@nexus-cortex/server 4.37.1 → 4.37.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.cortex/MEMORY.md +49 -0
- package/package.json +1 -1
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
# Memory — fresh install
|
|
2
|
+
|
|
3
|
+
This is a **fresh install of Nexus Cortex**, and you are likely the first agent
|
|
4
|
+
to run in it. This file is your persistent cross-session memory: it is injected
|
|
5
|
+
as a system message on every conversation turn, and YOU maintain it with your
|
|
6
|
+
file tools. Right now it contains onboarding notes; once you know the project,
|
|
7
|
+
**replace this content** with real memory (see "Memory discipline" below).
|
|
8
|
+
|
|
9
|
+
## First actions on a fresh install
|
|
10
|
+
|
|
11
|
+
1. **Run `/init`** (or the `InitCortexContext` tool). It scans the project and
|
|
12
|
+
generates `.cortex/CORTEX.md` — the project-context document auto-loaded
|
|
13
|
+
every turn — and establishes this memory file. `/init --global` does the
|
|
14
|
+
same for `~/.cortex/`, shared across projects.
|
|
15
|
+
2. **Discover your skills**: call `Skill` with `command: "list"`. The install
|
|
16
|
+
ships a skill library out of the box — highlights:
|
|
17
|
+
- `autoresearch` — act as the auto-research PM (plan-gated experiments,
|
|
18
|
+
varied subagent arms, holdout-verified winners). Pairs with the
|
|
19
|
+
`personas/` arm library inside the skill.
|
|
20
|
+
- `cortex-bench` — the multi-model benchmark methodology.
|
|
21
|
+
- `best-of-n` — implement one task N ways in parallel, keep the best.
|
|
22
|
+
- `verify-work` — adversarial verification before calling work done.
|
|
23
|
+
- `docx` / `xlsx` / `pptx` / `pdf-documents` — produce and edit real
|
|
24
|
+
document deliverables, with verification steps.
|
|
25
|
+
3. **Discover your agents**: call `Task` with `subagent_type: "list"` to see
|
|
26
|
+
the shipped agent profiles (code-reviewer, explore, plan, test-writer,
|
|
27
|
+
autoresearch-agent, …). Profiles live in `.cortex/agents/*.md`.
|
|
28
|
+
4. **Check the configuration**: `.env` at the repo root holds API keys and
|
|
29
|
+
`DEFAULT_MODEL_ID` (see `.env.example` for every option). The CLI offers
|
|
30
|
+
`--list-models` / `--list-tools`; the server exposes `/models` and `/tools`.
|
|
31
|
+
|
|
32
|
+
## Where things live
|
|
33
|
+
|
|
34
|
+
- `.cortex/` — sessions (JSONL), skills, agents, commands, permissions,
|
|
35
|
+
system-messages, and this memory. Runtime stores (router-matrix,
|
|
36
|
+
research-backlog, decisions) are created on demand.
|
|
37
|
+
- `.cortex/CORTEX.md` — project context, generated by `/init`, auto-loaded.
|
|
38
|
+
- Auto-research is gated by `AUTORESEARCH_AGENTS` (`off`|`native`|`mcp`) in `.env`.
|
|
39
|
+
|
|
40
|
+
## Memory discipline
|
|
41
|
+
|
|
42
|
+
Keep this file small, current, and factual — it costs context every turn.
|
|
43
|
+
After `/init`, replace the onboarding content above with:
|
|
44
|
+
|
|
45
|
+
## Project Patterns
|
|
46
|
+
|
|
47
|
+
## Key Files
|
|
48
|
+
|
|
49
|
+
## Known Issues
|