@@ -7,7 +7,46 @@
77
88 ---
99
10-## LSP Diagnostics
10+# Context window sizes are out of date
11+
12+CONTEXT_WINDOW_TABLE in providers.ts is out of date. Codex 5.3 and 5.2 have 400k tokens context limit not 128.
13+This list looks **VERY** wrong, let's get the facts from model.dev and update our list.
14+
15+`curl https://models.dev/api.json`
16+
17+Let's use the curl command above to update our table.
18+
19+---
20+
21+## Subagents are not really used that much.
22+
23+The main agent should be making use of these to ensure a clean and enduring context window. Search results, codebase reading, etc bloat the context.
24+Is short the main agent should delegate to subagents as much as he can. Some tools like the exa powered web tools, also generate big responses, which makes sense
25+since they are fetching content, but we should analyze them to see if we can trim the data.
26+
27+- Check the database for suabgent tool usage statistics
28+- Check the current language and instructions around the subagent tool
29+- Audit read/web tools responses for context bloat (anything that doesn't help the agent's decision making)
30+
31+Note: I've done some changes to the system-prompt.ts to try and ecourage more subagent usage.
32+
33+---
34+
35+# Post v1.0.0 work:
36+
37+## Custom commands vs built-in commands vs subagents
38+
39+We currently have a ton of overlap with these features, Commands run within subagents, some commands come from custom configs others are baked in.
40+
41+Let's clean things up:
42+
43+- `/review` command should use the same code as a custom-commands, consolidate the two. We can create the global `/review` command in `~/.agents/commands` at app start if it doesn't exist. That way it can be a pure custom command and the users are encouraged to edit it for their own custom reviews. Never overwrite the file if it exists. Print a line notifying the user that the command was created.
44+
45+- Custom commands and custom subagents: Custom commands just spawn generic subagents with a dedicated prompt, custom agents are just subagents with a dedicated system prompt and a main agent provided prompt. I believe these are already well implemented, and share most of the code.
46+
47+---
48+
49+## LSP Diagnostics (not very important, with tool-hooks and strong linting, this is not as necessary, but we should implement asap)
1150
1251 We should have a closed loop feedback for LSP diagnostics on edits/reads.
1352