verification-receipts

Attach a ledger-backed receipt to every completion claim so no Bot asserts success without proof the desk can re-check.

SKILL.md

GitHub

Verification Receipts

L1 - Summary

No completion claim without a receipt. Every Bot that reports work done attaches a receipt the desk can re-check with desk.receipt_check: what was claimed, which ledger rows prove it, and who approved it. An unreceipted claim is an unmade claim.

claim made -> receipt attached (desk.receipt_check exit 0) -> CHIEF accepts
claim made -> no receipt                       -> CHIEF rejects, work repeats

L2 - Method

  1. S1 - State the claim narrowly: ticket id, action taken, and the ledger event types that prove it (OrderPlaced, FillRecorded, JournalNote). Vague claims ("looks good") are not claims.
  2. S2 - Query the proof with desk.receipt_check: pass the ticket id and the expected event kinds. The tool returns PASS only when the ledger holds matching rows; otherwise it names what is missing.
  3. S3 - Record the approval: every receipt names approved_by (a seat for autonomous steps, human:<id> for human-gated steps per G-2). Self-approval is void.
  4. S4 - Chain the receipt: the receipt for step N references the receipt for step N-1, so a broken chain points at the exact step that lost its proof.
  5. S5 - Surface failures plainly: a FAIL names the missing row and the step to repeat. Never edit the ledger to manufacture a PASS; append-only always wins.

Failure modes

Symptom Cause Action
Receipt FAIL on a real fill Wrong ticket id or event kind in the query Re-query with the exact ticket id from pumpgrok ticket list; check the kind spelling
Missing approval Step ran without an approval record Route through RISK or the human; backdated approvals are void
Broken receipt chain An intermediate step skipped its receipt Re-run the missing step with its receipt; do not splice the chain
Ledger row absent The write never happened (crash before append) Treat the step as not done; repeat the step, never forge the row

References

  • Tool: desk.receipt_check (receipt query; resolvable by every seat via _core)
  • Receipt schema: contracts/tool-rosters/_core.yaml (core eight)
  • Gate G-2: approval integrity behind every receipted claim
  • Companion skills: desk-doctor (desk health proof), desk-post-trade-review (graded review rows)