From da60991b15cd62c724ceb54cddb79dde993e2cd4 Mon Sep 17 00:00:00 2001 From: Sven Date: Wed, 25 Mar 2026 19:32:21 +0100 Subject: [PATCH] docs(maxsim-batch): add tier selection table and graceful degradation to SKILL.md MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Adds the §7.2-required Tier Selection Logic decision table showing which workflows use Tier 1 subagents vs Tier 2 Agent Teams, and a Graceful Degradation section with the exact fallback message when Agent Teams are unavailable. Brings the §7.2 Parallelism Strategy audit to 100%. Co-Authored-By: Claude Sonnet 4.6 --- templates/skills/maxsim-batch/SKILL.md | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/templates/skills/maxsim-batch/SKILL.md b/templates/skills/maxsim-batch/SKILL.md index 219a0af3..75802d3b 100644 --- a/templates/skills/maxsim-batch/SKILL.md +++ b/templates/skills/maxsim-batch/SKILL.md @@ -89,12 +89,34 @@ Agent Teams (available since Claude Code v2.1.32, Feb 2026) enable inter-agent c **Current status:** Infrastructure is in place (env var, hooks). Workflow templates that invoke `TeamCreate`/`SendMessage` for Tier 2 patterns (competitive implementation, multi-reviewer code review, collaborative debugging) are planned but not yet implemented. All workflows currently use Tier 1 subagents. See PROJECT.md §7.2 for the full specification. +### Tier Selection Logic + +MaxsimCLI chooses the tier automatically based on the workflow: + +| Workflow | Tier | Reason | +|----------|------|--------| +| Phase execution (independent tasks) | Tier 1 (Subagents) | Tasks don't need to communicate | +| Codebase scanning | Tier 1 (Subagents) | Read-only, report back | +| Research gathering | Tier 1 (Subagents) | Collect and report | +| Competitive implementation | Tier 2 (Agent Teams) | Agents need to debate | +| Multi-dimensional code review | Tier 2 (Agent Teams) | Findings need cross-checking | +| Collaborative debugging | Tier 2 (Agent Teams) | Hypotheses need adversarial testing | +| Architecture exploration | Tier 2 (Agent Teams) | Requires discussion | + **When Tier 2 is ready, it will be used for:** - Competitive implementation with adversarial debate - Multi-dimensional code review (security + performance + test coverage) - Collaborative debugging with competing hypotheses - Cross-layer feature work (frontend + backend + tests) +### Graceful Degradation + +If Agent Teams are unavailable (env var `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS` not set, unsupported plan, or feature not yet stable), MaxsimCLI falls back to Tier 1 subagents for all workflows. Inform the user with this exact message: + +> "Competitive mode: using Tier 1 subagents (Agent Teams not available or not required for this strategy). Each executor works independently; verifier selects the best result." + +The user is informed but not blocked. All workflows remain fully functional via Tier 1. + ## Limits - Up to 30 parallel agents; typically 3-10 for manageable coordination