InsanusMokrassar/tgbotapi
1
0
Fork 0
mirror of https://github.com/InsanusMokrassar/TelegramBotAPI.git synced 2026-07-29 21:26:27 +00:00
Code Releases Activity

Document the plan/code/check agent pipeline in HELPERS

Browse Source 
Add a rule describing the three-phase subagent pipeline (Plan/Architect ->
Coding -> Check) and the best/high/medium model tiers (Fable/Opus/Sonnet).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
InsanusMokrassar
2026-07-20 11:23:13 +06:00
parent a2f670834e
commit bc52be7b40
1 changed files with 10 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
agents/HELPERS.md
View File

@@ -60,3 +60,13 @@ All work MUST be performed in the normal (main) working worktree of the reposito
---
When a prompt contains a list of several subtasks, run each subtask as its own subagent (one subagent per subtask) instead of doing all of them inline in the main thread.
---
For any substantial change, drive it through a three-phase agent pipeline, each phase run as its own subagent(s), and each section/subtask committed separately:
1. **Plan / Architect** the changes — model range best → medium, best is the most suitable.
2. **Coding / implementation** — model range medium → high, medium is the most suitable.
3. **Check / validation** — model range high → best, high is the most suitable.
Model tiers: **best = Fable**, **high = Opus**, **medium = Sonnet**. (Tier names, not the concrete models, are the contract — remap the models if the roster changes.)