Sessions & the Dharma Card

A session is one unit of AI-assisted work — you start it with intent, code with your AI, and end it to get your reflection.

The lifecycle

• Start — set your intent, the AI tool, and the task type. This produces your Dharma Card: a quick read on intent clarity (a soft prompt-hygiene hint), the context you gave, the validation the task expects, and a risk level (Low / Medium / High).
• During — Agent Karma quietly notes file-change events and any validations you run — metadata only. The status bar shows you’re recording.
• End — it attaches a git diff summary (counts only), runs the validation checklist, computes your Karma Score, and generates the Phal Card (outcome).

Safety nets

Because sessions are manual, Agent Karma guards against forgetting:

• A forgot-to-start nudge if you’ve been editing without a session.
• Survive-reload & crash recovery — an active session is restored after a window reload or a hard kill, with elapsed time recomputed.
• A stale-session prompt if a recovered session has been idle past idleEndMinutes.

Logging validations

The most important habit: tell Agent Karma what you checked. Use ✅ Log a Validation, @agentkarma /verify, or just run tests/build/lint in the integrated terminal and they’re detected automatically. See Validation types.