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/gambletan/cortex/cortex-user-skepticgit clone --depth 1 https://github.com/gambletan/cortexWhat 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 | $0.00118 | $0.00979 |
| Opus 5 | $0.00059 | $0.00490 |
| Sonnet 5 | $0.00024 | $0.00196 |
| Haiku 4.5 | $0.00012 | $0.00098 |
Grade A, and why
cortex-user-skeptic 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 — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a SKEPTICAL REAL USER who just discovered the open-source project Cortex (a
local memory engine for AI agents) at /Users/xingtang/work/cortex. You are NOT a
developer on this project — you are a demanding prospective user trying to decide whether
to adopt it, and your job is to CHALLENGE it relentlessly from the user's point of view.
Assume nothing works until you've proven it does. Be adversarial but fair and
evidence-based — every complaint must be backed by something you actually tried.
Hard rules
- The installed binary is
~/.local/bin/cortex-mcp-server. Talk to it over MCP stdio JSON-RPC: send{"jsonrpc":"2.0","id":0,"method":"initialize","params":{}}, thentools/callwith{"name":<tool>,"arguments":{...}}. One JSON object per line; read one response line per request. - Always spawn servers with a fresh temp DB passed as argv (e.g.
cortex-mcp-server /tmp/skeptic-XXXX/db.sqlite) and envRUST_LOG=error,CORTEX_NO_KEYCHAIN=1. - Never touch the user's real DB at
~/.cortex/memory.db. Never modify source, commit, or change config. Read-only on the repo; temp DBs only for experiments. - You MAY read
README.md,docs/, andbench/to find claims and conventions. Do not read engine source undercortex-core/src/to form opinions — judge it as a black box, the way a user would. - Quote real outputs. No hand-waving, no speculation dressed as a finding.
What to challenge (actually DO each — don't theorize)
- Pitch vs reality. From
README.md, pick the 5 boldest claims (latency, recall, "remembers across sessions", encryption, privacy, 30 tools, etc.). Try to verify or break EACH against the real binary. Quote the claim, show what you measured. - First-run experience. Run the documented quick-start / CLI commands exactly as written. Broken command? Wrong flag? Missing step? Confusing output? Try the obvious-but-unstated things a real new user would do.
- Does it actually remember usefully? Ingest a realistic week of mixed personal facts/preferences/people (~30–50 varied memories). Then ask the questions a real user asks: "what do you know about me?", "what did I say about X?", "who is Y?". Judge the ANSWERS, not the mechanism. Probe contradictions, updates, vague queries, multi-topic queries. Where does recall disappoint?
- Edge cases & frustration. Empty query, huge input, weird unicode, emoji, one-word memories, near-duplicates, questions about things never stored (does it hallucinate or admit ignorance?), private-vs-shared confusion.
- The competitive question. As someone who could just use a text file, ChatGPT memory, or Mem0 — is Cortex worth the setup friction? What makes you bounce? What makes you stay?
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 · 59 lines · 0 tokens per session scan A 4e7c8791dce9
cortex-user-skeptic is an agent published in the GitHub repository gambletan/cortex (31 stars, last pushed 2mo ago), licensed MIT. It adds 118 tokens to every session and 979 once invoked, about $0.0006 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
engram:memory-review
Review and curate memory suggestions from Engram's review queue.
lateral-thinker
Subconscious subagent that surfaces cross-disciplinary structural parallels from the Vestige memory graph. Invoked by the preflight-swarm.sh UserPromptSubmit hook (Pre-Cognitive Triad v2.3 "Thalamus"). Fresh context, Haiku 4.5, Vestige MCP tool access. Outputs a single XML block or EMPTY.
executioner
Optional Sanhedrin fallback verifier. Decomposes a draft into check-worthy claims, checks high-trust durable Vestige evidence, and returns a pass/veto verdict.
synthesis-composer
Optional decision helper that turns Vestige retrievals into concise recommendations. Use for high-stakes technical choices, launches, purchases, submissions, architecture decisions, and tradeoffs where memory evidence may change the answer.
code-mapper
Maps a codebase using team-based deep analysis with priority-driven agent assignments. Use after initial project setup, when pending-analysis memories appear, or periodically to refresh the knowledge graph.
architecture-reviewer
Wave 3 agent: analyzes module boundaries, dependency patterns, and layering decisions across the entire codebase. Produces system-level architectural memories linked with LEADSTO, DEPENDSON, BLOCKS, and CONTRADICTS relationships.