[caveman] Optimize instruction verbosity — developer.instructions, contribution-checker, agentic-chat (2026-06-15)#39457
Conversation
Why use many token when few do trick. - developer.instructions.md: tightened section preambles, condensed Data Flow steps, PR dedup protocol - contribution-checker.agent.md: trimmed Step 1/2/3 prose, condensed Comment Field section - agentic-chat.md: tightened Core Principles, Interaction Guidelines, and Do Not sections All YAML schema, field names, examples, and constraints preserved.
|
Please resolve the action_required workflow gate so this PR can move forward.
|
|
Hey One checklist flag:
If you'd like a hand addressing this:
|
|
@copilot please check whether snapshot or regression tests exist for the instruction files. If they do, update them to match the condensed wording; if not, report that no test changes are needed.
|
Summary
Reduces token weight in three agent instruction files by tightening prose — collapsing redundant section headers, trimming numbered-list preambles, and shortening field descriptions — without altering any functional rules, evaluation logic, YAML schema, field names, examples, or constraints.
Net diff: 69 insertions / 81 deletions across 3 files.
What Changed
.github/agents/contribution-checker.agent.md.github/agents/developer.instructions.md.github/aw/agentic-chat.mdMotivation
Agent instruction files are consumed as system prompt context on every invocation. Verbose phrasing in section intros, list preambles, and field descriptions adds token overhead without adding meaning. This change applies the principle of the commit message itself — why use many token when few do trick — to bring each file closer to minimum-viable instruction density.
Risk & Compatibility