AGENTS.md

1---
2name: sisyphus-junior-agent
3description: Developer reference for the Sisyphus-Junior category-spawned executor agent -- model variants and discipline.
4---
5 
6# src/agents/sisyphus-junior/ -- Category-Spawned Executor
7 
8**Generated:** 2026-05-18
9 
10## OVERVIEW
11 
1210 files. Sisyphus-Junior is a focused task executor spawned by `delegate-task` when category routing requires it. Runs in subagent mode with its own fallback chain. Does not delegate further; executes directly.
13 
14## FILES
15 
16| File | Purpose |
17|------|---------|
18| `agent.ts` | `createSisyphusJuniorAgentWithOverrides()` factory, model-variant routing, `SISYPHUS_JUNIOR_DEFAULTS` |
19| `index.ts` | Barrel exports |
20| `default.ts` | Base/Claude prompt: todo discipline, verification, termination rules |
21| `gemini.ts` | Gemini-optimized prompt variant |
22| `gpt.ts` | Base GPT prompt variant |
23| `gpt-5-4.ts` | GPT-5.4-native prompt variant |
24| `gpt-5-5.ts` | Shared GPT-5.5/GPT-5.6 model-aware prompt variant |
25| `kimi-k2-6.ts` | Kimi K2.6 prompt variant |
26| `kimi-k2-7.ts` | Kimi K2.7-native prompt variant |
27| `kimi-k3.ts` | Kimi K3-native prompt variant (reasoning depth with built-in stop conditions) |
28| `index.test.ts` | Unit tests |
29 
30## VARIANT SELECTION
31 
32Parent `agent.ts` selects prompt variant by model name:
33- Kimi K3 -> `kimi-k3.ts`
34- Kimi K2.7 -> `kimi-k2-7.ts`
35- Contains "kimi-k2" -> `kimi-k2-6.ts`
36- Contains "gpt-5.5" or "gpt-5.6" -> shared `gpt-5-5.ts` prompt family with model-aware identity
37- Contains "gpt-5.4" -> `gpt-5-4.ts`
38- Contains "gpt" -> `gpt.ts`
39- Contains "gemini" -> `gemini.ts`
40- Default -> `default.ts` (Claude, GLM, etc.)
41 
42## KEY BEHAVIORS
43 
44- Mode: `subagent` (uses own fallback chain, ignores UI selection)
45- Default model: `claude-sonnet-5`
46- Default temperature: `0.1` (`SISYPHUS_JUNIOR_DEFAULTS`)
47- Fallback chain: claude-sonnet-5 -> kimi-k3 -> gpt-5.6-sol medium -> minimax-m3 -> MiniMax-M3 -> minimax-m2.7 -> big-pickle
48- Blocked tools: `task` (all models); `apply_patch` also blocked for GPT models
49- `call_omo_agent` explicitly allowed so subagents can spawn explore/librarian
50- Max tokens: 64000
51- Thinking enabled for non-GPT/non-GLM models (budgetTokens: 32000)
52- Reasoning effort "medium" for GPT models
53