Amm recall

Skill frankenmason/amm-skills/amm-recall

Selective curated long-term memory skill bundle for agentmemory MCP. Works with Claude Code, Codex, Cursor, Gemini CLI, Copilot, Antigravity.

Install

npx -y skills add frankenmason/amm-skills --skill amm-recall

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Retrieve curated memory entries (directives, branch points, pendings, resume points) from agentmemory MCP to restore cross-session context. Trigger at session start, when the user asks "what did we leave off with?", or when the agent needs to check a past decision before acting.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.8 KB, 814 tokens by cl100k_base, as published. Nobody here has run it

AMM — Recall

Purpose

memory_save is useless without a disciplined recall pattern. This skill defines when and how to query memory_recall / memory_sessions / memory_smart_search so prior context surfaces at the right moments — session start, mid-session decision lookup, and before repeating work.

When to Trigger

Session start — always check for a resume point first:
```
memory_recall(query="resume_point", limit=3)
```
User asks "what were we doing?", "do we have a rule for X?", "did we decide on Y?".
Before repeating a decision — one recall to confirm no existing directive contradicts the new plan.
As a later step in a broader search workflow, when keyword search (e.g., qmd) and semantic search (vector store) did not surface the entry.

Apply Conditions

You can articulate a specific query (keyword or topic). Blind recall is noise.
You have reasonable concepts to query — "latest pending" is too broad; "pending P0 auth" is usable.

Execution Strategies

Strategy 1 — Session resume

memory_recall(query="resume_point", limit=3)
# Take the most recent entry, read the [next] field, execute it.
# Then open the files listed in [read] for context restore.

Strategy 2 — Topic-specific lookup

memory_recall(query="<topic> <concept_tag>", limit=10)
# e.g. "pending P1 caching"
# e.g. "user_directive python dependencies"

Strategy 3 — Semantic search (if available)

memory_smart_search(query="<natural-language query>", limit=5)
# Useful when you don't know the exact concept tags.

Strategy 4 — Session inventory

memory_sessions()
# Lists prior sessions with observation counts. Useful for picking which
# session's context to restore when resuming a multi-session project.

Interpreting Results

content — parse for the directive / pending / resume block.
concepts — filter by tag (keep only user_directive, drop pending, etc.).
files — open these in the order given.
Timestamp — prefer recent entries for volatile state; older entries may be stale.

Warnings

Recall may lag writes. Some builds return [] immediately after a save. If verification matters this turn, grep the storage file directly:
```
grep -F "<unique phrase>" ~/.agentmemory/standalone.json
```
Do NOT use AMM recall for bulk retrieval (1000+ entries). Use a proper observation system (claude-mem, qmd, a vector store) for that.
An empty memory_sessions() does not necessarily mean there are no entries — some builds separate raw save entries from session objects.

Example — Session start

# Turn 1
memory_recall(query="resume_point", limit=3)
# → top result: content includes "[next] run rollback test: psql staging < ..."

# Turn 2
# Open the files from [read]:
Read("/workspace/migrations/0042_down.sql")
Read("/workspace/docs/migration-playbook.md")

# Turn 3
# Execute the [next] action.

amm — umbrella (Bootstrap, references, license)
amm-save-resume — the writer side
amm-save-directive — directives that recall finds
amm-save-pending — pendings that recall finds

Change Log

1.0.0 — Initial release.