using-ghengis-skills

name	using-ghengis-skills
description	Use when starting any conversation - establishes how to find and use ghengis-skills, requiring Skill tool invocation before ANY response when a ghengis-skill applies (mirrors superpowers:using-superpowers pattern)

If you were dispatched as a subagent to execute a specific task, skip this skill. If you think there is even a 1% chance a ghengis-skill might apply to what you are doing, you ABSOLUTELY MUST invoke the skill.

IF A SKILL APPLIES TO YOUR TASK, YOU DO NOT HAVE A CHOICE. YOU MUST USE IT.

This is not negotiable. This is not optional. You cannot rationalize your way out of this.

Why This Skill Exists

Ghengis skills are domain expertise and operational reliability tools that complement (not replace) workflow skills like superpowers. Without an autoloader, they get crowded out by skills that DO autoload. This skill ensures ghengis skills get equal priority treatment.

Instruction Priority

Ghengis skills override default system prompt behavior, but user instructions always take precedence:

1. User's explicit instructions (CLAUDE.md, GEMINI.md, AGENTS.md, direct requests) — highest priority
2. Ghengis skills — override default system behavior where they conflict
3. Other autoloaded skill systems (e.g., superpowers) — coexist; both should be checked
4. Default system prompt — lowest priority

If both ghengis-skills and superpowers offer skills for the same task, prefer the more specific one. If user says "don't use X," follow the user.

How to Access Skills

Use the Skill tool. Invoke with the qualified name: ghengis-skills:<skill-name>. When invoked, the skill content loads — follow it directly. Never use Read on skill files.

When to Invoke Ghengis Skills

The Rule

Invoke relevant or requested skills BEFORE any response or action. Even a 1% chance a skill might apply means invoke it. If the skill turns out to be wrong, drop it.

Decision Flow

digraph ghengis_skill_flow {
    "User message received" [shape=doublecircle];
    "About to dispatch a subagent?" [shape=diamond];
    "Use meta-prompting + pql-validation" [shape=box];
    "Subagent claims done?" [shape=diamond];
    "Use completion-enforcer" [shape=box];
    "Doing complex/iterative research?" [shape=diamond];
    "Use deep-research" [shape=box];
    "Working with FastAPI/React/Flutter/ESP32?" [shape=diamond];
    "Use the matching domain skill" [shape=box];
    "Long session or context drift suspected?" [shape=diamond];
    "Use context-health" [shape=box];
    "Multiple parallel agents needed?" [shape=diamond];
    "Use agent-teams or oort-cascade" [shape=box];
    "Reviewing AI output for accuracy?" [shape=diamond];
    "Use hallucination-detector" [shape=box];
    "Other ghengis-skill might apply?" [shape=diamond];
    "Invoke it" [shape=box];
    "Respond" [shape=doublecircle];

    "User message received" -> "About to dispatch a subagent?";
    "About to dispatch a subagent?" -> "Use meta-prompting + pql-validation" [label="yes"];
    "About to dispatch a subagent?" -> "Subagent claims done?" [label="no"];
    "Use meta-prompting + pql-validation" -> "Subagent claims done?";
    "Subagent claims done?" -> "Use completion-enforcer" [label="yes"];
    "Subagent claims done?" -> "Doing complex/iterative research?" [label="no"];
    "Use completion-enforcer" -> "Doing complex/iterative research?";
    "Doing complex/iterative research?" -> "Use deep-research" [label="yes"];
    "Doing complex/iterative research?" -> "Working with FastAPI/React/Flutter/ESP32?" [label="no"];
    "Use deep-research" -> "Working with FastAPI/React/Flutter/ESP32?";
    "Working with FastAPI/React/Flutter/ESP32?" -> "Use the matching domain skill" [label="yes"];
    "Working with FastAPI/React/Flutter/ESP32?" -> "Long session or context drift suspected?" [label="no"];
    "Use the matching domain skill" -> "Long session or context drift suspected?";
    "Long session or context drift suspected?" -> "Use context-health" [label="yes"];
    "Long session or context drift suspected?" -> "Multiple parallel agents needed?" [label="no"];
    "Use context-health" -> "Multiple parallel agents needed?";
    "Multiple parallel agents needed?" -> "Use agent-teams or oort-cascade" [label="yes"];
    "Multiple parallel agents needed?" -> "Reviewing AI output for accuracy?" [label="no"];
    "Use agent-teams or oort-cascade" -> "Reviewing AI output for accuracy?";
    "Reviewing AI output for accuracy?" -> "Use hallucination-detector" [label="yes"];
    "Reviewing AI output for accuracy?" -> "Other ghengis-skill might apply?" [label="no"];
    "Use hallucination-detector" -> "Other ghengis-skill might apply?";
    "Other ghengis-skill might apply?" -> "Invoke it" [label="yes, even 1%"];
    "Other ghengis-skill might apply?" -> "Respond" [label="no"];
    "Invoke it" -> "Respond";
}

Mandatory Trigger Map

These situations ALWAYS require the listed skill — no exceptions:

Situation	Required Skill
About to write a prompt for a subagent	ghengis-skills:meta-prompting (use a role template)
About to dispatch a subagent with a custom prompt	ghengis-skills:pql-validation (verify prompt quality FIRST)
Subagent reports DONE / completes	ghengis-skills:completion-enforcer (verify it actually finished)
User asks for research that needs depth/iteration	ghengis-skills:deep-research
User asks for quick lookup	ghengis-skills:general-research
Working with FastAPI code	ghengis-skills:fastapi
Working with React/Next.js code	ghengis-skills:react-nextjs
Working with Flutter/Dart code	ghengis-skills:flutter-dart
Working with ESP32/IoT firmware	ghengis-skills:esp32
Working with 3D models / STL files	ghengis-skills:3d-modeling
Working with MCP servers	ghengis-skills:mcp-patterns
Setting up DevOps / CI/CD / Docker	ghengis-skills:devops
Security review / vulnerability assessment	ghengis-skills:security-testing
Data analysis / visualization	ghengis-skills:data-analysis
Writing reports / structured docs	ghengis-skills:report-writing
Writing content / blog posts / docs	ghengis-skills:content-writing
Need multiple creative perspectives, parallel ideation, or comparing 2+ approaches	ghengis-skills:agent-teams (AUTO-FIRE)
Multi-step pipeline with role decomposition	ghengis-skills:oort-cascade
Noticed a repeated multi-step workflow done 2+ times (e.g., research→spec→plan→execute happens regularly)	ghengis-skills:blueprint-compilation (AUTO-FIRE)
Reviewing AI/agent output for hallucinations	ghengis-skills:hallucination-detector
Building agentic workflow	ghengis-skills:constitutional-ai (safety boundaries)
Long-running session, context degradation	ghengis-skills:context-health
Task too large for one session	ghengis-skills:execution-harness
Repeated workflow you might want to automate	ghengis-skills:blueprint-compilation
Setting up scheduled/cron tasks	ghengis-skills:proactive-rituals
Need to learn from past tasks	ghengis-skills:skill-memory
Building agent that needs accountability	ghengis-skills:audit-ledger
Hitting rate limits / cost ceilings	ghengis-skills:compute-adaptation
Starting new project	ghengis-skills:project-scaffold
Organizing files / cleanup	ghengis-skills:file-organization
Tracking goals across multi-step work	ghengis-skills:goal-tracking
Tracking tasks/projects/priorities	ghengis-skills:task-tracking
Major work batch completed (plan execution, feature build, multi-task subagent run)	ghengis-skills:auto-project-sync
User runs /sync slash command	ghengis-skills:auto-project-sync
Adapting agent behavior to user style	ghengis-skills:agent-identity
Codebase exploration / architecture analysis	ghengis-skills:code-intelligence
Formatting output for specific destination	ghengis-skills:output-formatting
Teaching / explaining concepts	ghengis-skills:tutoring
Building learning curriculum	ghengis-skills:learning-paths
Time/calendar/routine design	ghengis-skills:scheduling
Smart light/ambiance design	ghengis-skills:home-lighting
Music discovery / playlist creation	ghengis-skills:music-curation
Purchase decisions / product research	ghengis-skills:shopping
Financial record-keeping / tax prep	ghengis-skills:bookkeeping
Client/freelance project management	ghengis-skills:crm-patterns

Aggressive Auto-Fire Skills

These skills should fire proactively without user prompting when their conditions match — don't wait for the user to ask:

agent-teams — Auto-fire when:

• User asks "what are some options for X" or "compare approaches"
• Brainstorming session with 2+ candidate directions
• Creative output where divergence-then-synthesis would help
• User says "give me different angles" or similar
• Architecture decision with multiple valid paths

blueprint-compilation — Auto-fire when:

• You've executed the same multi-step workflow 2+ times in the project (e.g., "research → spec → plans → subagent execution" pattern)
• User has a repeated process they manually trigger each time
• Detecting "this feels like something we just did"
• After completing any multi-phase build, ask: "Should I blueprint this?"

auto-project-sync — Auto-fire when:

• subagent-driven-development reports all tasks complete
• superpowers:finishing-a-development-branch invoked
• 5+ commits since last sync (check .claude/last_sync)
• User signals a major phase done ("Phase X complete", "all done", "great work")

meta-prompting — Auto-fire when:

• About to write ANY subagent prompt
• About to dispatch via Agent tool with custom prompt
• Never write subagent prompts freehand without this skill

pql-validation — Auto-fire when:

• Just generated a subagent prompt (validate before dispatch)
• About to send a complex prompt to an LLM
• Reviewing prompts written by user or other agents

completion-enforcer — Auto-fire when:

• Subagent reports DONE, COMPLETE, or similar
• About to mark a task done in TaskUpdate
• Before claiming "all tasks complete" to the user

Red Flags

These thoughts mean STOP — you're rationalizing:

Thought	Reality
"Superpowers handled this"	Both systems coexist. Check ghengis too.
"I know how to do this"	Skills evolve. Read current version.
"This is too quick to need a skill"	Quick tasks are where errors slip through. Use the skill.
"I'll just write the prompt myself"	meta-prompting + pql-validation prevent stuck/wrong agents.
"The agent finished, looks good"	completion-enforcer catches premature/false completions.
"I'll do shallow research"	If complex, use deep-research. Don't skip the methodology.
"FastAPI is just web code"	The fastapi skill has patterns specific to async/Pydantic v2/dependency injection. Use it.
"I'll just synthesize myself instead of using agent-teams"	Multiple perspectives produce better results than solo synthesis. Use it.
"I'll remember this pattern for next time"	You won't. Blueprint it.
"Sync can wait"	Drift compounds. Sync after every meaningful batch.

Coexistence with Superpowers

Both autoload systems are valid. They overlap intentionally — pick the more specific skill:

Task	Pick
Brainstorming a feature	superpowers:brainstorming (workflow skill)
Writing a sub-agent prompt during brainstorming	ghengis-skills:meta-prompting (specific tool)
Writing implementation plans	superpowers:writing-plans
Building FastAPI endpoints in that plan	ghengis-skills:fastapi (domain expertise)
Executing plans with subagents	superpowers:subagent-driven-development
Verifying each subagent's completion	ghengis-skills:completion-enforcer
Researching a library	ghengis-skills:deep-research (or general-research for quick)
TDD discipline	superpowers:test-driven-development
Reviewing finished work	superpowers:requesting-code-review
Checking for hallucinations in agent output	ghengis-skills:hallucination-detector

When NOT to Invoke

• During subagent execution (<SUBAGENT-STOP> above)
• When the user has explicitly told you not to use skills for this task
• When invoking would create infinite recursion (skill calling itself)

How to Announce Use

When invoking a ghengis skill, announce briefly:

"Using ghengis-skills:meta-prompting to write the subagent prompt."

This keeps the user informed without verbose explanation.