benbrastmckie/nvim/skill-consult

Claude Code and agent skill by benbrastmckie/nvim

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Route design consultations to domain-specific design partner agents

Install

npx vskill@latest install benbrastmckie/nvim/skill-consult

SKILL.md

Consult Skill

Thin wrapper that routes design consultation requests to domain-specific design partner agents.

IMPORTANT: This skill implements the skill-internal postflight pattern. After the subagent returns, this skill handles all postflight operations (artifact linking, git commit) before returning.

Context Pointers

Reference (do not load eagerly):

  • Path: .claude/context/formats/subagent-return.md
  • Purpose: Return validation
  • Load at: Subagent execution only

Note: This skill is a thin wrapper. Context is loaded by the delegated agent, not this skill.

Trigger Conditions

This skill activates when:

Direct Invocation
  • User runs /consult --legal command
Routing
  • domain=legal routes to legal-analysis-agent
  • Future: domain=investor, domain=technical, domain=competitor
When NOT to trigger

Do not invoke for:

  • Contract review (use skill-legal / legal-council-agent)
  • Market research (use skill-market)
  • Competitive analysis (use skill-analyze)

Execution Flow

Stage 1: Input Validation

Validate required inputs:

  • domain - Required, currently only "legal" supported
  • input_type - Required, one of: file_path, inline_text, design_question, task_number
  • task_number - Required (always provided by consult.md Gate In)
  • session_id - Required
bash
# Validate domain
case "$domain" in
legal) agent="legal-analysis-agent" ;;
*) return error "Unsupported domain: $domain. Currently supported: legal" ;;
esac
# Validate input type
case "$input_type" in
file_path|inline_text|design_question|task_number) ;;
*) return error "Invalid input_type: $input_type" ;;
esac

Stage 2: Resolve Task Context

Resolve task directory for artifact storage. task_number is always provided by the command's Gate In:

bash
task_data=$(jq -r --argjson num "$task_number" \
'.active_projects[] | select(.project_number == $num)' \
specs/state.json)
if [ -z "$task_data" ]; then
return error "Task $task_number not found"
fi
project_name=$(echo "$task_data" | jq -r '.project_name')
padded_num=$(printf "%03d" "$task_number")
task_dir="specs/${padded_num}_${project_name}"
metadata_file="${task_dir}/.return-meta.json"
mkdir -p "$task_dir"

Stage 3: Prepare Delegation Context
json
{
"task_context": {
"description": "{description or input summary}",
"task_type": "founder"
},
"domain": "legal",
"input_type": "{file_path | inline_text | design_question | task_number}",
"file_path": "{file path if applicable}",
"inline_text": "{inline text if applicable}",
"design_question": "{design question if applicable}",
"task_number": "{task_number}",
"metadata_file_path": "{metadata_file}",
"metadata": {
"session_id": "{session_id}",
"delegation_depth": 1,
"delegation_path": ["orchestrator", "consult", "skill-consult"]
}
}

Stage 4: Invoke Agent

CRITICAL: You MUST use the Agent tool to spawn the agent.

Required Tool Invocation:

Tool: Agent (NOT Skill, NOT Plan)
Parameters:
- subagent_type: "{agent}" (e.g., "legal-analysis-agent")
- prompt: [Include all delegation context fields]
- description: "Legal design consultation"

The agent will:

  • Understand user intent via Socratic dialogue
  • Translate document language to attorney perspective
  • Identify translation gaps
  • Suggest reframings with rationale
  • Write consultation report
  • Write metadata file
  • Return brief text summary

Stage 4b: Self-Execution Fallback

CRITICAL: If you performed the work above WITHOUT using the Agent tool (i.e., you read files, wrote artifacts, or updated metadata directly instead of spawning a subagent), you MUST write a .return-meta.json file now before proceeding to postflight. Use the schema from return-metadata-file.md with the appropriate status value for this operation.

If you DID use the Agent tool, skip this stage -- the subagent already wrote the metadata.


Postflight (ALWAYS EXECUTE)

The following stages MUST execute after work is complete, whether the work was done by a subagent or inline (Stage 4b). Do NOT skip these stages for any reason.

Stage 5: Read Metadata File
bash
if [ -f "$metadata_file" ] && jq empty "$metadata_file" 2>/dev/null; then
status=$(jq -r '.status' "$metadata_file")
artifact_path=$(jq -r '.artifacts[0].path // ""' "$metadata_file")
artifact_type=$(jq -r '.artifacts[0].type // ""' "$metadata_file")
artifact_summary=$(jq -r '.artifacts[0].summary // ""' "$metadata_file")
else
status="failed"
fi

Stage 6: Link Artifacts

Link the consultation artifact to the task in state.json:

bash
if [ -n "$artifact_path" ]; then
# Add consultation artifact to state.json
bash .claude/scripts/state-write.sh \
'(.active_projects[] | select(.project_number == $num)).artifacts += [{"path": $path, "type": $type, "summary": $summary}]' \
--session-id "$session_id" \
--arg path "$artifact_path" \
--arg type "$artifact_type" \
--arg summary "$artifact_summary" \
--argjson num "$task_number"
fi

Stage 7: Git Commit

Apply targeted staging per .claude/context/standards/git-staging-scope.md — never a repo-wide add:

bash
bash .claude/scripts/git-commit-scoped.sh \
--message "task ${task_number}: legal design consultation" \
--session "${session_id}" \
--honest-index-rows "${task_number}" \
-- "${task_dir}/" "specs/TODO.md" "specs/state.json"

Stage 8: Return Brief Summary
Design consultation complete ({domain}):
- Input: {input type and description}
- Translation gaps: {N} identified
- Consultation report: {artifact_path}
- Task: #{task_number}
- Advisory: recommend attorney review for high-stakes materials

Return Format

Brief text summary (NOT JSON).


Error Handling

Unsupported Domain

Return immediately with supported domain list.

Task Not Found

Return error, do not proceed.

Metadata File Missing

Return partial status with progress from agent output.

Git Commit Failure

Non-blocking: Log failure but continue.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/benbrastmckie/nvim/skill-consult/badge)](https://verified-skill.com/skills/benbrastmckie/nvim/skill-consult)

Something wrong with this skill or its page? Tell us.