tricorder · impact-advocate

Make the case for your impact, from your own GitHub record

impact-advocate turns your commits, pull requests and reviews into a document for a self-review, promotion case or manager 1:1. It shows where you exceed and where you meet the expectations of your role. Every claim is linked or counted.

3 stepsClaude CodeNo API key, no PythonMarkdown version

Private by design. It runs inside your own Claude Code session with your own GitHub sign-in. No server, no telemetry, no API key. It reads your repositories and changes nothing in them. On an employer's repository, check that company policy lets your Claude plan read the code.

Only for yourself. It assesses the person running it and declines to assess anyone else (constitution §6).

Install

In Claude Code (v2.1.275 or later), run:

/plugin install impact-advocate --marketplace dhk/tricorder

Confirm the marketplace, pick Install for you, and you're done. No clone, no Python.

Older Claude Code, or installing from your shell

claude plugin marketplace add dhk/tricorder
claude plugin install impact-advocate@tricorder

The second command installs it for the next session; in an open one, run /reload-plugins.

Run

Open Claude Code in a repository you work on (or anywhere) and type:

/impact-advocate:impact-advocate

Typing /impact is enough to find it. You can also just ask, for example "help me write my self-review from GitHub". The skill takes it from there, and you'll always see where you are:

Step 2 of 4: Confirm  ·  Next: build the document  ·  Undo: say "start over"

Check. It reads what's already set up: GitHub sign-in, your login and commit emails, local clones, incomplete (shallow) clones. Read-only. You do nothing.
Confirm. It shows what it found, then asks only what it couldn't find out: time window, purpose and level, what to measure against, your proudest work. One question at a time, each with a recommended answer you can accept in a word.
Build. It gathers the evidence, rates each expectation Exceeds, Meets or Not visible in GitHub, and writes impact-<login>-<start>-to-<end>.md.
Review. It tells you what to check before you share. Ask for changes in plain words.

Your answers are saved to impact-profile.md, after you see them. Next time, setup is one question. Edit that file to change an answer; delete it to start fresh.

Not signed in to GitHub? It still works from local git history, but reviews and PR timing will be missing. The skill tells you when that's the case; run gh auth login and say "re-check".

Review before you share

Open every link. The skill never invents evidence, but your name is on the document.
Re-run two or three commands from the methodology appendix and check the numbers.
Check that no colleague is named, ranked or quoted.
Check how co-authored and AI-assisted commits are described. The document discloses them because anyone can see them in the log.
Fill Not visible in GitHub with what you'll bring instead: design docs, incident records, peer feedback.

Optional

Make it stronger with tricorder

Counts flatten code review, which is often where senior impact shows. After the document is written, the skill may offer tricorder, a separate tool that maps ownership, expertise and how you review. It names which of your ratings tricorder would help with and installs it with one command if you agree. Skip it and the document is still complete.

Undo

Undo and uninstall

To	Do
Change an answer	Say "change the window to 12 months", or edit `impact-profile.md`
Start setup over	Say "start over", or delete `impact-profile.md`
Remove the document	Delete the `impact-…md` file; nothing else was written
Uninstall the skill	`/plugin uninstall impact-advocate@tricorder`
Uninstall tricorder	`pipx uninstall tricorder` or `uv tool uninstall tricorder`

Help

Troubleshooting

Symptom	Fix
`/plugin install … --marketplace` isn't recognized	Update Claude Code, or use the two shell commands under Install.
`/impact` shows nothing	Run `/reload-plugins`, then check `claude plugin list` shows `impact-advocate@tricorder`.
No review or PR timing data	It had no GitHub access. Run `gh auth login`, then say "re-check".
Commit counts look low	It flags shallow clones and unknown emails during Check; accept its fix, or add the email to `impact-profile.md`.
It won't assess a colleague	Working as designed: it only builds your own packet.

Other agents and Claude.ai

Claude.ai. Turn on code execution (Settings → Capabilities; on Team and Enterprise an admin enables skills). Download skills/impact-advocate, zip it so impact-advocate/ is the top level, and upload it under Customize → Skills. Claude.ai can't see your local clones, so it works only from GitHub data. (Using skills in Claude)

Any other agent. Give it skills/impact-advocate/SKILL.md as instructions and the files in references/ as context.

Limits. GitHub isn't the whole job: incidents, design reviews and mentoring often leave little trace. Team comparisons need several contributors; in a solo repository the skill reports no baseline. The skill has been evaluated on two scenarios, both on a single-contributor repository using git only.