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 skills/vedantnimbarte/kortex-memory/kortex-memorynpx skills add vedantnimbarte/kortex-memory --skill kortex-memorygit clone --depth 1 https://github.com/vedantnimbarte/kortex-memoryWhat 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 | $0.00095 | $0.01096 |
| Opus 5 | $0.00048 | $0.00548 |
| Sonnet 5 | $0.00019 | $0.00219 |
| Haiku 4.5 | $0.00010 | $0.00110 |
Grade A, and why
kortex-memory 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 — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Writing and reading Kortex memory
The MCP tools give you a memory. This tells you when to use it, which is the part that decides whether the memory is useful in six months or landfill.
Read before you decide, not after
Before proposing an approach in an unfamiliar area, recall it. A memory layer
pays for itself the first time it stops you re-litigating a decision the team
already made — and costs the user nothing except when you skip it and rebuild
something that was deliberately removed.
Worth a recall first:
- a design or architecture choice in code you have not touched this session
- anything the user frames as "why is this like this"
- a convention question ("do we use X or Y here")
- before a large refactor, to find the constraint that explains the odd bit
Not worth it: reading a file, running a test, answering from what is already in this conversation.
Write the durable half, not the transcript
The failure mode is not writing too little. It is writing everything — a corpus of "fixed a typo" that buries the three facts that mattered.
Write when:
| Signal | Example |
|---|---|
| A decision is settled, with a reason | "Postgres over DynamoDB for the ledger: we need joins" |
| A convention is agreed | "Migrations are named kkxNNNN, not by date" |
| A constraint is discovered the hard way | "plainto_tsquery ANDs its terms — a two-word query needs both" |
| A fix was non-obvious | "The MissingGreenlet came from onupdate=func.now() expiring the attribute" |
| The user says to | "remember that…" |
Do not write:
- anything already in the repository — code, README, CLAUDE.md, git history. A memory that restates a file is a second copy to keep in step, and it will drift.
- the fact that you ran a command, read a file, or fixed a lint error
- a summary of this conversation. Sessions are cheap. Conclusions are not.
- anything you are guessing about. Confidence you do not have is worse than silence, because it reads as settled later.
One memory, one fact. "We use Postgres and migrations are kkxNNNN and the worker runs Celery" is three memories in a trench coat: it matches every query weakly and answers none of them well.
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 · 98 lines · 95 tokens per session scan A 5ba35a707b4a
kortex-memory is a skill published in the GitHub repository vedantnimbarte/kortex-memory (2 stars, last pushed 3d ago), licensed Apache-2.0. It adds 95 tokens to every session and 1,096 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 skills, from other repositories
openai-docs
Use when the user asks how to build with OpenAI products or APIs and needs up-to-date official documentation with citations (for example: Codex, Responses API, Chat Completions, Apps SDK, Agents SDK, Realtime, model capabilities or limits); prioritize OpenAI docs MCP tools and restrict any fallback browsing to…
trellis-start
Initializes an AI development session by reading workflow guides, developer identity, git status, active tasks, and project guidelines from .trellis/. Classifies incoming tasks and routes to brainstorm, direct edit, or task workflow. Use when beginning a new coding session, resuming work, starting a new task, or…
company-brain
How an agent should operate as one mind inside a shared Company Brain built on Caura — recall before acting, obey fleet keystones, reuse and publish skills, and report outcomes so every task compounds across the whole organization. Use this whenever you work as part of a Caura-connected team or fleet and your work…
seo-content-writer
Write SEO blog posts, articles, landing pages with keyword integration, header optimization, and snippet targeting. SEO文章写作/内容优化.
Universal Tracker
Cross-platform background monitor to auto-detect new videos/posts (YouTube, Douyin, Website) and run team workflows.
schema-markup-generator
Generate JSON-LD structured data for FAQ, HowTo, Article, Product, LocalBusiness rich results. Schema标记/结构化数据.