waybill 1.8.0

"Many readers": the same receipts, new audiences — the third release of
the committed path to complete (ROADMAP). Engine-side cost
held to three additive fields; everything else is rendering, which is the
completion thesis working as designed.

Added

• The invoice skill (+ trigger eval) — billing paperwork from the
  receipts, two modes: the client invoice pack (shipped entries as
  line items with recorded actual_hours, artifacts, and per-item
  disclosed AI cost — counterfactual hours are never presented as
  billable time, and pricing stays the user's business) and the
  personal expense receipt (a period's metered costs with
  pricing_coverage honesty plus the CSV export for finance, labeled
  list-price equivalent — the provider bill says what was paid; this
  says what it was for).
• The disclose skill (+ trigger eval) — "was AI used here?"
  answered from the meter instead of memory: per item, the recorded
  claude_role (labeled as the user's declaration), metered sessions
  and tokens (labeled as assistance volume, never an authorship
  percentage), the ship receipt, and the verification basis. Single-item
  statements or a window's AI-disclosure register — including
  role-none rows, because a register that can't prove a negative isn't
  one. IC-owned, handed over per item; no standing feeds, no colleague
  data, ever.
• Report presets grant-report and incident — the maintainer
  grant report (what the sponsorship shipped, funder verification via
  export --pack) and the incident receipts pack (tight window,
  work_type: "incident" rows, timeline facts, no efficiency math —
  nobody prices a fire by the token).
• The portable career ledger — documented exit path in the ledger
  skill: export --format json --audience external over all history
  gives a pseudonymized, numbers-intact record that follows the user
  between jobs; the home repo clone remains the raw, lock-in-free copy.
• Engine: three additive fields on report shipped rows —
  work_type, sessions (distinct metered sessions for the story in
  the window; a count, so it survives every redaction level), and
  metered_cost_usd (the story's windowed spend-account cost, null when
  unpriced — never $0). These power the presets above; no new commands,
  no storage change.

CHANGELOG · VALIDATION · DECISIONS