Everything it does

Agent Karma is small on the surface and deliberate underneath. Here’s the whole surface in one place — grouped by what you’ll actually reach for.

Run a session

• Start / End Session — bracket a piece of work with an intent. At the end you get a Karma Score and a Phal Card. Why: it turns “I used AI” into “I used AI and here’s whether I checked it.”
• Status-bar toggle — one click to start or end from anywhere. Why: zero friction means you actually do it.
• Ambient mode (optional) — keeps one rolling session per day and captures saves and validations hands-free. Why: track the habit without remembering to press start.

Log what you validated

• Log a Validation — record a check you ran when shell integration can’t see it. Why: covers browser AI, copy-paste, and work outside the IDE.
• @agentkarma /verify — the same, from the chat box: “I ran the tests, they passed.” Why: log a validation without leaving your train of thought. (See the chat participant.)

See your practice — the dashboard

The dashboard turns your sessions into a picture you can read in seconds:

• At a glance — rolling Karma + trend, validation rate, sessions, and a tally of Phal outcomes.
• Karma sparkline — your recent scores at a glance.
• Consistency strip — your last 24 sessions as green (validated) / red (unvalidated) / grey (nothing to validate) cells. Why: streaks and slips become obvious.
• Risk × validation heatmap — which checks you run for which kinds of task. Why: find the blind spot (“I never lint my migrations”).
• Validation habits & trend — per-check rates (Test, Build, Lint, Type Check) and whether each is rising, steady, or slipping.

…plus derived insights that need no extra capture:

• High-risk watchlist — high-risk sessions you never validated, as a to-do list.
• Score composition — how much of your score is real verification vs. cheap points.
• Usage attribution (with Claude Code reading on) — tokens spent on validated vs. unvalidated work.
• Karmic message — one human line that reflects your mood (luminous → on the path) and ties back to your latest Dharma and Phal.

Know if you can validate — readiness

• Check Validation Readiness — scans the repo for test/build/lint/type-check tooling, CI, a pre-commit hook, and whether your agent guidance (CLAUDE.md / AGENTS.md) even mentions validation. Gives you a 0–100 readiness score and the single biggest gap. Why: before it coaches whether you validated, it checks whether you’re set up to.
• Find Validation Gaps — suggests the next validation skill to build from your own habits.

Safety nets

• Pre-commit nudge (opt-in) — a non-blocking reminder at git commit if the files you’re committing were never validated. It’s change-aware, and it won’t clobber an existing husky/lefthook setup. Why: a last, gentle check exactly when it counts. (See Readiness, nudge & chat.)
• @agentkarma /summary — your latest Karma, this week’s nudge, and your readiness status, on demand in chat.

Reflect & share

• Generate Karma Card — your shareable validation credential as SVG or printable PDF.
• Why did my Karma move? — compares your last two scored sessions and explains the change.
• Weekly reflection — the week’s highest-leverage nudge.
• Export — save any session as JSON or Markdown (timeline + summary).

Own your data

• Reset history — clear sessions and your Karma trend, keep your settings.
• Delete all local data — a full wipe, no trace.
• Settings — every capture toggle in one place. Why: nothing is captured that you didn’t agree to. (See Settings and the full command list.)

---

Want the why behind all of this? Read How it brings value. Want to see exactly what each feature records? Read What’s captured — and what never is.