Features Cards

📖 This page mirrors the canonical Agent Karma docs. Edits should be made there, not in the wiki.

Your cards: Dharma, Karma & Phal

The artifacts Agent Karma generates around a session — what each one shows, when it appears, and the value it gives you.

Every session is bracketed by small, readable artifacts that turn raw events into something you can act on and share. They follow the arc the whole tool is built on — Dharma (the intent you set), Karma (the validation you did), Phal (the outcome you can trust).

Dharma Card — your intent, made explicit

Generated the moment you start a session, the Dharma Card reflects what you set out to do and how risky it is — before any code is written.

It shows	Meaning
Task & task type	What you're working on (Bug Fix, Security Fix, Refactor…)
Intent clarity	How clear your stated intent is (from the prompt-clarity hint)
Context provided	None / Partial / Good — did you give the AI files, errors, specifics?
Expected validation	Not Mentioned / Recommended / Explicit — does the task call for testing?
Risk level	Low / Medium / High — derived from the task type

Why it helps: it makes the stakes visible up front. A High-risk task with "Not Mentioned" validation is a flag you see before you trust the output — not after.

Karma Card — your shareable credential

The Karma Card is the one you show the world: a self-contained SVG that celebrates your validation practice — never usage volume, never a leaderboard rank. You generate it on demand; it renders locally with the font embedded, and nothing leaves your machine.

It carries four honest numbers:

• Karma — your rolling, self-comparative score (0–100)
• Validation — the share of your sessions with a real validation
• Best streak — your longest run of consecutive validated sessions
• Sessions — how many you've completed

…wrapped around a chakra wax-seal and a credential tied to your mood:

Mood	Credential	The line it carries
Luminous	Luminous Validator	"I write with AI — and I verify every line."
Steady	Steady Validator	"I use AI — and I check its work."
Forming	Forming Validator	"Building the habit: validate what the AI writes."
Dim	On the Path	"Learning to verify my AI-assisted work."

Why it helps: it's proof of practice you can drop in a PR, a standup, or a profile — evidence that you're holding the last line of trust, without exposing a line of your code. Save it as an SVG or a printable PDF.

Phal Card — the outcome, ready or not

Generated when you end a session, the Phal Card answers the only question that matters at the finish: is this safe to ship?

It shows	Meaning
Outcome	Ready for Review / Needs Review / High Risk / Informational
Files & test files changed	The size and shape of the change
Validation detected	Did any test/build/lint/type-check actually run?
Commands detected	The validation types and their results (never raw strings)
Recommendations	Concrete next steps ("no lint ran", "run your test command")

The verdict is earned, not flattering: Ready for Review requires both that validation ran and that your Karma Score cleared 75. A change with no validation reads as High Risk — not because you did something wrong, but because the question is still open.

Two quieter signals

• Karma Trace — a plain, chronological timeline of a session's events (14:24 Validation: Test (passed), 14:27 Commit abc1234). Metadata only, perfect for exports and audits.
• Risk Alignment — a coaching line that compares the task's risk to what you validated. It congratulates a validated high-risk change and gently flags an unvalidated one — but stays silent on low-risk work. It coaches; it never nags.

---

See these in motion in Sessions & the Dharma Card and on the Dashboard & insights — or earn your own Karma Card by practicing in the Validation Dojo.