report

Report

Turn ledger facts into a report a decision-maker can verify in two minutes. Brevity is part of the credibility: one page hard cap, every section has a line budget below, and when in doubt cut the sentence, not the number. The detail level scales the budgets (config.detail_default, or --detail / the user asking for a terse or full version): terse = the headline numbers in 1–3 lines per section (never dropping unattributed %, confidence, low_confidence labels, evidence tiers, or ranges — shorter never means less honest); standard = the budgets below; full = no collapse, every receipt and caveat rendered, still one claim per line. Read skills/ledger/references/methodology.md first and enforce its language rules throughout; read references/schema.md for field semantics.

The numbers come from the engine, never from estimation:

"${CLAUDE_PLUGIN_ROOT}/bin/waybill" mine --all                 # catch-up metering first, always
"${CLAUDE_PLUGIN_ROOT}/bin/waybill" query report --from <iso> --to <iso> [--audience <a>]
"${CLAUDE_PLUGIN_ROOT}/bin/waybill" query spend  --from <iso> --to <iso>    # spend-ledger detail
"${CLAUDE_PLUGIN_ROOT}/bin/waybill" query story <KEY>                       # "what did KEY cost"

You write the prose; the engine writes the numbers. Never recompute or adjust a number the query returned.

Resolve the request

• Window: default to the current sprint for "recap", the last full quarter for "quarterly", the review period the user names for a perf review, and the period since the last allocation (from config.json.allocations) for a token pitch. Confirm the window in the output header, not with a question, unless genuinely ambiguous.
• Audience preset (D12 redaction levels):
  • token-pitch, perf-review → --audience internal (keys and URLs stay; machine-local detail dropped).
  • anything the user will paste outside the org → --audience external (keys, titles, repos pseudonymized; numbers survive). Say in the footer that identifiers are pseudonymized and the internal version is available.
  • exploring their own data → self (the default).
• Presets: token-pitch (default when the goal is more tokens), perf-review, sprint-recap, quarterly, grant-report (maintainers: what the sponsorship shipped), incident (a windowed receipts pack over an incident timeframe). (Billing paperwork belongs to the invoice skill; per-item AI-involvement statements to disclose.)