echo-memory-agent

name	echo-memory-agent
description	Scope Cursor Agent prompts for Echo-Memory development — memory families, entry scripts, project skills, and public-repo constraints. Use when vibe coding, writing .cursor/rules, or planning multi-file agent tasks in this repo.

Echo-Memory Agent development

Project skills (read before large tasks)

Skill	Path	When
Training	.cursor/skills/echo-memory-train/	Memory baselines, context rows, train/
Evaluation	.cursor/skills/echo-memory-eval/	eval/v2, HF checkpoint checks
Release & site	.cursor/skills/echo-memory-release/	gh-pages, i18n, checkpoints doc

Invoke by name in chat: e.g. use echo-memory-eval to add a context_k1 replay check.

Prompt template

Include in Agent requests:

1. Memory family — Context, Compression, Spatial, State-Space (or row id)
2. Entry script — e.g. train/.../run_spatial_memory_baseline.sh
3. Eval branch — replay / in-domain / open-domain
4. Pool — static in-domain vs dynamic training (DATASET_BASE_PATH)

Example prompts

Using echo-memory-eval: download context_k1 from Echo-Team/Echo-Memory and
run eval/v2/run_basic_replay_gt.sh with the static in-domain pool.

Using echo-memory-train: add OUTPUT_BASE_ROOT override docs to
run_ablation_block_wise_ssm_two_chunk.sh without changing defaults.

Trace env/memory_baseline_runtime.py for spatial_mem → inject flags;
update doc/checkpoints.md with a short mapping table.

Rules (optional)

Add .cursor/rules/echo-memory.mdc for standing constraints:

• Echo pool naming in public markdown
• No upload scripts, internal benchmark names, or /pfs/ paths in GitHub
• Minimal diff; match existing bash/Python patterns in train/ and eval/v2/

Modes

• Agent — multi-file implementation
• Ask — read diffsynth/, trace configs, no edits