mex is a repository-local knowledge system for AI coding agents that turns codebase discoveries into connected Markdown documentation and a code graph. It helps agents and developers retain architectural decisions, conventions, edge cases, and project context across coding sessions while detecting documentation affected by code changes. The catalogue skills and instructions are add-ons for using mex with coding agents.
Borrowing it
Nothing to install: this file belongs to mex-memory/mex. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/mex-memory/mex/main/.agents/skills/mex-inbox/SKILL.mdgit clone --depth 1 https://github.com/mex-memory/mexWrote 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/skills/mex-memory/mex/mex-inbox)<a href="https://agentmods.dev/skills/mex-memory/mex/mex-inbox"><img src="https://agentmods.dev/badge/skills/mex-memory/mex/mex-inbox/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/skills/mex-memory/mex/mex-inbox"><img src="https://agentmods.dev/badge/skills/mex-memory/mex/mex-inbox.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 25 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00081 | $0.00819 |
| Opus 5.5 | $0.00032 | $0.00328 |
| Sonnet 5.5 | $0.00016 | $0.00164 |
| Haiku 4.5 | $0.00008 | $0.00082 |
Grade A, and why
mex-inbox 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 1mo 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 — 45 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MEX Inbox
Turn an explicit request to retain project knowledge into a focused contribution for review. Inbox proposals are review artifacts; accepted knowledge belongs in the existing Wiki Markdown. Ordinary GROW upkeep can continue directly.
Keep the scope honest
- Create one
knowledge.createorknowledge.updatechange per draft, for an existing kind:architecture,component,convention,decision,pattern, orguide. - Existing Spec workflows also support
spec.createandspec.updateforspec,requirement,constraint, andacceptance_criterion. Choose these only for actual Spec-family intent. - Capture durable conclusions, their rationale, and useful evidence. Do not dump the conversation or present unresolved ideas as agreed facts.
- Do not route session logs, Relays, or routine GROW edits through Inbox merely because they contain context.
Prepare a draft
- Distill what future agents or teammates need to know from the user's request and discussion.
- Search existing knowledge first. Prefer correcting or extending the relevant record or section when it already covers the subject; create a new entry when the claim has no suitable home.
- Read the exact target and current revisions before drafting a correction. Preserve unrelated knowledge when replacing its body. Never guess an ID or revision.
- Preserve useful files, code, commits, entities, and external links as evidence only when actually available. Never invent provenance, authors, dates, events, or history.
- Build one request against the action-scoped runtime contract.
- Run the exact preview flow and summarize the human effect before technical details.
- When the user already asked to create, save, or draft it, apply that exact successful checkout-local draft preview without asking again.
- Return
/inbox?view=drafts&draft=<id>and state that the draft is checkout-local, nothing was published or shared, and review or publication happens in Hub.
Read references/cli-workflows.md before executing any Inbox mutation. Load only the operation being performed.
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 1mo ago Changed · -35 tokens per session 39cf595ac482
- 1mo ago First seen · 45 lines · 116 tokens per session scan A ec0e61b45c9f
mex-inbox is a skill published in the GitHub repository mex-memory/mex (1,779 stars, last pushed yesterday), licensed MIT. It adds 81 tokens to every session and 819 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-09-03.
Other skills, from other repositories
layered-context
L0/L1/L2 three-layer context loading protocol — reduces token consumption during /resume boot.
semantic-search
Zero-dependency TF-IDF search across MUSE memory, roles, and skills. Use when user wants to find information across their project context.
contextstream-coach
Coach this agent on work the person needs again and again, such as a weekly note, an article, a recap or an image series. Use it when the person starts a job for one of their workflows, reacts to a draft, or asks what was learned from earlier jobs. Goals, drafts and the person's own words are saved in their…
strategic-compact
Suggests manual context compaction at logical intervals to preserve context through task phases rather than arbitrary auto-compaction. Supports focus-aware compaction for targeted context preservation.
engram
Local-first personal AI identity and memory layer for MCP-compatible coding tools (Claude Code, Codex, Cursor, and others). Use this skill when the user wants to continue from a previous session ("continue from last session", "pick up where we left off"), recall a past decision ("what did we decide", "what was our…
context-receipts
Emit privacy-safe receipts for context selection, deferral, hydration, compaction, pruning, delegation, usage attribution, and boundary handoffs.