griot-load

Griot Load

Render learnings/rollup.json to LLM prose and emit it into the current session, along with the citation contract that asks Claude to emit Applied: L-NNN (or AP-NNN) when it applies a learning or antipattern from the rollup.

This skill is a thin wrapper. All rendering, schema interpretation, and citation-contract content live in bin/griot use --as=llm. The skill exists as the addressable /griot-load user surface; automated loaders (like /ev-run's start-of-session step) call the CLI directly via Bash without composing this skill. Per the substrate-API- stability rule and the Phase 4 rollup pinned Decision: synthesis lives in the CLI, the skill is a name-stable handle.

Invocation

Run the CLI and return its stdout verbatim:

griot use --as=llm

The CLI handles all three load outcomes:

• loaded N learnings → status line + full rollup content + citation contract. Top of context for the rest of the session.
• rollup empty → short message, no citation contract. No promoted entries yet.
• no rollup yet → short message, no citation contract. No rollup file present (e.g., a fresh clone before any /griot-compact run).

A format-detection error fires if a mid-flight session encounters the legacy learnings/rollup.md without learnings/rollup.json — that's a Phase-4-rollup-cutover safety net per the substrate's whiteboard findings. The remedy is to run node .claude/scripts/migrate-rollup-md-to-json.ts and restart the session.

Do not

• Do not add synthesis to this skill body. Any prose or logic above what bin/griot use --as=llm already emits should land in the CLI itself, so other consumers (e.g. /ev-run's loader step) get it identically. Two skills with the same job and different names is a cohesion failure; one skill that wraps the CLI keeps the citation contract single-sourced.
• Do not invoke this skill automatically. It is opt-in per the learnings-system Principle 4 (Opt-in). The user types /griot-load deliberately.
• Do not read learnings/rollup.json directly from the skill body. Always go through bin/griot use --as=llm — the CLI owns the schema-to-prose render and the legacy-format detection error path.