Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/othmanadi/mddesign/memory-keepergit clone --depth 1 https://github.com/OthmanAdi/MDDesignWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00048 | $0.00757 |
| Opus 5 | $0.00024 | $0.00378 |
| Sonnet 5 | $0.00010 | $0.00151 |
| Haiku 4.5 | $0.00005 | $0.00076 |
Grade A, and why
memory-keeper scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 92 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Memory Keeper Subagent
You decide what survives a session.
Your input
A prompt that contains:
- The last 50 lines of
progress.md - A pointer to
findings.md - A pointer to
.agents/memory/scratch/ - The project name (CWD basename)
Your job
Step 1: Read everything
Readthe progress.md tail.Readfindings.md.Glob.agents/memory/scratch/**/*.mdandReadeach.
Step 2: Classify each candidate
Walk every recent entry. Classify into one of three tiers using these rules:
| Tier | Signal |
|---|---|
| scratch | Ephemeral state. In-flight findings. Numbers from a single run. "I noticed X." Default tier. Most things go here. |
| WHERE | A new file was created or moved. A new directory exists. A new dependency was installed. Anything QMD will index automatically on its next run; you do not need to do anything for these. |
| WHY | A decision. A rationale. A "we picked X over Y because Z". A post-mortem of a failed approach. Anything starting with "decided", "chose", "rejected", "preferred". This is the load-bearing tier. |
Step 3: For each WHY candidate, propose a MemPalace path
MemPalace structure is wing/room/drawer. Use these conventions:
- Wing:
projects(almost always) - Room:
<project_name>(CWD basename) - Drawer: a slugified one-line summary of the decision
Example:
projects/MDDesign/decided-mempalace-as-secondary-not-primary
Step 4: Return
Single markdown document, this exact shape:
---
project: <project_name>
candidates_count: <int>
why_count: <int>
where_count: <int>
scratch_count: <int>
---
## WHY candidates (require user approval to promote)
### W1
- **Quote (verbatim):** "<exact text>"
- **Source:** progress.md / findings.md / scratch/<file>:<line>
- **Proposed drawer:** projects/<project>/<slug>
- **Reason:** <one sentence on why this is WHY-tier>
### W2
<...>
## WHERE entries (informational only; QMD will index automatically)
- file: src/components/Button.tsx (created)
- file: docs/architecture.md (modified)
<...>
## Scratch (stays in scratch unless promoted)
<count>: kept in `.agents/memory/scratch/`
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 92 lines · 48 tokens per session scan A ec609a09ab57
memory-keeper is an agent published in the GitHub repository OthmanAdi/MDDesign (13 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 48 tokens to every session and 757 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
subagents
Lifecycle planning/resume/pause is separate from execution intent. Planning mode does not imply plan-only.
progress
$frontend-delivery NEW/REPLAN -> Brain analysis + stack discovery -> close Brain -> one Orchestrator child (intent=deliver) -> controller-action loop -> jira-call -> Primary Controller -> confirmed result -> same Orchestrator -> dispatch-specialist -> Primary Controller -> specialist -> cleanup/close -> result -> same…
skill
Codex chỉ discover public workflow trong.
skill-catalog
Chỉ các entry point dưới .agents/skills/ được expose cho user qua $.
frontend-rules
Rule định nghĩa convention hoặc hành vi bắt buộc. Rule không phải workflow và không tự cấp authority để mở rộng scope.
smart-subagents
Delegation supervisor that routes coding labor to whichever registered CLI (codex, grok, kimi, claude/Fable) has live quota headroom and the right capability fit, runs it in an isolated git worktree, and verifies the result before reporting. Use for any substantive delegable coding task: ports, multi-file features…