name: chl-memory
description: Use when working with the CHL MCP for everyday memory tasks: remember, recall, infer, learn, inspect lexicon, or inspect state/resources.
CHL Memory
Use this skill when the task is about storing, retrieving, or reasoning over CHL memory.
Default flow
- Use
chl_remember when the user gives new facts, decisions, preferences, or task context that should persist.
- Use
chl_recall when the user wants related memories, similar past items, or supporting context.
- Use
chl_infer when the user wants the best answer synthesized from stored memories instead of a raw list.
- Use
chl_learn when the user corrects, reinforces, or weakens an association.
Tool choice
- Keep
chl_recall queries short and specific.
- Use
topK only as high as needed.
- Put structured data in
payload and lightweight provenance in metadata.
- Use positive
reward to reinforce, negative reward to suppress, and 0 for neutral feedback.
Inspect when needed
Prefer the lightest inspection that answers the question:
chl_snapshot for a compact current view.
chl_profile for the active tuning preset.
chl_lexicon for learned concepts and phrases.
chl_state for a full diagnostic dump.
chl_entries when you need every stored entry.
chl_journal when you need mutation history.
chl_bucket_stats when debugging retrieval behavior.
Resources
Use MCP resources directly when a read-only view is enough:
chl://memory
chl://profile
chl://state
chl://entries
chl://journal
chl://lexicon
Guardrails
- Do not clear memory unless the user explicitly asks for a reset.
- Do not invent memory structure; mirror the user’s content and the MCP schema.
- Prefer one focused tool call over several broad ones.