page:docs-harness-features-backlog-gaps-prompt-engineering-gap-prompt-011
GAP-PROMPT-011: Output Efficiency Rules reference
Last refreshed: 2026-04-25
GAP-PROMPT-011: Output Efficiency Rules
| Field | Value |
|---|---|
| Category | prompt-engineering |
| Priority | Medium |
| Effort | S |
| Status | Missing |
Last refreshed: 2026-04-25
Description
Include output efficiency guidelines in orchestration prompts to reduce token consumption from verbose delegated agent output. Long-running orchestrations accumulate context quickly; concise output from each iteration preserves token budget for actual work.
Current State
Iteration prompts generated by session:iteration-message do not include output efficiency guidelines. Delegated agents often produce verbose explanations, repeat user instructions, and include unnecessary preamble. This wastes token budget, especially in multi-iteration orchestrations where context accumulates.
Target State
An outputEfficiency prompt section injected into iteration prompts and task prompts. Content: lead with action not reasoning, skip filler and preamble, do not restate instructions, focus on decisions needing input / status updates / errors, one sentence over three when possible. Adapted from CC's output efficiency rules. Configurable verbosity level per process definition (verbose for debugging, concise for production).
Dependencies
- GAP-PROMPT-001 -- strata model for section placement
- GAP-PERF-002 -- session compaction for token management
Key Files
| Component | Path |
|---|---|
| Prompts module | packages/sdk/src/prompts/ |
| Session management | packages/sdk/src/session/ |
| CC prompt phrasing analysis | `11-prompt-phrasing-analysis.md` -- Section 5 (Output Efficiency Rules) has exact CC phrasing |
Recommendation
Phase 2 implementation. Small effort -- extract efficiency rules, inject into iteration prompt template. Measurable impact on token usage in multi-iteration runs.