From 432282f186018d0616e89ceb88bb3463275c0bb5 Mon Sep 17 00:00:00 2001 From: artale Date: Mon, 29 Jun 2026 12:11:13 +0200 Subject: [PATCH] docs(tac): add agent harness loop guidance Capture the Sean's AI Stories loop-engineering video in the TAC meta-prompt. Require plans to name harness context, memory, guardrails, traces, evals, and promotion criteria, and lock those markers with pytest. --- pipeline/test_loop_engineering_meta_prompt.py | 4 ++++ plans/meta-prompts/loop_engineering.md | 16 ++++++++++++++++ 2 files changed, 20 insertions(+) diff --git a/pipeline/test_loop_engineering_meta_prompt.py b/pipeline/test_loop_engineering_meta_prompt.py index a2cd995..907ac5a 100644 --- a/pipeline/test_loop_engineering_meta_prompt.py +++ b/pipeline/test_loop_engineering_meta_prompt.py @@ -12,6 +12,10 @@ def test_loop_engineering_meta_prompt_locks_safety_fields(): "git-proxy:8099/deploy", "auto_patch_proven", "rsi_canary_recovery_evidence", + "Agent harness", + "end-loop guardrails", + "latency, token use", + "YouTube `GrNbuWWJYiI`", "Model Workspace Protocol", "00-intake/", "40-ship/", diff --git a/plans/meta-prompts/loop_engineering.md b/plans/meta-prompts/loop_engineering.md index 3d52ae6..b58cc14 100644 --- a/plans/meta-prompts/loop_engineering.md +++ b/plans/meta-prompts/loop_engineering.md @@ -34,6 +34,22 @@ At minimum, plans must name the checks that prove: Prefer the boring loop that compounds over the clever prompt that works once. If the plan needs a new framework, first prove a markdown file, JSON receipt, and pytest assertion cannot hold the invariant. +## Agent harness note + +A loop is not just repeated model calls. Every agent plan must name the harness around the model: + +- context and working memory; +- durable semantic memory / RAG sources; +- episodic memory / traces from prior runs; +- tool allowlist and signed deploy boundaries; +- end-loop guardrails that define when to stop; +- tracing for retrievals, tool calls, latency, token use, and errors; +- evals/receipts that decide whether prompt or config changes can feed back into the next run. + +LLMOps/hill-climbing only promotes a changed prompt, retrieval config, tool config, or model parameter when traces plus evals show the new loop is safer or better. + +Reference: Sean's AI Stories, "You Can Learn AI Agent Harness & Loop Engineering In 19 Min" (YouTube `GrNbuWWJYiI`). + ## Model Workspace Protocol note For sequential workflows with human review between stages, prefer folder-structured orchestration before multi-agent framework code. The Model Workspace Protocol pattern treats numbered folders as stages, markdown files as role/context carriers, and local scripts as the boring mechanical layer.