amadeus-session-cost

AI-DLC Session Cost

Purpose

Give the team a transparent, deterministic view of what the current workflow has consumed: how long it has run, how many stages have cleared their gates, how much the orchestrator wrote to its observation diaries, how often sensors fired, and how many learnings were captured.

Every number this skill prints comes from bun {{HARNESS_DIR}}/tools/amadeus-runtime.ts summary --json — the materialised, event-sourced view over runtime-graph.json. This skill does no counting of its own. It does not estimate tokens, does not walk the artefact tree, and does not read audit.md. If a number isn't in the tool's output, this skill does not invent it.

Classification

Read-only. This skill never advances the workflow stage pointer, never emits an audit event, and never writes a file. It is safe to run at any point in a workflow, including mid-stage.

Steps

Step 1: Read the aggregates

Run:

bun {{HARNESS_DIR}}/tools/amadeus-runtime.ts summary --json

If the command exits non-zero (no runtime-graph.json yet — the workflow hasn't compiled a graph), print:

No session data yet.

Session cost becomes available once a workflow has started and its
first stage transition has compiled runtime-graph.json. Run /amadeus to
begin, then re-run /amadeus-session-cost.

and STOP.

Otherwise parse the JSON. The shape is:

{
  "workflow_id": "...",          // ISO timestamp of the live workflow
  "scope": "...",
  "started_at": "...",
  "duration_minutes": 40,         // null when nothing has completed yet
  "stages":   { "total": N, "approved": N, "failed": N, "pending": N },
  "by_phase": { "<phase>": { "total": N, "approved": N, "failed": N, "pending": N }, ... },
  "memory":   { "total": N, "interpretations": N, "deviations": N, "tradeoffs": N, "open_questions": N },
  "sensors":  { "total": N, "passed": N, "failed": N, "budget_override": N, "incomplete": N },
  "learnings":{ "from_orchestrator": N, "from_user_addition": N }
}