Evidence-Before-Edit
Do you read and test before changing code, or edit blind? Tracked across every session — the habit that separates understanding from shipping.
Not what you shipped. How you got better. Gems turns how you build with Claude Code into a shareable record of evidence — not hype.
Already published? View profile
Install once. Build normally. Your journey accumulates in the background.
Add the Gems marketplace and install into Claude Code. One command — then every session is captured when it ends.
No new workflow. Gems reads session transcripts locally and derives metrics — evidence-before-edit, steering, tool failures, evolution.
Run /gems publish when you are ready. A redacted artifact lands on your public profile — metrics and achievements, never raw transcripts.
Install command
/plugin marketplace add NitishKumar-ai/Gems/plugin install gems@gemsAll derived from your session transcripts. None of it self-reported. Earned from evidence, never granted for showing up.
Do you read and test before changing code, or edit blind? Tracked across every session — the habit that separates understanding from shipping.
How often you stop and redirect the agent instead of letting it run. A measure of active collaboration, not passive acceptance.
Failed tool calls and retries as a share of total work. Execution hygiene over time — not a single bad day, but a trend.
All of the above, plotted longitudinally. Six weeks ago you edited before reading; now you read first 90% of the time. That is the story.
Gems captures sessions locally. Metrics are derived on your machine. Publishing sends only a redacted artifact — never raw conversation history. Privacy is not a footnote; it is how the product works.
“The interesting claim is not this repo scored 82. It is: six weeks ago you edited before reading the file; now you read first 90% of the time.”
Install the plugin, build as you already do, and let the evidence accumulate. When you are ready, publish a profile that shows how you got better — not just what you shipped.