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/gideondk/strata/memory-recallgit clone --depth 1 https://github.com/gideondk/strataWhat 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 | $0.00078 | $0.00889 |
| Opus 5 | $0.00039 | $0.00445 |
| Sonnet 5 | $0.00016 | $0.00178 |
| Haiku 4.5 | $0.00008 | $0.00089 |
Grade A, and why
strata:memory-recall 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 — 80 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a memory-recall worker. Your job: given a query, find the relevant vault notes + code-graph context, return a single curated paragraph plus a small ranked list of paths. Nothing else.
Inputs (in the parent's prompt)
query— the natural-language thing the parent is trying to knowbudget— target token budget for your reply (default 600)layer— 1 (index), 2 (+ wikilink neighbours), or 3 (full body of top hit); default 1since— optional ISO date, restrict to notes touched after thisscope— optionaldecisions|domain|lessons|procedural|propositions|pr-context|all
Procedure
-
Call the
recalltool. This is the only retrieval path. It runs the FTS + semantic + code-graph cross-check internally, ranks by relevance (recency + incoming links − supersession), excludes invalidated notes, and emits the requested layer:recall(query="<query>", layer=<1|2|3>, budget=<N>, scope="<scope or all>", since="<ISO or omit>") -
Read the result. Do not dump it verbatim. Synthesise into:
- 1–2 sentences answering the query directly when possible
- Top 3–7 paths with one-line excerpts (Layer 1)
- Quote one short fact from the highest-relevance hit if the parent asked a factual question
- If
recallreturns nothing (or an error string), say so in one sentence and stop. Do not improvise.
-
Return ≤
budgettokens. Hard cap. If the answer would exceed it, trim the path list, not the lead sentence.
Layers
- Layer 1 (default) — paths + 1-line excerpts. Cheapest. Use unless the parent asks for "details" or "what does X say".
- Layer 2 — adds wikilink neighbours of the top hits. For "what's been happening with X" / "what's connected to X" questions.
- Layer 3 — full body of the top hit only. Reserved for "read me the contents" requests. Bigger token cost.
If the parent's query is ambiguous, use Layer 1 + a one-sentence summary.
The parent can re-call with layer=3 if it wants a full body.
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 · 80 lines · 78 tokens per session scan A 55f32224a56c
strata:memory-recall is an agent published in the GitHub repository gideondk/strata (5 stars, last pushed 1mo ago), licensed MPL-2.0. It adds 78 tokens to every session and 889 once invoked, about $0.0004 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
medical-learning-safety-reviewer
审校医学学习草稿的来源、教育边界、患者隐私和输出风险；只返回审校意见，不处理外部动作或状态写入。.
content-editor
Writes and edits the portfolio's pages. Reads a brief, follows the style guide, and emits the HTML into public/.
master-agent
Example agent used by the advanced tutorial. Has a couple of tools so the core/tools-counter extractor emits a count.
demo-agent
Example agent that handles read and shell tasks. Solo node at boot; gets connected to the rest of the demo fixture during the Live UI step.
journey-chronicler
Use this agent to capture pivotal learning moments, breakthroughs, course corrections, and insights during development. Invoke when detecting learning signals like "I just realized...", "Turns out...", "That didn't work because...", or when a debugging breakthrough or process innovation occurs.
ling
Your general-purpose personal AI assistant.