ops-knowledge-ingest

Orchestrates ingestion of upstream documentation and operator repositories into an OKF knowledge base.

Updated Sep 15, 2026
One-click install
npx skills add https://github.com/WangWindow/CANN-BatchMatMulMaxsum --skill ops-knowledge-ingest-wangwindow
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ops-knowledge-ingest
Source: https://github.com/WangWindow/CANN-BatchMatMulMaxsum/tree/main/.agents/skills/ops-knowledge-ingest
Command: npx skills add https://github.com/WangWindow/CANN-BatchMatMulMaxsum --skill ops-knowledge-ingest-wangwindow

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Keeping a structured knowledge base of Ascend NPU / Ascend C operator development documentation in sync with upstream sources is error-prone: new sources need onboarding, upstream commits need incremental syncing, and major version upgrades need careful migration. This Skill provides the top-level orchestration for all three ingestion routes while delegating actual card authoring to per-tree production skills. ## Core Features & Use Cases - Three-route ingestion orchestration: Routes work to new-source onboarding, commit-level incremental sync, or major version upgrade pipelines based on the nature of upstream changes. - Shared graph engine: Ships okf_graph.py and okf_judge_aggregate.py scripts that build, judge, inject, visualize, and verify the cross-tree knowledge graph with fingerprint-based incremental re-judging. - Governance invariants: Enforces provenance discipline (single primary source per card, upstream @sha URLs), watermark-based landing semantics, and red-line rules across all routes. - Use Case: When the asc-devkit upstream repo advances from one commit to another, use this Skill to diff the range, adjudicate scope with the user, delegate card updates to the reference-ingest skill, rebuild the graph, and advance the watermark atomically. ## Quick Start Ask the orchestrator to sync the asc-devkit reference bundle from its recorded watermark commit to the latest upstream commit and rebuild the knowledge graph.

Frequently Asked Questions about ops-knowledge-ingest

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I sync a knowledge base with new upstream commits?

Use the commit-level incremental sync route: read the watermark sha from the bundle root index.md, run git diff between old and new sha, adjudicate scope with the user, delegate card updates to the production skill, then rebuild the graph and advance the watermark.

What is the difference between the three ingestion routes?

New source onboarding handles first-time ingestion of an upstream repo or doc bundle. Commit-level incremental sync handles same-ref commit advancement. Major version upgrade handles ref switches like 9.0.0 to 9.1.0 with possible architectural changes.

Does this skill write knowledge cards itself?

No. It is a pure orchestrator that holds the routing principles and shared graph engine scripts. Card authoring is delegated to per-tree production skills such as ops-knowledge-reference-ingest, ops-knowledge-vv-ingest, and ops-knowledge-cv-ingest.

How does the knowledge graph stay up to date after card changes?

After any card addition or modification, rerun the graph pipeline: candidates, judge, inject, viz, verify. Judgments are cached in graph/edge_judgments.json keyed by content fingerprints, so unchanged cards cost zero LLM calls.

What happens if the watermark commit is missing?

The process stops and asks the user for the baseline commit instead of guessing with HEAD~N or recent tags. The watermark is the only authoritative signal of what has been landed.