/devpersona:stats
The numeric usage profile: projects, sessions, prompts, tool calls, errors, permission denials and your peak hours. Numbers only — no interpretation.
00 hailneed suite · module 06 · developer profile
Claude Code, Codex CLI and Gemini CLI keep a record of every session on your disk. devpersona reads that record through one shared adapter and builds a profile in three layers — numbers first, then an evidence-cited character card, then concrete suggestions. Nothing leaves your machine.
01 Evidence
Abridged from an actual run of node scripts/scan.mjs --md on the
machine this site was built on. Project names generalized; every number is what
the scanner printed. Nothing in it ever left that machine.
# Coding agent usage profile Source: Claude Code, Codex CLI · 187 sessions scanned ## Totals | Projects | 53 | | Sessions | 187 | | Human prompts | 644 | | Tool calls | 15072 | | Tool errors | 609 | | Permission denials | 55 | ## Work rhythm - Peak hours: 11:00 (95), 16:00 (82), 15:00 (80) - Peak days: Monday (184), Friday (151), Tuesday (106) ## Projects (by prompt count) | Project | Agents | Prompts | Languages | |----------------|--------------------|---------|-----------------------| | aspnet-lob-app | claude-code | 85 | C#, JavaScript, Razor | | media-pipeline | claude-code, codex | 78 | Markdown, Python | | client-site | codex | 57 | HTML, CSS, JavaScript | > This report was generated entirely locally; no data left this machine.
Projects are grouped by working directory, so one worked on in two agents — like
the pipeline above — appears once, with the agents named, never double-counted.
The adapter layer is
one file you can open,
and --selftest proves the classifier against 15 fixtures without
reading your data at all.
02 Install
# once > /plugin marketplace add hailneed/plugins # the profiler > /plugin install devpersona@hailneed # and try it > /devpersona:analyze
node scripts/scan.mjs --md gives you the numbers without the plugin.--agent codex limits to one agent, --days 90 narrows the window, --lang tr switches the report to Turkish.03 Commands
/devpersona:statsThe numeric usage profile: projects, sessions, prompts, tool calls, errors, permission denials and your peak hours. Numbers only — no interpretation.
/devpersona:analyze
A persona card — your stack, working style, communication style, strengths and
risk patterns — written to ~/.claude/persona/persona.md. Every
observation cites the evidence it came from: the numbers, the project memory
files, a small prompt sample.
/devpersona:advise
Turns the card into concrete suggestions: agent skills worth authoring — drafts
included — coding-habit fixes and architecture recommendations. Focus with
skills, code or architecture. A
suggestion without evidence does not make the report.
04 Trust
--prompts <project> prints them to your terminal and nowhere else.--selftest runs 15 fixtures without touching your history or the network, and prints the score.05 Local
There is no devpersona Cloud, and none is planned. A profile built from your own chat history is exactly the kind of thing that should never sit in someone else's dashboard.
~/.claude/persona/persona.md. Edit it, share it, delete it; it answers to you.06 Family
What your agent actually did — and what can still be undone.
Is the skill written right — and does it ever actually fire?
Failures that keep repeating, written down — then measured.
What CLAUDE.md costs you on every single request.
Public complaints plus verified revenue — themes with links.