NOTE · DEEP PROMPT
Read how agents are designed.
An introduction to Deep Prompt, a source-grounded archive for comparing autonomy, permissions, memory, and collaboration across coding agents.
Deep Prompt is a reading map.
It keeps source evidence and interpretation apart, so every design claim stays inspectable.
- Fixed snapshotsRead normalized prompt text pinned to a known source revision.
- Anchored annotationsFollow commentary from the exact rule that motivates it.
- Cross-agent comparisonCompare recurring design choices without flattening their context.
Four questions to take with you.
Open one agent page and read with a single lens at a time.
- Autonomy
- What can the agent do without asking?
- Permissions
- What needs approval, and what stays off-limits?
- Memory
- What survives between steps and sessions?
- Collaboration
- How are people and agents expected to work together?
Continue with the living archive.
Keep the source visible while moving from observation to comparison.
- Choose an agentStart with the system whose behavior you already know.
- Read the sourceScan the preserved prompt before opening the commentary.
- Follow annotationsStay anchored to the exact rule that triggered each note.
- Compare themesTest the same design question across several agents.