--- name: ultimate-skill-creator description: > The definitive end-to-end pipeline for creating the highest quality portable AI skills. Orchestrates six individual skills in sequence — each run fully, separately, and in its entirety — to produce elite, production-grade skills. Use when the user explicitly requests the full pipeline, says "ultimate skill", "ultimate skill creator", "ultimate skill creation", "best skill", "best skill creator", "full pipeline", "full skill pipeline", or specifically asks for maximum quality skill creation with all optimization stages. For simple skill creation without the full pipeline, use skill-creator instead. --- # Ultimate Skill Creator Execute a comprehensive six-stage pipeline to create, optimize, triage, connect, debug, and deploy a production-grade portable AI skill. Each stage runs a fully independent skill through its complete workflow — no shortcuts, no combining, no skipping. ## Pipeline Overview | Stage | Skill | Purpose | Exit Criteria | |---|---|---|---| | 1 | `skill-creator` | Full 6-step creation process | Passes `quick_validate.py`, all scripts execute | | 2 | `prompt-engineer` | Deep multi-model SKILL.md optimization | Optimized SKILL.md passes validation | | 3 | `master-skill-orchestrator` | Skill Triage Protocol for domain analysis | Triage documented in `/tmp/triage_$SKILL_NAME.md` | | 4 | `skill-connection-map` | Connection Discovery Protocol | Connections documented in `/tmp/connections_$SKILL_NAME.md` | | 5 | `skill-debugger` | Deep dual-model debug + fix cycle | No critical/high issues after max 3 cycles | | 6 | `skill-sync` | Push to GitHub + local installation | Deployed and verified | ## Pre-Flight Check Run the pre-flight script before starting the pipeline: ```bash python3 /home/ubuntu/skills/ultimate-skill-creator/scripts/preflight_check.py ``` This verifies all six required skills are installed, at least one API key is available, and GitHub CLI is authenticated. If it fails, fix the reported errors before proceeding. If it passes with warnings, the pipeline can proceed with limited functionality. ## Operational Rules 1. **Read each skill's SKILL.md fresh** at the start of each stage. Do not rely on memory. 2. **Complete each stage fully** before advancing. Do not skip steps within a skill's workflow. 3. **Announce each stage transition**: "Starting Stage N: [skill-name] — [purpose]." 4. **Stage transition checkpoint** — Before starting each new stage, confirm the stage number, skill name, and purpose from the Pipeline Overview table above. Do not re-read the entire SKILL.md unless uncertain about requirements. 5. **Ask clarifying questions ONLY during Stage 1** (Understand step). Proceed autonomously thereafter. 6. **On fatal errors**, inform the user immediately with error context and recovery options. 7. **Clean up temporary files** after each stage, including on failure. 8. **Sequential execution only** — Complete each stage fully before starting the next. Do not run stages concurrently. --- ## Stage 1: skill-creator — Full Creation **Read**: `/home/ubuntu/skills/skill-creator/SKILL.md` Execute the skill-creator's complete 6-step process: 1. **Understand** — Gather concrete examples of how the skill will be used. Ask the user clarifying questions if the skill's purpose is unclear. Do not proceed until you have a clear sense of the functionality. 2. **Plan** — Identify reusable resources (scripts, references, templates). Determine what goes in each directory. 3. **Initialize** — Run `python3 /home/ubuntu/skills/skill-creator/scripts/init_skill.py $SKILL_NAME`. Clean up example files. 4. **Edit** — Read the skill-creator's reference files (`references/workflows.md`, `references/output-patterns.md`, `references/progressive-disclosure-patterns.md`) for design patterns. Write all scripts, references, templates. Write the SKILL.md with proper frontmatter and comprehensive instructions. Test all scripts. 5. **Validate** — Run `python3 /home/ubuntu/skills/skill-creator/scripts/quick_validate.py $SKILL_NAME`. Fix any errors and re-validate. 6. **Confirm** — Inform the user that Stage 1 is complete. Show the skill structure. **Exit**: Skill passes `quick_validate.py` and all scripts execute without errors. --- ## Stage 2: prompt-engineer — Deep Optimization **Read**: `/home/ubuntu/skills/prompt-engineer/SKILL.md` Execute the prompt-engineer's complete workflow. Replace `$SKILL_NAME` with the actual skill name throughout. 1. **Prepare** — Copy the skill's SKILL.md to a temp file: ```bash cp /home/ubuntu/skills/$SKILL_NAME/SKILL.md /tmp/pe_input_$SKILL_NAME.txt ``` 2. **Verify flags** — Check the script's supported arguments: ```bash python3 /home/ubuntu/skills/prompt-engineer/scripts/optimize_prompt.py --help ``` Adjust the flags in step 3 if any are not supported. 3. **Optimize** — Run at deep depth: ```bash python3 /home/ubuntu/skills/prompt-engineer/scripts/optimize_prompt.py \ --file /tmp/pe_input_$SKILL_NAME.txt \ --depth deep --show-analysis --show-comparison \ -o /tmp/pe_output_$SKILL_NAME.md ``` 4. **Synthesize** — Read the optimized output. Rewrite the SKILL.md incorporating the best improvements. Preserve all script paths, command references, and technical accuracy. Do not blindly replace — merge intelligently. 5. **Validate** — Re-run `quick_validate.py` to confirm the optimized SKILL.md passes. 6. **Clean up** — Remove `/tmp/pe_input_$SKILL_NAME.txt` and `/tmp/pe_output_$SKILL_NAME.md`, even if earlier steps failed. **Exit**: SKILL.md is rewritten with prompt-engineer improvements and passes validation. --- ## Stage 3: master-skill-orchestrator — Full Triage **Read**: `/home/ubuntu/skills/master-skill-orchestrator/SKILL.md` Execute the orchestrator's complete Skill Triage Protocol: 1. **Domain Classification** — Identify ALL domains the new skill touches using the orchestrator's domain table. 2. **Complexity Assessment** — Classify as simple/moderate/complex/wicked. 3. **Phase Classification** — Determine which phases the skill covers (Understand → Reflect). 4. **Skill Selection** — Using the Priority Stack, Common Task Patterns, Prerequisite Chains, and Emergent Combinations, identify additional installed skills whose knowledge should inform the new skill. 5. **Apply insights** — If relevant skills or patterns are found, read those skills and incorporate their knowledge into the SKILL.md or reference files. 6. **Anti-Pattern Check** — Verify no Skill Overload, Tunnel Vision, Cargo Culting, or Skipping. 7. **Document** — Save the triage analysis to `/tmp/triage_$SKILL_NAME.md` with domain classification, complexity level, phase classification, and any insights applied. **Exit**: Triage document exists at `/tmp/triage_$SKILL_NAME.md` with all required sections. --- ## Stage 4: skill-connection-map — Connection Discovery **Read**: `/home/ubuntu/skills/skill-connection-map/SKILL.md` Execute the Connection Discovery Protocol: 1. **Name the problem domain** — What field is the new skill obviously in? 2. **Name three distant domains** — What fields have nothing to do with this skill? 3. **Find one skill per distant domain** — What installed skill has a structural analogy? 4. **Apply distant frameworks** — Translate the distant skill's concepts into the new skill's domain. 5. **Evaluate all seven connection types** — Structural Analogy, Inverse/Complement, Prerequisite Chain, Cross-Domain Transfer, Emergent Combination, Tension/Dialectic, Recursive/Fractal. 6. **Incorporate discoveries** — Apply any valuable cross-domain insights to the SKILL.md or references. 7. **Document** — Save the connection analysis to `/tmp/connections_$SKILL_NAME.md` listing all seven connection types evaluated and any improvements applied. **Exit**: Connection document exists at `/tmp/connections_$SKILL_NAME.md` with all seven types evaluated. --- ## Stage 5: skill-debugger — Full Debug Cycle **Read**: `/home/ubuntu/skills/skill-debugger/SKILL.md` This is the most time-intensive stage (2-5 minutes per run, potentially multiple cycles). Set expectations accordingly. Execute the skill-debugger's complete workflow: 1. **Run deep analysis**: ```bash python3 /home/ubuntu/skills/skill-debugger/scripts/debug_skill.py $SKILL_NAME --deep ``` Wait for completion. 2. **Read the full report** — Read `/home/ubuntu/skills/$SKILL_NAME/DEBUG_REPORT.md`. Understand every finding. 3. **Apply ALL fixes** — Fix in severity order (critical → low). Fully rewrite affected files — no partial edits. 4. **Re-run debugger** — Verify fixes. If new issues emerge, fix those too. **Maximum 3 debug cycles.** If critical/high issues persist after 3 cycles, list all unresolved issues, inform the user, and ask whether to proceed to Stage 6 or abort the pipeline. Do not proceed silently. 5. **Clean up** — Remove `/home/ubuntu/skills/$SKILL_NAME/DEBUG_REPORT.md` and `/home/ubuntu/skills/$SKILL_NAME/.debug_raw.json`. 6. **Final validation** — Run `quick_validate.py` one last time. **Exit**: No critical/high severity issues (or user confirmed proceed after 3 cycles), passes `quick_validate.py`. --- ## Stage 6: skill-sync — Full Deployment **Read**: `/home/ubuntu/skills/skill-sync/SKILL.md` Execute the full deployment. Replace `$SKILL_NAME` with the actual skill name. Verify the name is correct before executing any destructive commands. 1. **Push to GitHub**: ```bash # Ensure repo exists if [ ! -d /tmp/universal-ai-skills-library ]; then gh repo clone onfire7777/universal-ai-skills-library /tmp/universal-ai-skills-library fi cd /tmp/universal-ai-skills-library git pull --rebase || { git rebase --abort; git pull --no-rebase; } # Guard against empty skill name SKILL_NAME="the-actual-skill-name" if [ -z "$SKILL_NAME" ]; then echo "ERROR: SKILL_NAME is empty"; exit 1; fi # Deploy skill rm -rf "${SKILL_NAME}" cp -r "/home/ubuntu/skills/${SKILL_NAME}" . git add -A "${SKILL_NAME}/" git diff --cached --quiet || git commit -m "feat: ${SKILL_NAME} - created via ultimate-skill-creator pipeline" git push || { echo "ERROR: git push failed"; exit 1; } ``` If `git push` fails, report the error to the user with the git error message and suggest manual resolution. 2. **Run skill-sync** — Verify installation: ```bash python3 /home/ubuntu/skills/skill-sync/scripts/sync_skills.py ``` 3. **Verify** — Confirm the skill exists at `/home/ubuntu/skills/$SKILL_NAME/SKILL.md`. 4. **Clean up** — Remove `/tmp/triage_$SKILL_NAME.md` and `/tmp/connections_$SKILL_NAME.md`. **Exit**: Skill pushed to GitHub and confirmed installed locally. --- ## Completion Deliverables After all six stages, deliver to the user: 1. **Attachment**: Final `SKILL.md` file 2. **Summary table**: | Stage | Skill | Key Actions | Result | |---|---|---|---| | 1 | skill-creator | [what was created] | [validation result] | | 2 | prompt-engineer | [improvements applied] | [improvement score] | | 3 | master-skill-orchestrator | [domains/skills identified] | [insights applied] | | 4 | skill-connection-map | [connections discovered] | [improvements made] | | 5 | skill-debugger | [issues found/fixed] | [final health] | | 6 | skill-sync | [deployed] | [commit ref] | 3. **Key metrics**: Prompt-engineer improvement score, debugger findings resolved, files created, GitHub commit SHA 4. **Reflection**: Brief summary of what was learned during the pipeline — what worked well, what was challenging, and any insights for future skill creation ## Constraints - NEVER skip or combine stages. Each skill runs fully and independently. - NEVER partially edit files during debug fixes. Fully rewrite affected files. - Maximum 3 debug cycles in Stage 5 to prevent infinite loops. - Sequential execution only — no concurrent stages. - Clean up all temporary files after each stage, including on failure. - Preserve all original script paths, commands, and technical details exactly. - Verify $SKILL_NAME is set correctly before any destructive operations.