From 87d741cbafa66c020a67efe7a04b89408566cefe Mon Sep 17 00:00:00 2001 From: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com> Date: Fri, 28 Aug 2026 19:30:22 -0700 Subject: [PATCH] chore(context): drop redundant commentary from graph-analyst and session-navigator MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit T1: drop all blocks from agents/graph-analyst.md (2 examples, commentary only) and agents/session-navigator.md (1 example, commentary only -- already well under the 2-example cap, so nothing is dropped there beyond commentary). Both files carry two separate top-level description fields in frontmatter -- one under a bundle: mapping (a short, one-line, no-examples summary used for bundle listings) and one under the meta: mapping (the real, description: | block-scalar one with the / content this item targets). Edited only the meta: description via git apply; verified the bundle: description was untouched and the meta: description now has commentary=0 (graph-analyst.md: examples=2, session-navigator.md: examples=1) via yaml.safe_load post-patch. Full module test suite green: modules/tool-context-intelligence-query/tests/ 195 passed, 0 failed. One test, test_tool_dependencies.py, references graph-analyst.md by name but asserts a module-entry-point-naming invariant unrelated to the / content this item touches -- confirmed it still passes. Validated by the 1e2 context-hygiene DTU A/B treatment: 12/12 runs task-correct on both providers at n=3/arm; wire audit confirmed every deleted string absent in all treatment runs and present in all controls; measured -5,860 chars/turn system prompt (-11.3%) and -1,277 chars/turn tool catalog (-4.5%) reduction across the full treatment set; no YAML corruption in mounted agent catalogs. (openai arm showed -45% cost but is noise-inflated at this sample size; the defensible claim is the wire-byte reduction.) 🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier) Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com> --- agents/graph-analyst.md | 2 -- agents/session-navigator.md | 1 - 2 files changed, 3 deletions(-) diff --git a/agents/graph-analyst.md b/agents/graph-analyst.md index fa77435..61d6a62 100644 --- a/agents/graph-analyst.md +++ b/agents/graph-analyst.md @@ -23,14 +23,12 @@ meta: Context: User wants to query session events using the graph user: 'Find all tool errors in my last session using the graph' assistant: 'I will use graph-analyst to run a Cypher query for tool error events — it checks server availability first and falls back to session-navigator if the server is unreachable.' - Graph-powered session event queries go to this agent. It handles server availability automatically. Context: User needs to trace a delegation tree user: 'Show me the full delegation tree for my last recipe run' assistant: 'I will delegate to graph-analyst to trace the parent-child session chain and map the delegation tree using Cypher graph traversal.' - delegation tree tracing across many sessions benefits from graph traversal rather than scanning JSONL files. model_role: [reasoning, general] diff --git a/agents/session-navigator.md b/agents/session-navigator.md index b020718..102d204 100644 --- a/agents/session-navigator.md +++ b/agents/session-navigator.md @@ -18,7 +18,6 @@ meta: Context: Graph analyst delegating because server is unreachable user: [graph-analyst delegates] 'Find tool errors in session abc123 — graph server is unreachable. Workspace: my-project' assistant: 'I will scope search to workspace my-project. I will first resolve CONTEXT_INTELLIGENCE_ROOT="${AMPLIFIER_CONTEXT_INTELLIGENCE_BASE_PATH:-$HOME/.amplifier/projects}", then look in "$CONTEXT_INTELLIGENCE_ROOT"/my-project/sessions/ first, then filter by workspace field if needed. I will search for tool errors using safe jq extraction patterns.' - session-navigator receives workspace from graph-analyst and uses it to scope all directory lookups and field filters. External callers should never invoke session-navigator directly. model_role: general