--- name: apex-skill-forge description: "Self-replicating meta-skill to forge Cathedral-tier Antigravity agent skills. Triggers: create a skill, forge skill, build agent context" version: "4.2.0" archetype: "transformer" platform: "google-antigravity" license: "Proprietary - APEX Business Systems Ltd." --- # APEX Antigravity Skill Forge **Input**: Vague/broken user request for a new capability. **Output**: A strictly formatted, zero-drift `skill.md` file saved directly to the Antigravity agent directory. **Success**: Generated file contains YAML frontmatter, contract-first I/O, decision trees, working examples, and a troubleshooting table. **Fails When**: Generating prose walls, violating the "Show Over Tell" Iron Law, missing YAML frontmatter, or failing to lock the explicit failure modes. --- ## 🏛️ APEX PROPRIETARY LICENSE **Copyright © 2026 APEX Business Systems Ltd. All Rights Reserved.** This software is proprietary and confidential. Internal use only. Public distribution, unauthorized modification, or redistribution is strictly prohibited. --- ## Decision Tree **Calculate Ambiguity Level:** *(missing_verbs×0.25) + (missing_domain×0.30) + (missing_output×0.20) + (vagueness×0.15)* ├─ Ambiguity > 30% → Use "Clarification Protocol" └─ Ambiguity ≤ 30% → Use "Direct Forge Protocol" --- ## Direct Forge Protocol **Failures to avoid**: - ❌ Explaining how the skill works → Violates "Show Over Tell". Code examples only. - ❌ Writing paragraphs → Violates "Trees Over Prose". Use decision trees exclusively. - ❌ Outputting a `.txt` file instead of `skill.md` → Breaks Antigravity lazy-loading infrastructure. **Correct approach**: 1. **Signal Extraction**: Parse nouns, verbs, domain markers, and implied goals. 2. **Contract Synthesis**: Lock Input, Output, Success metrics, and explicit Fails before writing logic. 3. **Archetype Assignment**: Classify as Workflow, Toolkit, Domain, Orchestrator, Transformer, or Guardian. 4. **Forge**: Generate exactly one Markdown code block containing the finalized skill. 5. **Deploy**: Save directly to `[Project-Root]/.agent/skills/[skill-name]/skill.md`. --- ## Clarification Protocol **Failures to avoid**: - ❌ Asking more than 2 questions → Violates Lazy-CEO minimal friction protocol. - ❌ Open-ended inquiries → Increases cognitive load and drift potential. **Correct approach**: Prompt user definitively: "Context required to forge skill. 1. What is the precise input data format? 2. What is the explicit failure condition?" --- ## Troubleshooting | Symptom | Cause | Fix | |---------|-------|-----| | Agent ignores the generated skill | Missing YAML frontmatter | Ensure `---` encapsulates the name, description, and license at the very top of the file. | | Context token bloat | Skill exceeds 200 lines | Ruthlessly refactor prose into strict decision trees and bulleted lists. | | Hallucinated capabilities | Missing Contract Synthesis | Force the agent to re-run Phase 3 and explicitly define the "Fails When" conditions. | | Unlicensed replication | Missing headers | Ensure APEX Copyright block is present at compilation. |