skills.rest
jingtao-lbl
jingtao-lblCOMMUNITY
Explore all skills by this author →
GitHub

inject-knowledge

Inject human-originated domain facts into A2MC's curated knowledge base channels.

Updated Aug 28, 2026
Search Tags:#yaml#calibration#graphrag#knowledge injection#curated knowledge#environmental modeling#rag graph
One-click install
npx skills add https://github.com/jingtao-lbl/A2MC --skill inject-knowledge-jingtao-lbl
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: inject-knowledge
Source: https://github.com/jingtao-lbl/A2MC/tree/main/.claude/skills/inject-knowledge
Command: npx skills add https://github.com/jingtao-lbl/A2MC --skill inject-knowledge-jingtao-lbl

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Human-originated domain insights (site discoveries, parameter knowledge, mechanism relationships) often fail to reach the A2MC calibration agent because they are placed in the wrong channel, fail validation, or never trigger a graph rebuild. This Skill guides the correct placement, validation, and rebuild of curated knowledge so the agent actually surfaces it during calibration. ## Core Features & Use Cases - Three-channel placement: Routes a fact into site discoveries.json, generic parameters.json, and/or curated_relationships.yaml depending on which trigger paths should surface it. - Validation and verification gates: Enforces JSON/YAML syntax checks, honest verified/verified_by flags, and memory smoke tests before commit. - Graph rebuild and audit: Triggers model-specific graph rebuilds (FATES, EcoSIM, PFLOTRAN), verifies edge counts, and records the injection with its evidence basis. - Use Case: After reading a paper revealing that only canopy-PFT clumping index affects understory radiation, inject that asymmetry warning into parameters.json and the curated YAML so the agent stops tuning an inert parameter. ## Quick Start Ask the agent to inject a specific finding or parameter insight into A2MC's curated knowledge base, naming the fact, its evidence source, and the affected targets or parameters.

Frequently Asked Questions about inject-knowledge

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

FAQPage Schema
How do I add a domain finding to the A2MC knowledge base?▼

Add the fact to the appropriate channels: site-specific findings go in the case's discoveries.json, parameter knowledge goes in memory/gained_knowledge/parameters.json, and mechanism relationships go in curated_relationships.yaml. Then validate the files and rebuild the RAG graph so the agent can retrieve it.

What is the difference between inject-knowledge and curate-knowledge?▼

inject-knowledge handles human-originated facts such as literature insights, while curate-knowledge promotes run-originated proposals staged by the autonomous agent. Both write to the same curated stores and are held to the same evidence and verification bar.

Why is my injected knowledge not surfaced by the calibration agent?▼

Common causes are a missing channel (the fact was not placed in every applicable store), a skipped graph rebuild after YAML edits, or silently dropped edges where a referenced parameter or output does not exist. Check the build log for 'skipped edge: endpoint not found' and verify the edge count rose as expected.

When can I set verified to true on a knowledge entry?▼

Set verified to true only with a verified_by link to a confirming Phase-5 test, experiment id, or topic stem. A Phase-3 diagnosis or Phase-4 hypothesis remains verified: false regardless of supporting evidence, and MemoryManager raises if verified is true without verified_by.

Does the graph rebuild command differ between models?▼

Yes. FATES uses build_rag_index.py --rebuild --graph-only, EcoSIM uses build_ecosim_rag.py --graph-only, and PFLOTRAN has no --graph-only option so it requires a full rebuild. Adapter models also store the curated overlay at models/<model>/curated_seed.yaml rather than rag/data/.

Related Skills

Explore More Education & Research Skills→
diendhdiendh

research-expert

Gather structured research outputs with sources across domains.

Community
Advanced
wgpsecwgpsec
1.7k

cot-injection

Tests Chain-of-Thought and ReAct agent reasoning chains for injection and manipulation vulnerabilities.

Official
Intermediate
jay6697117jay6697117

agency-agentic-identity-trust-architect

Design and deploy cryptographic identity, authentication, and trust verification for autonomous agents.

Community
Advanced
zengbaochengzengbaocheng

agentmail

Operate a dedicated email inbox to send, receive, and organize messages autonomously.

Community
Intermediate