level (scope, goals, constraints) to specific (implementation details, trade-offs)
- The first 2-3 questions should establish context and scope; detailed questions come after
- If the user's project already has constraints (existing codebase, tech stack), detect and incorporate them rather than asking about known facts
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: VdustR
- Source: VdustR/vp-claude-code-marketplace
- License: MIT
- Homepage: https://github.com/VdustR/skills
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.