From a4852a574abb2e3377f148ba93354768a651c5cc Mon Sep 17 00:00:00 2001 From: Jacob Hagstedt Date: Wed, 3 Jun 2026 22:45:41 +0200 Subject: [PATCH] docs: fix OpenCode subagent terminology in AGENTS.md Replace "sub-agent" (hyphenated) with "subagent" and update "spawn a sub-agent" to "invoke a subagent via the Task tool" to match OpenCode's actual documented terminology. Co-Authored-By: Claude Sonnet 4.6 --- AGENTS.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/AGENTS.md b/AGENTS.md index 5be8a01..68e1019 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -126,10 +126,10 @@ tree -I 'node_modules|.git|dist|build|__pycache__|venv|.venv|env|.pytest_cache|. **Context:** The main OpenCode agent context window is our most critical resource. Executing complex tasks, debugging cross-file issues, or writing massive features directly in the main thread causes severe context bloat, leading to code hallucinations and dropped instructions. -* **Delegate Aggressively:** For any complex implementation, structural change, or task larger than a single file edit, you **MUST spawn a sub-agent**. Act strictly as the architectural orchestrator; let sub-agents handle the iterative coding labor. -* **Scope Narrowly:** Give sub-agents highly specific, isolated, and atomic objectives (e.g., *"Read `backend/app/routers/auth.py` and summarize endpoints"* or *"Write a unit test for the frontend sleep tracker component"*). -* **Strict Summarization:** Sub-agents are **STRICTLY FORBIDDEN** from passing raw terminal dumps, unedited file blocks, or their internal step-by-step logs back to the main thread. -* **Return Format:** Sub-agents must return *only* clear, actionable results back to the orchestrator: a concise summary of findings, the exact lines of code that require modification, or direct confirmation of a successful execution. +* **Delegate Aggressively:** For any complex implementation, structural change, or task larger than a single file edit, you **MUST invoke a subagent via the Task tool**. Act strictly as the architectural orchestrator; let subagents handle the iterative coding labor. +* **Scope Narrowly:** Give subagents highly specific, isolated, and atomic objectives (e.g., *"Read `backend/app/routers/auth.py` and summarize endpoints"* or *"Write a unit test for the frontend sleep tracker component"*). +* **Strict Summarization:** Subagents are **STRICTLY FORBIDDEN** from passing raw terminal dumps, unedited file blocks, or their internal step-by-step logs back to the main thread. +* **Return Format:** Subagents must return *only* clear, actionable results back to the orchestrator: a concise summary of findings, the exact lines of code that require modification, or direct confirmation of a successful execution. ## NOTES