Mentoring
Turn monitoring into growth — coaching routed to whoever can act on it, the skill gaps behind it, and the fix library below.
Coaching queue
Highest value first — named coaching, visible under Accountability-focused
Named coaching renders in an engineer's own view only — in either trust mode. At the org-wide altitude you see coaching throughput instead: 34 recs adopted · 2.1B tokens tokens recovered (30d). Individuals stay visible in their own view only.
Skill gaps
Across 69 engineers — where coaching pays off most
- Tool selection70
- Context management71
- Diff review discipline71
- Planning / plan mode72
- Security hygiene72
- Test-driven flow73
- Prompt clarity74
Recommended program
Biggest-lever intervention
Tool selection is the weakest skill in scope (avg 70/100). A 45-min clinic + paired follow-up moves the needle fastest.
Est. +9% acceptance across 69 engineersFix library
The org's best-practice fixes — each tied to the value it delivers when adopted. Assign one straight from a coaching finding.
Context-first prompting
Give Claude the files, constraints, and acceptance criteria before asking for code.
Plan mode for multi-file changes
Get an approved plan before edits on anything spanning 3+ files.
Test-driven with Claude
Write the failing test first, then let Claude implement to green.
Secret & secrets-file hygiene
Never paste secrets; use allowed-paths and env references.
Token-efficient long sessions
Scope context, /clear between tasks, prefer targeted reads.
Reviewing diffs before accept
A 30-second pre-accept checklist that catches most defects.