Make Runtime State Agent Legible
make-runtime-state-agent-legible
Summary
Agents can only validate behavior reliably when the relevant runtime state is observable through their tools. Expose the smallest useful signal first, such as app boot, UI inspection, logs, metrics, screenshots, traces, or generated artifacts.
Problem
Agents cannot reliably validate behavior they cannot observe. If UI state, logs, metrics, traces, screenshots, or local app instances are only available to humans, the agent must guess whether the change worked.
Preferred Move
Expose runtime state through tools the agent can query. Make local app boot, UI inspection, logs, metrics, traces, screenshots, and generated artifacts available per task or worktree when they are needed for validation.
Tradeoff
Do not build a full observability harness for every small repo. Start with the smallest signal that lets the agent prove the changed behavior, then add richer runtime visibility for repeated or high-risk workflows.
Agent Instruction
When behavior depends on runtime state, use or create a legible validation path. Report the runtime
signal used, not only the code change.Examples
Bad: the agent claims a UI fix without seeing the UI.
Changed the modal state handling; the dialog should close now.
Good: the agent uses runtime evidence.
Launched the local app, opened the modal, clicked Save, and captured before/after screenshots showing
the dialog closes and the success toast appears.
References
| Source | Use | Note |
|---|---|---|
| OpenAI application legibility | supports |
Agents need direct access to app state and observability signals. |