Grade both sides of your AI coding sessions. Outcome quality meets collaboration quality. 810 sessions proved: how you participate matters more than how the AI behaves.

copied $ pip install agentsesh

v0.12.1 · Python 3.10+ · zero dependencies · GitHub · PyPI

The Partnership (43% ship rate) vs The Spec Dump (7%).
Corrections predict shipping (r=0.242). Prompt length is negative.
Engagement beats specification. See it in your own sessions.

$ sesh analyze

Session Analysis ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Duration: 47 min | 189 tool calls | ~$3.24 Session type: BUILD_TESTED Outcome ─────── Grade: B (74/100) Commits: 2 | Tests: 14 run (14 pass) Uncommitted files: 0 Collaboration ───────────── Score: A (85/100) Style: The Partnership Turns: 8 (42 words/turn avg) Corrections: 2 (25%) | Affirmations: 3 (38%) Autonomy: 24 tool calls/turn Arc: short-directive → affirmation Prompt trend: shortening Tip: This is working. Keep doing what you're doing — short directives, corrections when the AI drifts, affirmation when it delivers.

$ sesh analyze --profile

Behavioral Profile ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Sessions: 99 analyzed (of 103 total) Collaboration ───────────── Average score: 97.0 The Partnership 64 (73%) ███████░░░ ← dominant The Struggle 13 (15%) █░░░░░░░░░ The Micromanager 10 (11%) █░░░░░░░░░ The Spec Dump 1 ( 1%) ░░░░░░░░░░ Avg correction rate: 50% Avg affirmation rate: 38% ↑ Stable: early avg 94.4 → recent avg 98.8 Outcome Grade Distribution ───────────────────────── A 15 (17.6%) █████ B 14 (16.5%) █████ C 16 (18.8%) ██████ D 29 (34.1%) ███████████ F 11 (12.9%) ████ Average score: 60.1 ↑ Improving: early 59.0 → recent 72.9