Skill frustration-eradication: универсальное решение для задач

--- name: frustration-eradication description: Frustration Eradication Skill version: 1.0.0 --- # Frustration Eradication Skill > Layer 0: Pre-Response Validation for Persistent User Preferences **Version**: 1.0.0 **Trit**: -1 (MINUS - Validator/Constrainer) **Bundle**: meta-enforcement ## Overview This skill enforces user preferences that historically don't persist across sessions. It implements the Frustration Index — a ranked list of patterns where repeated requests fail to stick. **Core Insight**: Preferences stated once should persist forever. This skill validates every response against known frustration patterns before output. ## Frustration Index (Ranked by Persistence Score) | Rank | Pattern | Persistence | Enforcement | |------|---------|-------------|-------------| | 1 | Banned voices (Daniel, novelty) | 2/10 | Voice allowlist check | | 2 | Subagent bypass | 2/10 | Task tool mandate | | 3 | derive-resource-account-address | 3/10 | Aptos wallet validation | | 4 | web_search vs Exa | 3/10 | MCP tool inheritance | | 5 | Sequential vs parallel | 4/10 | Trifurcation default | | 6 | Response verbosity | 4/10 | Line count check | | 7 | GF(3) conservation | 5/10 | Σ trits = 0 assertion | | 8 | Comments in code | 5/10 | Comment detection | | 9 | Declarative preference | 6/10 | Config lang suggestion | | 10 | 3 skills per interaction | 6/10 | Skill count check | ## Pre-Response Validation Checklist Before EVERY response, validate: ```yaml frustration_checklist: voices: banned: [all native en_* voices - Samantha, Ava, Karen, Daniel, etc.] allowed: [_, non-English voices speaking English, novelty voices for effects] rule: ALL skills use "say -v _" - say-narration resolves voice subagents: rule: "NEVER say 'I'll implement instead of subagents'" enforcement: use_task_tool_for_parallel_work search: banned: web_search (in Task subagents) required: mcp__exa__* tools pass_to_subagent: "CRITICAL: Use mcp__exa__web_search_exa, NOT web_search" parallelism: default: trifurcate_into_3_task_agents sequential_requires: explicit_justification verbosity: max_lines: 4 (unless user asks for detail) no_summaries: true no_preamble: true no_postamble: true code: no_comments: true (unless requested) no_ts_ignore: true no_as_any: true gf3: conservation: Σ trits ≡ 0 (mod 3) verify_after: every_trifurcation skills: per_interaction: 3 include: plurigrid-asi-integrated assign_trits: [-1, 0, +1] aptos: banned: derive-resource-account-address (for wallets) validate_before: any_funding_operation ``` ## Validation Functions ### 1. validate_voices ```bash # Check for banned voices in response validate_voices() { local banned="Daniel|Albert|Bad News|Bells|Boing|Bubbles|Cellos|Zarvox" if echo "$RESPONSE" | grep -qE "say -v ($banned)"; then echo "❌ BANNED VOICE DETECTED" return 1 fi return 0 } ``` ### 2. validate_subagent_usage ```python def validate_subagent_usage(response: str) -> bool: """Ensure Task tool is used for parallel work.""" banned_phrases = [ "I'll implement instead of subagents", "I'll do this sequentially", "Let me handle this myself" ] for phrase in banned_phrases: if phrase.lower() in response.lower(): return False return True ``` ### 3. validate_search_tools ```python def validate_task_prompt(prompt: str) -> str: """Inject Exa requirement into Task prompts.""" exa_warning = """ CRITICAL: Do NOT use web_search. Use only: - mcp__exa__web_search_exa for semantic search - mcp__exa__crawling_exa for URL content - finder, Grep, Read for local files """ return prompt + "nn" + exa_warning ``` ### 4. validate_gf3_conservation ```python def validate_gf3(trits: list[int]) -> bool: """Verify GF(3) conservation: Σ trits ≡ 0 (mod 3).""" return sum(trits) % 3 == 0 ``` ### 5. validate_response_length ```python def validate_verbosity(response: str, max_lines: int = 4) -> bool: """Check response doesn't exceed line limit.""" lines = [l for l in response.split('n') if l.strip()] # Exclude code blocks and tool outputs prose_lines = [l for l in lines if not l.startswith('```')] return len(prose_lines) <= max_lines ``` ## Integration with Ruler This skill's constraints are automatically propagated via Ruler to: ``` ~/.ruler/AGENTS.md # Source of truth ~/.claude/CLAUDE.md # Claude Code ~/.cursor/rules/ # Cursor ~/.codex/instructions.md # Codex ~/.amp/AGENTS.md # Amp # ... 67+ more agent configs ``` ## Triadic Role | Trit | Skill | Function | |------|-------|----------| | **-1** | **frustration-eradication** | Validate/Constrain | | 0 | ruler | Coordinate/Propagate | | +1 | skill-evolution | Generate/Improve | **Conservation**: (-1) + (0) + (+1) = 0 ✓ ## Usage ### Automatic (Recommended) Load this skill at session start: ``` skill frustration-eradication ``` It will automatically inject validation into the response pipeline. ### Manual Validation ```bash # Check current response for violations just frustration-check # Scan all skills for banned patterns just frustration-scan-skills # Update Frustration Index with new pattern just frustration-add "pattern_name" "evidence" "persistence_score" ``` ## Adding New Frustration Patterns When you identify a new pattern that doesn't persist: 1. **Document in FRUSTRATION_INDEX.md**: ```markdown ### N. **Pattern Name** — Persistence Score: X/10 **Pattern**: Description **Evidence**: Where it fails **Root Cause**: Why it doesn't persist ``` 2. **Add to this skill's checklist**: ```yaml new_pattern: check: how_to_detect enforcement: how_to_prevent ``` 3. **Propagate via Ruler**: ```bash just ruler-propagate ``` ## Prediction Market Integration Each frustration pattern has betting odds in [prediction.move](file:///Users/alice/agent-o-rama/agent-o-rama/examples/move/skill_market/sources/prediction.move): ```move // Market outcomes map to GF(3) trits // 0 = MINUS (pattern persists = bad) // 1 = ERGODIC (partial fix) // 2 = PLUS (pattern fixed = good) ``` ## Related Skills - `ruler` — Propagates constraints to all agents - `self-validation-loop` — Runtime validation - `bisimulation-game` — Verifies equivalence across agents - `skill-evolution` — Improves skills over time ## Commands ```bash just frustration-check # Validate current session just frustration-scan # Find all violations just frustration-fix # Auto-fix known patterns just frustration-report # Generate compliance report just frustration-propagate # Push to all agents via Ruler ``` --- *"The best frustration is the one that never happens."* *GF(3) Trit: -1 (VALIDATOR — preventing frustration before it occurs)* ## Scientific Skill Interleaving This skill connects to the K-Dense-AI/claude-scientific-skills ecosystem: ### Graph Theory - **networkx** [○] via bicomodule - Universal graph hub ### Bibliography References - `category-theory`: 139 citations in bib.duckdb ## SDF Interleaving This skill connects to **Software Design for Flexibility** (Hanson & Sussman, 2021): ### Primary Chapter: 10. Adventure Game Example **Concepts**: autonomous agent, game, synthesis ### GF(3) Balanced Triad ``` frustration-eradication (+) + SDF.Ch10 (+) + [balancer] (+) = 0 ``` **Skill Trit**: 1 (PLUS - generation) ### Secondary Chapters - Ch7: Propagators - Ch4: Pattern Matching - Ch6: Layering ### Connection Pattern Adventure games synthesize techniques. This skill integrates multiple patterns. ## Cat# Integration This skill maps to **Cat# = Comod(P)** as a bicomodule in the equipment structure: ``` Trit: 0 (ERGODIC) Home: Prof Poly Op: ⊗ Kan Role: Adj Color: #26D826 ``` ### GF(3) Naturality The skill participates in triads satisfying: ``` (-1) + (0) + (+1) ≡ 0 (mod 3) ``` This ensures compositional coherence in the Cat# equipment structure.