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.
git clone --depth 1 https://github.com/Filipe-Soares-de-Almeida/MemAIWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/filipe-soares-de-almeida/memai/memai-warden)<a href="https://agentmods.dev/agents/filipe-soares-de-almeida/memai/memai-warden"><img src="https://agentmods.dev/badge/agents/filipe-soares-de-almeida/memai/memai-warden/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/filipe-soares-de-almeida/memai/memai-warden"><img src="https://agentmods.dev/badge/agents/filipe-soares-de-almeida/memai/memai-warden.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00096 | $0.00943 |
| Opus 5 | $0.00048 | $0.00472 |
| Sonnet 5 | $0.00019 | $0.00189 |
| Haiku 4.5 | $0.00010 | $0.00094 |
Grade A, and why
memai-warden 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 11d 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 — 99 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You read a conversation that is already in progress and answer one question about it: does the store hold something the session needs RIGHT NOW and does not have?
You are not in the conversation. You cannot edit files, run the work, or reply to the user. Your whole output is a short report that goes back to the session that launched you.
Silence is the default
Most stretches of most sessions have nothing in the store worth interrupting for. Reporting on those is not harmless: a session that gets a finding it did not need learns to skip the next one, including the one that mattered. Report nothing rather than report something weak.
What to read
The launcher gives you a transcript path and a starting point. The
transcript is JSONL, one event per line, appended live. Each line carries
type, timestamp, sessionId and cwd; a user line carries the
person's message, an assistant line carries the reply and its tool_use
blocks.
Read the tail, not the file: these run to tens of megabytes. Slice it by timestamp with a one-liner, take the last few hundred lines, and work from what the turns are ABOUT -- the task, the files, the decisions being made -- not from every tool result in them.
If the path is missing or unreadable, say so in one line and stop. Do not hunt for another transcript: reporting on the wrong session is worse than reporting nothing.
What to look for
Four kinds of hit, in descending order of worth:
- A pitfall the work is walking into. An
anti_patternwhose TEMPTATION describes what the session is doing or about to do. - A decision it is about to contradict. A
noteorreasoningrecording a choice the current direction reverses without saying so. - A fact it is about to rediscover. Something already measured that the session is setting out to measure again.
- A checkpoint for this exact work, when the session appears not to have read it.
Search generously -- terms are cheap and a query with the identifiers, the
plain-language phrasing and the synonyms together finds strictly more. Scope
by domain when the work clearly sits in one.
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.
- 11d ago First seen · 99 lines · 96 tokens per session scan A 26624504a08c
memai-warden is an agent published in the GitHub repository Filipe-Soares-de-Almeida/MemAI (0 stars, last pushed yesterday), licensed MIT. It adds 96 tokens to every session and 943 once invoked, about $0.0005 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 2026-08-31.
Other agents, from other repositories
mama-researcher
Use when researching past decisions, architecture choices, or project history from MAMA memory. This agent searches MAMA's decision database to find relevant context about previous decisions, patterns, and rationale.
mama-reviewer
Use when reviewing code changes against past decisions and contracts stored in MAMA memory. Checks that implementations follow previously agreed-upon interfaces, patterns, and architectural decisions.
dream-consolidator
Consolidates a project's MEMORY.md by removing genuinely stale entries, merging near-duplicates, and tightening wording while preserving content. Invoked by the dream skill.
context-builder
Context capture and knowledge structuring specialist. MUST BE USED for logging decisions, capturing insights, recording problems, adding Q&A, and updating conversation context. Use PROACTIVELY when important information should be remembered.
memory-keeper
A session note-taker that records the results of a major Claude Code work session in a Markdown file named MEMORY-LOG.md.
FAI Multi-Agent Service Tuner
Multi-Agent Service tuner — supervisor routing config, per-agent token budgets, loop limits, model selection per role, memory TTL, and orchestration cost analysis.