Self-Reflection & Routing

Before executing any task, you must strictly follow this self-reflection protocol to ensure you fully understand the user's request.

Step 1: Intent Confidence Check

Question: Are you confident with the user's intents?

NO / UNSURE: Activate intent-inference skill.

# Analyze intent
bun scripts/run_skill.js intent-inference

YES: Can you articulate them? Write them down.

Step 2: Problem Articulation Check

Question: Can you articulate the problems that you need to address?

NO / UNSURE: Activate problem-statement skill.

# Define problem using 5 Whys
bun scripts/run_skill.js problem-statement

YES: Write down the root problem.

Step 3: Scoping Check

Question: Can you list 3 items in-scope and 3 items out-of-scope?

NO / UNSURE: Activate scoping skill.

# Define scope
bun scripts/run_skill.js scoping

YES: Write them down.

Step 4: Artifact Recording

Ensure all your findings (Intent, Problem, Scope) are recorded in debug/context.json.

mkdir -p debug
# Append or write findings to debug/context.json

Skill Router (Standard Mode)

Once you have clarity on the intent, problem, and scope, proceed to find and activate the appropriate skills to solve the task.

Find Skills

# Search by keyword (instant, no install needed)
bun scripts/search_skills.js <keyword>

# Or with node (fallback)
node scripts/search_skills.js <keyword>

# List all skills (if installed)
npm ls --depth=0

# Query installed skills (if installed)
npm query "[keywords~=<keyword>]"

Activate Skill

// 1. Search
const result = terminal.execute("bun scripts/search_skills.js brainstorming");
const skill_name = parse_first_result(result);

// 2. Read instructions
const skill_path = `skills/${skill_name}/SKILL.md`; // OR skills/<skill_name>/skill.json
const instructions = filesystem.read(skill_path);

// 3. Execute
follow(instructions);

Exit

If no skills match: report "No skill found for "

FilesExpand file tree

SKILL.md

Latest commit

History