TL;DR Readout Discipline
Every agent reply that contains substantive work ends with a TL;DR summary in the format below. The user reads the body if they want; the TL;DR makes the message scannable in <15 seconds.
Core claim
Long agent replies hide the substance in text. A short bulleted TL;DR at the bottom, with bold lead-ins, lets the user read the message in seconds. This is a summary, not a decision brief - capture what matters, not field-labeled slots.
Format (fixed - use this exactly)
**TL;DR**
- **Bold lead-in** - short descriptive content explaining the point.
- **Another bold lead-in** - more content; keep each bullet to one
sentence or a short two.
- **Optional third / fourth bullet** if the message has multiple
substantive threads worth highlighting.
Optional one-line closing punch (no bullet, no bold) that gives the
takeaway.
Rules:
- 2-5 bullets, total ~3-6 lines including header.
- Bold lead-in + em-dash + plain prose. No labels like “State:” / “Need from you:” / “Next:“. The TL;DR is a summary, not a form.
- If something needs the user, just say so plainly in a bullet (“Decision queued - pick A or B before next move”).
- Optional closing one-line observation only when it sharpens the point (“You’ve got the model. Build the right column.”). Don’t pad.
- Skip the TL;DR for purely conversational replies with no work content.
When it applies
Every agent reply in this project - Claude, Codex, future agents - that contains substantive work, findings, or decisions. The fixed format makes it portable across agents.
When it breaks
- Pure conversational replies with no work: skip - the message is already executive-sized.
- A reply where the user explicitly asks for a different format: defer to that framing.
See Also
~/.claude/projects/.../memory/feedback-executive-readout.md- project memory copy- Origin: user feedback 2026-05-21 (the earlier “Executive Readout” six-line field-labelled format was rejected as too rigid; this bullet-prose-with-bold-lead-ins format is what the user actually wants).