メニュー
Advanced semantic vocabulary, holographic contracts, and structural patterns for OCTAVE. REQUIRES octave-literacy. Extends literacy with mythology, archetype annotation, v6 contracts, and anti-pattern rules.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | octave-mastery |
| description | Advanced semantic vocabulary, holographic contracts, and structural patterns for OCTAVE. REQUIRES octave-literacy. Extends literacy with mythology, archetype annotation, v6 contracts, and anti-pattern rules. |
| allowed-tools | ["Read","Write","Edit"] |
| triggers | ["octave architecture","agent design","semantic pantheon","advanced octave","OCTAVE mastery","holographic patterns","archetypes","high-density specifications","system architecture","holographic contracts","archetype annotation"] |
| version | 3.2.0 |
STATUS::["ACTIVE"∧ENUM[ACTIVE,DRAFT,DEPRECATED]→§LIFECYCLE] PRIORITY::["HIGH"∧ENUM[HIGH,MEDIUM,LOW]] LATENCY::["<200ms"∧REGEX[<\d+ms]]
CONSTRAINT_TYPES:
CORE::[
REQ,
OPT,
CONST,
ENUM,
TYPE,
REGEX,
DIR,
APPEND_ONLY
]
EXTENDED::[
RANGE,
MAX_LENGTH,
MIN_LENGTH,
DATE,
ISO8601
]
§4c::CONTRACT_BLOCK_EXAMPLE
// Full META block using CONTRACT for a typed document
EXAMPLE:
```
===MY_DOCUMENT===
META:
TYPE::DECISION
VERSION::"1.0.0"
CONTRACT::HOLOGRAPHIClatest@localSTATUS::["ACTIVE"∧ENUM[ACTIVE,DRAFT,DEPRECATED]]
OWNER::["team-name"∧REQ]
DECISION::["adopt microservices"∧REQ]
===END===
§5::BLOCK_NOTATION_RULE // Hierarchical content MUST use block notation RULE::"Nested structures (maps containing maps) MUST use BLOCK notation: single colon + indented children" NEVER::"Inner value is itself an inline map: KEY::[outer::[inner::val]] — error E_NESTED_INLINE_MAP in strict mode, warning W_NESTED_INLINE_MAP in lenient mode" PRIMARY_CASE::"Agent §2::BEHAVIOR definitions — CONDUCT, PROTOCOL, OUTPUT blocks require block notation" CORRECT:
CONDUCT:
TONE::"Precise"
PROTOCOL:
MUST_ALWAYS::[rule_a, rule_b]
```
WRONG::"CONDUCT::[TONE::"Precise", PROTOCOL::[MUST_ALWAYS::[rule_a]]]"
RECOVERY_ON_E_NESTED_INLINE_MAP::"Fix is BLOCK form — NOT flatten-to-scalars (creates FLAT_PREFIX_SCALARS anti-pattern, §6), NOT collapse-to-blob (creates W_SNAKE_CASE_BLOB). The wrong escapes produce a different error class, not a solution."
§6::ANTI_PATTERNS
// Each has a concrete example of what NOT to do
ISOLATED_LIST::"[auth, payments, users] with no relationships — use DECISION::microservice_extraction[auth⊕payments→independent_services] instead"
FLAT_HIERARCHY::"All keys at top level with no grouping — group related keys under a parent BLOCK"
BURIED_NETWORK::"RELATED_TO::other_service hidden in prose comment — use explicit operator: auth→payments[dependency]"
OPERATOR_SOUP::"RESULT::A+B->C~D all in one expression — break into separate keyed fields"
PROSE_BLEED::"Stopword-laden English sentences as values — quoted telegraphic phrases ARE valid (see §6a). Ban applies when operators could compress the same semantics."
INLINE_ARRAY_ROOT::"TOKEN::[KEY::v, KEY2::v2] — inline map used as multi-field token root. Non-canonical for map tokens even when values are atomic; BLOCK form is mandated. Triggers E_NESTED_INLINE_MAP as soon as any child value needs nesting (§5::BLOCK_NOTATION_RULE). Fix: use BLOCK form (TOKEN: + indented children) from the start."
FLAT_PREFIX_SCALARS::"PARENT_CHILD::v, PARENT_CHILD2::v2 — flattened hierarchy via key name prefixes instead of BLOCK nesting. Destroys grouping and LLM attention. Fix: group under a PARENT: block with CHILD::v children."
§6a::TELEGRAPHIC_PHRASE
DEFINITION::"Quoted value with stopwords dropped; operators ⊕ ⇌ ∧ ∨ → carry relational meaning English connectives would spell out"
RULES::[
"Drop stopwords (is, at, for, the, a, with, by, of) — operators carry connectives",
"Use ⇌ for tension ('security ⇌ usability' not 'security at odds with usability')",
"Use → for causality ('migration → conflict' not 'migration causes conflict')",
"Preserve named entities, thresholds, IDs — never compress these",
"Use ∧ for joint conditions, ∨ for alternatives, ⊕ for emergent synthesis"
]
EXAMPLE_PAIR:
BEFORE::""natural language at odds with OCTAVE because stopwords" (~13 tokens)"
AFTER::""natural language ⇌ OCTAVE → stopword overhead" (~5 tokens)"
WHY::"operators are parse-efficient for LLM attention — same fidelity, lower token cost"
SEE_ALSO::"octave-compression §4::R3a for full rule set"
§7::TIER_NORMALIZATION_AUDIT_CHANNEL
// ADR-0006 SR1-T1 Step 3 (v1.12.0): centralised audit channel for I4 completeness
MODULE::"octave-mcp:src/octave_mcp/core/grammar/tier_normalize.py"
// ^ path is in the octave-mcp repo (upstream OCTAVE implementation), not this repo.
API:
LOG_REPAIR::"log_repair(log, rule_id, before, after, *, safe=True, semantics_changed=False) — single precise entry point appending a RepairTier.NORMALIZATION entry"
ACTIVE::"active(log) context manager — binds log as the ContextVar-scoped sink so pipeline-internal sites (notably emitter identifier-dequoting) can record receipts without threading RepairLog through emit()'s public signature"
RECONCILE::"reconcile_canonical_emission(log, baseline_bytes, canonical_bytes) — reconciler bridge"
RULE_IDS:
PRECISE::"TN_IDENTIFIER_DEQUOTE — was_quoted-driven dequoting at emit_assignment"
BRIDGE::"TN_RECONCILE_CANONICAL — coarse-grained post-emit baseline-vs-canonical receipt"
RECONCILER_BRIDGE_PATTERN:
PURPOSE::"Closes the audit-cardinality gap for diffs that upstream precise loggers do not yet cover (blank-line stripping pending Sprint 3+ trivia population; triple-quote collapse pending new lexer W-code)"
DEDUP_DISCRIMINANT::"Only RepairTier.NORMALIZATION entries on the log suppress the bridge. RepairTier.REPAIR (schema repairs) and RepairTier.FORBIDDEN entries do NOT count — they are orthogonal dimensions"
SELF_DEPRECATION::"When Sprint 3+ trivia and the new triple-quote lexer W-code land, precise loggers will cover their respective diffs, the dedup precondition fails, and the reconciler no-ops without any code change"
INTEGRATION_POINTS:
EMITTER::"emit_assignment: precise log when assignment.was_quoted is True AND emitter chose to emit bare (identifier-shape dequoting)"
WRITE_PY::"octave-mcp:src/octave_mcp/mcp/write.py — tier_normalize.active(tier_normalize_log) wraps each _emit_with_style call; reconcile_canonical_emission runs after final emit; entries drain into result[corrections]"
===END===