feat(rules): add subagents.md — when, how, and when not to delegate

Cross-cutting rule covering subagent hygiene:

- Decision table: parallel-safe (read-only investigation, independent
  test failures, multi-domain research) vs sequential-with-review
  (plan execution, coupled edits) vs never-parallel (concurrent writes)
  vs don't-subagent-at-all (target known, work fits in ~10 tool calls).
- Prompt contract: scope / pasted context / explicit 'do NOT' list /
  required output format. Missing any field yields shallow work.
- Context-pollution rule: subagents absorb noise the main thread
  shouldn't carry. When one fails, dispatch a fix-agent with the
  failure report — do not retry in the orchestrator.
- Review-gate cadence: after each sequential task, or every ~3 in
  parallel batches.
- Anti-patterns including parallel implementation on overlapping
  files, broad 'fix all the tests' prompts, timeout-tuning to mask
  flakes, and letting the agent decide scope.

Added one-line cross-references to subagents.md from debug/,
review-code/, and finish-branch/ SKILL.md — the skills that most
rely on delegation.

Clean-room synthesis from NeoLabHQ/context-engineering-kit's
subagent-driven-development pattern (MIT).

1 files changed, 2 insertions, 0 deletions

diff --git a/review-code/SKILL.md b/review-code/SKILL.md
index 6366345..fdf5cfb 100644
--- a/review-code/SKILL.md
+++ b/review-code/SKILL.md
@@ -62,6 +62,8 @@ If skipping, report: "Skipped — <reason>" and stop.
 
 Each perspective is a distinct review angle. For substantial changes, dispatch them as parallel sub-agents. For small changes, run sequentially inline.
 
+Follow `subagents.md` for the dispatch contract — each perspective's prompt needs explicit scope, pasted diff context, constraints (don't flag unmodified lines, don't rewrite the PR), and a required output format. Perspectives are read-only and independent, so parallel fan-out is always safe here.
+
 ### Perspective A — CLAUDE.md Adherence
 
 Read the CLAUDE.md files collected in Phase 1. For each rule they state, check whether the diff honors it. CLAUDE.md is writing guidance, so not every rule applies at review time — focus on rules that are *assertions about what committed code should look like*.