multimedia-intelligence-forge
Ingest, transcribe, and distill multimedia lectures, technical videos, and literature into deep agent skills and cryptographically verified Knowledge Vault items. Do not use for generic audio/video editing or entertainment media consumption.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- multimedia-intelligence-forge
- description
- Ingest, transcribe, and distill multimedia lectures, technical videos, and literature into deep agent skills and cryptographically verified Knowledge Vault items. Do not use for generic audio/video editing or entertainment media consumption.
02 资源文件 4
- 参考文档CARD.md3.6 KB
- 参考文档config.default.yaml758 B
- 参考文档references/multimodal-distillation.md1.1 KB
- 脚本scripts/multimedia_distiller.py8.4 KB
03 SKILL.md 正文
Multimedia Intelligence Forge: Audio, Video & Knowledge Distillation
multimedia-intelligence-forge is an authoritative composite meta-skill that transforms unstructured multimedia assets—YouTube lectures, technical keynotes, conference recordings, and multimedia literature—into production-grade, architecturally deepened agent skills and cryptographically verified Knowledge Vault items.
It coordinates five specialized capabilities:
- Subprocess Transcript Ingestion (
youtube-transcript-fetcher) - Dual-Lens Cognitive Distillation (
media-mind-forge) - End-to-End Deep Skill Synthesis (
deep-skill-forge) - Cryptographic Isnad Verification (
epistemic-isnad-audit) - Epistemic Memory Promotion & Vault Retention (
epistemic-memory-lifecycle)
See CARD.md for the companion summary card, 5-stage progression matrix, and invariants checklist.
Consult /media-mind-forge for dual-lens extraction heuristics, /epistemic-isnad-audit for cryptographic claim verification, and multimodal-distillation.md for multimodal distillation heuristics.
The 5-Stage Multimedia Distillation Progression
[1. Transcript Ingestion] ──► [2. Dual-Lens Distillation] ──► [3. Consolidated Master Checkpoint]
│
▼
[5. Epistemic Vault Commit] ◄── [4. Deep Skill Scaffolding & Repair] ◄────┘
1. Multimedia Ingestion & Subprocess Transcription
Fetch and normalize spoken transcripts or written media via isolated subprocess pipes without shell injection vulnerabilities:
- Subprocess Audio/Transcript Acquisition (Rule 5 & Rule 15):
- Extract full timed captions or transcripts via isolated JSON-RPC runners.
- Prevent argument injection by passing sanitized video IDs/URLs directly to subprocess argv lists.
- Deterministic Pre-LLM Normalization (Rule 9):
- Deduplicate filler phrases, vocal tics, and sponsor breaks.
- Segment long-form transcripts into semantically coherent chapters anchored by timestamps.
- Primary Source Hashing:
- Compute the SHA-256 digest of the raw transcript text for immutable lineage tracking.
Completion criterion: Clean, timestamp-segmented transcript generated with SHA-256 digest recorded.
2. Dual-Lens Cognitive Distillation
Apply the Dual-Lens Cognitive Distillation protocol (Rule 41), strictly bifurcating the raw material into two distinct output streams:
- Lens A: Epistemic Introspection (Ground-Truth Beliefs):
- Extract core mental models, conceptual axioms, and philosophical heuristics.
- Pin each claim directly to an exact timestamp or line quote (chain-of-custody lineage).
- Lens B: Procedural Skill Synthesis (Shu-Ha-Ri Execution):
- Extract operational stages, binary completion criteria, diagnostic interview questions, and anti-patterns.
- Cognitive Salvage (Rule 42):
- Cleanse extracted JSON schemas through multi-pass repair (strip fences, fix trailing commas, unquote keys).
Completion criterion: Epistemic claims and procedural skill blueprints extracted and validated.
3. Visual Brief & Consolidated Checkpoint
Synthesize and present findings to the user before mutating persistent state:
- Interactive HTML Visual Brief:
- Render a dark-themed visual summary in
%TEMP%showing transcript metrics, extracted mental models, and target skill architecture.
- Render a dark-themed visual summary in
- Consolidated Master Checkpoint:
- Author
implementation_plan.mdwithRequestFeedback: truedetailing both proposed Knowledge Items and new agent skill files. - Await explicit user approval before proceeding to file creation.
- Author
Completion criterion: Interactive HTML visual brief rendered and user sign-off obtained at master checkpoint.
4. Deep Skill Scaffolding & Bounded Self-Repair
Generate the complete production skill package following deep-module standards:
- Slotted & Frozen Dataclass Domain Models (Rule 12):
- Author
@dataclass(slots=True, frozen=True)schemas with__post_init__invariant guards.
- Author
- Craft Standards Specification:
- Generate
SKILL.md(bounded description 100-350 chars with negative boundary, Rule 44). - Generate
CARD.mdwith single-pipe ASCII borders and blocking checklist invariants (Rule 37). - Generate
config.default.yamlfor 3-tier zero-fork configuration.
- Generate
- Bounded In-Flight Self-Repair:
- Execute contract tests in-flight; repair syntax or schema failures automatically up to 3 attempts.
Completion criterion: Compliant skill package scaffolded with 100% passing tests within $\le 3$ repair attempts.
5. Epistemic Isnad Lineage & Knowledge Vault Commit
Commit ground-truth learnings to persistent memory according to repository invariants:
- Canonical Dual-File Directory Format (Rule 40):
- Write
.harness/knowledge/<ki_id>/metadata.json(SHA-256, source URI, claims, verification status). - Write
.harness/knowledge/<ki_id>/summary.md(distilled narrative, mental models, decision heuristics).
- Write
- Ecosystem Context Map Registration:
- Register forged skills in
CONTEXT-MAP.mdunder their appropriate domain.
- Register forged skills in
- Skill Knowledge Graph Indexing:
- Index the new skill nodes and dependency edges into
skill_knowledge_graph.
- Index the new skill nodes and dependency edges into
Completion criterion: Canonical dual-file Knowledge Vault item committed, CONTEXT-MAP.md updated, and skill graph indexed.
The Three Foundational Pillars
1. The Visual Brief Pillar
Every distillation run renders an interactive HTML visual brief in %TEMP% displaying speech timelines, isnad claim matrices, and synthesized skill component graphs.
2. The Mandatory Checkpoint Pillar
The agent must never author skill files or mutate persistent memory without first presenting implementation_plan.md with RequestFeedback: true and awaiting explicit user confirmation.
3. Explicit Anti-Patterns
Rigid architectural boundaries prevent ungrounded transcript hallucinations, flat vault pollution, and unbounded repair loops.
Anti-Patterns
- Ungrounded Transcript Hallucination — Distilling claims or heuristics without verbatim quote evidence or timestamp references.
- Single-File Vault Pollution — Dumping single flat JSON files into
.harness/knowledge/instead of the canonical dual-file directory (metadata.json+summary.md). - Collapsing Dual-Lens Seams — Blending epistemic belief models and executable procedural skills into an undifferentiated prose blob.
- Unbounded Repair Thrashing — Retrying failing contract tests in an infinite loop without an explicit circuit breaker limit of 3 attempts.
- Unverified Shell Redirection — Piping transcript downloads through bare shell redirects instead of safe subprocess UTF-8 streams.
- Missing Negative Deflection Boundary — Scaffolding skills without explicit
Do not use for...constraints in frontmatter descriptions.
04 同仓库更多 Skills
ui-behavior-primitive-architect
Architect, audit, and verify zero-dependency headless UI behavior primitives, WebMCP agent bridges, WAI-ARIA/SSR/RTL tri-modal checks, and framework expert routing. Do not use for static CSS theme styling or backend database migrations.
trustgraph-ontology-governor
Govern knowledge graphs, SHACL-AF triple policies, OntologyRAG subclass validation, PROV-O dual named graphs, and Laya entropy decision scoring. Do not use for relational SQL queries or generic web scraping.
shim-harness-governor
Govern universal CLI agent loops, 4-tier tool permissions, ancestor-symlink workspace sandboxes, deterministic-first command routing, collision-safe MCP namespacing, and layered skill loading. Do not use for static UI styling or external SQL analytics.
graphiti-zep-memory-governor
Govern bi-temporal knowledge graphs with Graphiti future-edge clamping, MinHash/LSH entity dedup, RRF/MMR reranking, and Zep typed ontologies. Do not use for static SQL joins or generic web crawling.
agora-mcp-workbench-architect
Architect, audit, and orchestrate Agora Workbench MCP servers, loopback sidecars, unified state graphs, gateway/dispatcher connectors, and policy-enforced data lakes. Do not use for static frontend CSS styling or non-MCP desktop GUI automation.
skill-eval-assurance-runner
Evaluate and assure agent skills using small-N Wilson confidence intervals, unbiased pass^k reliability, deterministic trajectory graders, preflight boundary honesty, and sealed digest verification. Do not use for subjective prose review or unconstrained production deployment.