sage-igv

name	sage-igv
description	AI-native access to Sage's IGV engine. Use when an agent needs to inspect current intents, generated artifacts, verification state, state graph, data sources, or briefing context through MCP/CLI/HTTP instead of a UI.
version	0.1.0
metadata	{"openclaw":{"emoji":"◎","homepage":"https://github.com/EvanL1/sage","tags":["igv","cli","status","agent"]}}

Sage IGV

Use this skill when you need Sage's product state without opening a UI.

IGV means:

• Intent: what was intended, usually open tasks or explicit user goals.
• Generation: what Sage or agents produced, such as reports and high-signal memories.
• Verification: what still needs trust checking, plus recent outcomes and evidence refs.

Query Order

1. Prefer Sage MCP tools when the host exposes them:

• sage_igv_status
• sage_list_intents
• sage_list_generations
• sage_list_verifications
• sage_run_command
• sage_schema
• sage_search
• sage_record_observation

For Claude Code in this repo, the project .mcp.json starts Sage with:

cargo run -q -p sage-cli -- mcp serve

2. If MCP tools are unavailable, use the local CLI:

sage status --json
sage schema notion
sage notion push all --dry-run --limit 100
sage review --json
sage /sage:review --json
sage igv --json

3. Narrow the query when possible:

sage igv intent list --json
sage igv generation list --json
sage igv verification list --json
sage igv graph --json
sage igv sources --json
sage igv brief --json

4. If the light runtime server is already running, HTTP is equivalent:

curl -s http://127.0.0.1:18522/api/status
curl -s http://127.0.0.1:18522/api/schema/notion
curl -s http://127.0.0.1:18522/api/igv/status
curl -s "http://127.0.0.1:18522/api/command?command=/sage:review"
curl -s "http://127.0.0.1:18522/api/search?query=Sage&limit=5"

How To Use Results

• Treat ref_id as the stable handle in your response or follow-up action.
• Treat evidence_ref as required context for any trust-sensitive claim.
• Do not infer that a generation is true just because it exists; check verification.
• If verification has pending items, surface them before recommending irreversible action.
• Use synthesis.briefing.headline for short status answers.

Write Paths

Use existing Sage write commands instead of editing the database:

sage mcp serve
sage serve
sage learn "fact to remember"
sage observe "behavior or event to record"
echo "raw context" | sage pipe "source label"

For task, report, or memory changes that are not exposed by CLI or runtime HTTP, use the existing Sage API layer rather than inventing a new storage path. Treat Tauri as an optional surface, not the owner of Sage lifecycle.