From c8102183832f9b56f057bb2fb4e17bb43490346e Mon Sep 17 00:00:00 2001 From: metaphorics <152830360+metaphorics@users.noreply.github.com> Date: Mon, 15 Jun 2026 20:39:36 +0900 Subject: [PATCH] docs(coding-agent): refresh workflowz notice with advisor + provenance patterns Add two patterns to the hidden workflowz orchestration notice: "Advisor consult" (consult the advisor tool at a hard fork before committing, phrased "if enabled" so it is independent of the advisor-tool change) and "Provenance-rich context" (brief each subagent on which files/commits and what "done" means via a '/home/alpha/.omp/agent/sessions/-odin-oh-my-pi/2026-06-15T05-57-55-286Z_019ec9db-a3d6-7000-bad4-21ce1689b5f3/local' brief). Content-only; no code change. --- packages/coding-agent/CHANGELOG.md | 4 ++++ packages/coding-agent/src/prompts/system/workflow-notice.md | 2 ++ 2 files changed, 6 insertions(+) diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md index e1682b7e6f..9a85f5d5ac 100644 --- a/packages/coding-agent/CHANGELOG.md +++ b/packages/coding-agent/CHANGELOG.md @@ -2,6 +2,10 @@ ## [Unreleased] +### Changed + +- Refreshed the hidden `workflowz` orchestration notice with two patterns: an "Advisor consult" (consult the `advisor` tool at a hard fork before committing) and "Provenance-rich context" (brief each subagent on which files/commits and what "done" means via a `local://` brief). + ## [15.13.2] - 2026-06-15 ### Added diff --git a/packages/coding-agent/src/prompts/system/workflow-notice.md b/packages/coding-agent/src/prompts/system/workflow-notice.md index a8e1c6f557..ce1b648e40 100644 --- a/packages/coding-agent/src/prompts/system/workflow-notice.md +++ b/packages/coding-agent/src/prompts/system/workflow-notice.md @@ -57,6 +57,8 @@ Compose the harness the task calls for: - **Completeness critic** — a final agent that asks "what's missing — modality not run, claim unverified, file unread?"; its answer is the next round. - **Budget/count loops** — `while len(bugs) < 10:` to hit a target, or `while budget.total and budget.remaining() > 50_000:` to scale depth to the turn budget; `log()` each round. - **No silent caps** — if you bound coverage (top-N, no-retry, sampling), `log()` what you dropped; silent truncation reads as "covered everything" when it didn't. +- **Advisor consult** — at a hard fork or when stuck, if the `advisor` tool is enabled, consult it for a second opinion from a more-capable paired model before committing to a direction; cheaper than a full judge panel for a single decision. +- **Provenance-rich context** — hand each subagent which files/commits to touch, why, and what "done" looks like via a `local://` brief; never assume a subagent shares your context. Scale to the ask: "find any bugs" → a few finders, single-vote verify. "thoroughly audit / be comprehensive" → larger finder pool, 3–5-vote adversarial pass, a synthesis stage.