agentsclimarketplace

Dispute respond

Skill runxhq/runx/skills/dispute-respond

the governed runtime for agent skill workflows, off the leash but on the record

Install
npx -y skills add runxhq/runx --skill dispute-respond

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Prepare and explicitly file an evidence-bound payment dispute response using native receipt proof, approval, provider execution, and independent readback.

SKILL.md

5.3 KB, as published. Nobody here has run it

Dispute Respond

Turn one provider dispute event and the exact Runx payment history behind it into a reviewable response posture. A good dispute packet does not merely tell a persuasive story: it proves which charge and refund receipts were consulted, which external evidence may be cited, what remains unknown, and why the operator should accept, contest, review, or acknowledge an existing refund.

The default runner prepares evidence and wording without external mutation. The explicit file runner can submit that exact packet through a configured payment provider, but only after human approval and only when independent provider readback confirms what was filed. Neither runner closes a dispute, moves money, or invents provider acceptance.

When to use it

Use dispute-respond after a provider has created a bounded dispute event and the original charge should exist in Runx receipt history. Use refund when the operator chooses a new reversal rather than a dispute response. If the original settlement cannot be proven, gather that evidence before contesting.

How it works

  1. Normalize provider, dispute, charge, amount, currency, reason, and optional response deadline.
  2. Resolve the exact original charge and every prior refund by content-addressed Runx receipt id. Caller-supplied receipt rows or summary projections are not accepted as substitutes.
  3. Require complete production-signature verification and redacted native detail for each linked receipt tree.
  4. Admit external evidence only as bounded ref, SHA-256 digest, kind, and summary. Raw provider payloads and credentials remain outside the packet.
  5. Draft one posture using only admitted receipt ids and evidence refs.
  6. Deterministic finalization rejects invented citations, effect claims, and any completion state the evidence cannot support.
  7. When the operator deliberately invokes file, Runx validates the unchanged ready packet, binds it to a SHA-256 digest, obtains approval, and calls the configured provider under the narrow dispute.file scope.
  8. Runx projects only the filing reference, dispute id, request digest, and status from the provider result, then performs an independent dispute.read and requires those fields to match before reporting success.

A contest posture must cite delivery or consent evidence. A verified prior refund forces refund_already_sent or operator_review; the model cannot ignore it to produce a stronger contest narrative.

Inputs and result

The input includes the normalized dispute event, exact original receipt id, prior refund ids, and bounded external evidence. The result contains the posture, response summary, cited receipt ids and evidence refs, open questions, validation, and an exact provider-filing handoff for later review.

ready_for_review means only that the packet is complete enough for an operator. The preparation result always reports filing.status: not_filed, provider_status: not_called, and approval_status: not_requested.

Use the named file runner only after reviewing that packet. Supply the exact packet, the configured provider name, and one stable idempotency key for that filing. Runx injects the key into the provider request, rejects a conflicting nested key, refuses a packet whose decision or filing state has drifted, and does not accept the mutation response as proof by itself. A successful run must also return matching dispute.read evidence. Provider credentials remain in the configured adapter boundary; they never belong in skill inputs or output.

Stop conditions

  • Return needs_more_evidence for missing, invalid, local-development, or incompletely verified receipt trees.
  • Refuse caller-authored receipt detail, raw provider evidence, credentials, or citations not present in the admitted set.
  • Do not contest without evidence of delivery, consent, or the provider-specific fact on which the response depends.
  • Do not file a packet that is incomplete, failed validation, already marked as filed, or different from the packet approved by the operator.
  • Do not claim filing from the mutation response alone. Require the independent provider readback, and do not equate a filed response with dispute acceptance, closure, settlement, or a refund.

Example

A dispute alleges an unrecognized charge. Native history proves the charge but contains no consent evidence. The packet may recommend operator_review and name the missing evidence; it should not contest by inference. If a verified refund receipt already covers the amount, the posture becomes refund_already_sent and cites that exact receipt. If the operator later invokes file, the approved packet—not a rewritten summary—is what reaches the provider, and matching provider readback is the completion proof.

Agent task contract

dispute-response-draft

Choose accept, contest, refund_already_sent, needs_more_evidence, or operator_review from the admitted dispute, evidence refs, and native redacted receipt detail. Cite only exact admitted ids and refs. Never claim filing, provider acceptance, closure, settlement, or a refund not proven by the packet.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.