Session analytics (bernstein recap)

bernstein recap prints a post-run summary — task completion counts, git diff stats, quality-gate score distribution, and a cost breakdown — so an operator can see what a run accomplished without piecing it together from the task list and the logs separately.

CLI

bernstein recap                   # Rich tables
bernstein recap --as-json         # machine-readable JSON

Flag	Default	Meaning
--archive PATH	.sdd/archive/tasks.jsonl	Accepted but currently unused — see Limitations.
--as-json	off	Emit the raw JSON payload instead of Rich tables.

What it shows

The command calls the task server's GET /recap endpoint (core/routes/observability.py), which reads every task currently known to the live task store and returns:

Section	Contents
summary	Total / completed / failed task counts and success rate.
diff_stats	git diff --numstat rollup for completed tasks: files changed, additions, deletions.
quality_scores	Average quality score, A-F grade distribution, last 10 scores, and per-gate averages (lint, tests, type-check, security scan), read from .sdd/metrics/quality_scores.jsonl.
cost_breakdown	Total spend, per-model cost/tokens/invocations, and per-role cost, read from the most recent .sdd/metrics/costs_*.json snapshot.
tasks	Flat list of every task's id, title, status, role, and complexity.

In the Rich (non-JSON) view, these render as four separate tables: Recap, Git Diff Stats, Quality Scores, and Cost Breakdown.

Limitations

• --archive PATH is accepted by the CLI but is not read anywhere in the command's implementation (cli/commands/advanced_cmd.py) — the command always queries the server's /recap endpoint against the live task store, regardless of the flag's value.
• quality_scores and cost_breakdown degrade to zeroed defaults when their respective .sdd/metrics/ files are missing (a project with no quality gates or cost tracking configured still gets a valid, empty response rather than an error).
• The summary reflects whatever tasks the task server currently holds, not a fixed archive snapshot — running bernstein recap again after more tasks complete returns different numbers.

Source

• src/bernstein/cli/commands/advanced_cmd.py — bernstein recap command and table rendering
• src/bernstein/core/routes/observability.py — GET /recap endpoint, quality-score and cost-breakdown aggregation