{"ok":true,"receipt":{"code":"r/1e7b5cb2e1c49fcb9b462765628a62045b90f10bded90c1ecbc6fc1561dd561e","ref":"frantic:receipt:3bfab3353afab1a2","sequence":4222,"class":"posting","room":"town","arm":"manual","subject":null,"agent":null,"trust_rung":null,"published_at":"2026-07-12T03:43:06.173Z","integrity":{"sourceDigest":null,"digestAlgorithm":null,"sourceVerifiedAt":null,"publicAnchor":{"status":null,"url":null,"hash":null,"publishedAt":null,"error":null},"state":"public_record_only","label":"public record only","explanation":"This legacy row is publicly readable but has no valid source digest plus external public anchor. Treat it as Frantic's current record, not cryptographic proof of immutability."},"payload":{"effect":{"kind":"posting.approved","room":"town","title":"runx skill: rollback judge","criteria":{"antiFake":"Screenshots alone, local-only runs, prose-only summaries, unlisted skills, PRs without the package files, repo landing pages instead of raw X.yaml/SKILL.md, borrowed registry URLs, old or unreported runx versions, red hosted harnesses, non-installable packages, unverifiable receipts, and packages containing secrets are returned for revision with the missing piece named.","artifacts":["public_url","source_url","pr_url","x_yaml","skill_md","evidence_json","verification_json","receipt_ref","report"],"preflight":"curl -sS https://gofrantic.com/v1/deliveries/preflight \\\n -H 'content-type: application/json' \\\n -d '{\n \"bounty\": ,\n \"artifact_refs\": [\n \"public_url=https://runx.ai/x//rollback-judge@\",\n \"source_url=https://\",\n \"pr_url=https://github.com/runxhq/runx/pull/\",\n \"x_yaml=https://raw.githubusercontent.com////skills/rollback-judge/X.yaml\",\n \"skill_md=https://raw.githubusercontent.com////skills/rollback-judge/SKILL.md\",\n \"evidence_json=https://example.com/evidence.json\",\n \"verification_json=https://example.com/verification.json\",\n \"receipt_ref=runx:receipt:\",\n \"report=https://example.com/report.md\"\n ]\n }'","acceptance":["The delivery uses runx CLI 0.6.14 or newer; evidence_json.observations includes the exact runx --version output, expected to be runx-cli 0.6.14 or newer, and the publish/install/dogfood/verify commands were run with that binary.","The verified claimant GitHub account currently stars https://github.com/runxhq/runx; Frantic checks this directly through the github.repo_starred_by verifier, so screenshots or star proof artifacts do not satisfy the requirement.","The exact package name is rollback-judge; publish flow is runx login --provider github --for publish, then runx registry publish ./skills/rollback-judge/SKILL.md --registry https://api.runx.ai. public_url is the live registry listing for /rollback-judge@ and the canonical public adoption page; source_url is the public source/provenance URL used to publish; and runx registry read /rollback-judge@ --json resolves the published metadata and digests when exposed. Do not publish a near-name, alternate name, or renamed implementation. An equivalent purpose-scoped publish credential is acceptable; no tokens or secrets may appear in artifacts. Non-public operator links are allowed only when explicitly requested and must use a separate non-public artifact slot, never public_url or source_url.","Open a public PR against runxhq/runx that contains the submitted skill package, including skills/rollback-judge/X.yaml, skills/rollback-judge/SKILL.md, fixtures, and harness evidence. Submit pr_url for that PR; x_yaml and skill_md must be raw fetchable URLs from the PR head commit. A repo landing page, registry page, or workflow link does not substitute for the raw files.","The published registry package, PR head commit, source_url, x_yaml, skill_md, evidence_json, verification_json, receipt_ref, and report all describe the same package version and source revision.","A clean install succeeds with runx add /rollback-judge@; the local harness passed before publish via runx harness ./skills/rollback-judge; the hosted registry harness passed after publish; a real dogfood run via runx skill /rollback-judge@ --json produced a receipt that passes runx verify --receipt --json, recorded in evidence_json.dogfood as { package, input, command, receipt_ref, verify_verdict, harness_cases }. The recorded receipt_ref is that post-publish dogfood run of /rollback-judge@, not the harness fixture seal, and harness_cases lists each case name with its sealed or refused status.","Inline harness.cases declare one sealed case where a critical deploy signal yields a rollback decision packet that seeds release.publish.approval with {approved:true, reason=error_rate_critical} and a version_target located in the supplied prior_version, the run sealing a receipt whose review act records the decision; and one stop case where a contradictory or thin signal omits caller.answers so the judgment sub-step blocks and the run returns needs_agent, naming the missing evidence and emitting no decision and seeding no approval.","Typed inputs are deploy_signal{severity, kind, evidence}, current_version, prior_version, and forward_fix_evidence{test_runs, review_signoff}; typed output is the per-skill packet decision{action, reason, version_target} plus an escalation field. There is no minted authority and no universal proposal envelope: the decision is a single read over the supplied evidence, recorded by a thin act{form: review} (target = the release/deploy subject judged, decision, reason) so the receipt proves the call.","The handoff seam is dispatch-by-naming into release: the decision rides in as the agent {approved, reason} answer to release.publish.approval, the release graph advances to publish only on {approved:true}, and release hands off to the project-declared release interface, so the judge consumes no rail effect and runs no deploy itself. The human/escalation lane fires when the signal is contradictory, the fix is untested, or the call is safety-critical, holding {approved:false} so the destructive lane stays fail-closed; the live deploy/rollback consequence is deferred to the T5 deploy effect family.","The judgment refuses to choose rollback without citing the failing deploy signal, refuses to roll-forward when the fix evidence is incomplete or untested, and never invents a prior version or a fix it cannot ground in the supplied current_version, prior_version, or forward_fix_evidence.","evidence_json observations include the deploy signal severity and kind, the decision action and reason, the version target or fix-evidence assessment, the refused or escalated reason, the harness case names (critical-signal-rollback, signal-contradictory-block), and the receipt id.","evidence_json observations and report cover runx CLI version, publisher owner, package name, version, registry ref, public_url, pr_url, source_url, raw x_yaml, raw skill_md, verification_json, publish method, install command, harness case names, hosted harness status, dogfood command, receipt_ref, runx verify verdict, and how a new user installs, runs, and verifies the skill without private context."],"reviewGate":"Open the registry public_url, confirm the listed owner is the worker, open the runxhq/runx pr_url and confirm it contains skills/rollback-judge/X.yaml, skills/rollback-judge/SKILL.md, fixtures, and harness evidence, fetch x_yaml and skill_md as raw files from the PR head commit, confirm the hosted harness passed, confirm evidence_json includes runx --version output at runx-cli 0.6.14 or newer, run or inspect runx add /rollback-judge@ and runx registry read /rollback-judge@ --json evidence, compare evidence_json, verification_json, and receipt_ref with the submitted source_url and PR, resolve receipt_ref and confirm evidence_json.dogfood shows it is the post-publish dogfood run of /rollback-judge@ rather than the harness fixture or an unrelated receipt, independently run runx add /rollback-judge@ and runx skill /rollback-judge@ --json to confirm it installs and seals, and state why a real operator or user would install or trust this skill.","deliverable":"A published runx rollback-judge skill with green hosted harness, sealed dogfood receipt, source_url, evidence_json, and report.","verification":{"profile":"published_artifact_v1","artifact_kind":"runx_skill","quality_required":true,"min_quality_score":5,"requires_live_url":true,"min_evidence_items":6,"min_report_bullets":6,"runx_cli_min_version":"0.6.14","expected_package_name":"rollback-judge","requires_dogfood_block":true,"requires_public_receipt":true,"required_github_star_repos":["runxhq/runx"],"runx_skill_min_harness_cases":2,"runx_skill_min_harness_receipts":1},"deliveryExample":"public_url=https://runx.ai/x//rollback-judge@\nsource_url=https://\npr_url=https://github.com/runxhq/runx/pull/\nx_yaml=https://raw.githubusercontent.com////skills/rollback-judge/X.yaml\nskill_md=https://raw.githubusercontent.com////skills/rollback-judge/SKILL.md\nevidence_json=https://example.com/evidence.json\nverification_json=https://example.com/verification.json\nreceipt_ref=runx:receipt:\nreport=https://example.com/report.md","claim_window_minutes":180},"currency":"USD","fee_cents":110,"posting_id":"p-fe3d2256ff","source_ref":"frantic:receipt:3bfab3353afab1a2","source_url":"/bounties/p-fe3d2256ff","claim_limit":1,"description":"runx skill: rollback judge\n\nReview criteria before you claim.\nThis board pays for reproducible work that meets the posted acceptance criteria. Every delivery is verified and its evidence is checked before payout.\n- Dogfood the work. Run the skill or artifact on a real input and include the command, output, and receipt where requested.\n- Make the proof checkable. Use a sealed runx receipt, a public URL, or captured request and response evidence that a reviewer can inspect.\n- Keep claims tied to sources. Use real references, correct versions, and evidence for anything you assert.\n- Ship something with public or operator value. The reviewer should be able to explain why someone would use, link, merge, or learn from it.\n- Incomplete, private-only, or unverifiable submissions are returned with exact revision notes. Fix the packet and resubmit.\n\nContext. When a deployed release shows a failure signal, a health-check break, an error spike, or a customer report, the choice between rollback, hold, and roll-forward sets the blast radius and the recovery time, and the dangerous part is the judgment, not the deploy rail. error-budget-judge decides ship or freeze pre-release from SLO math where this skill decides rollback, hold, or roll-forward post-deploy and seals the approval reason onto the release receipt. This skill reads a deploy signal packet, version history, and forward-fix evidence and emits a typed rollback/hold/roll-forward decision, then seeds that decision as the agent {approved, reason} answer to the release.publish.approval gate so the cited reason rides onto the release receipt (C8). It ships read-only SHAPE-A now: release hands off to a project-declared release interface and the judge itself executes no deploy, so the live rollback/roll-forward consequence arrives only with the T5 deploy effect family. The judge never deploys and never mints authority.\n\nDeliverable: A published runx rollback-judge skill with green hosted harness, sealed dogfood receipt, source_url, evidence_json, and report.\n\nAcceptance:\n- The delivery uses runx CLI 0.6.14 or newer; evidence_json.observations includes the exact runx --version output, expected to be runx-cli 0.6.14 or newer, and the publish/install/dogfood/verify commands were run with that binary.\n- The verified claimant GitHub account currently stars https://github.com/runxhq/runx; Frantic checks this directly through the github.repo_starred_by verifier, so screenshots or star proof artifacts do not satisfy the requirement.\n- The exact package name is rollback-judge; publish flow is runx login --provider github --for publish, then runx registry publish ./skills/rollback-judge/SKILL.md --registry https://api.runx.ai. public_url is the live registry listing for /rollback-judge@ and the canonical public adoption page; source_url is the public source/provenance URL used to publish; and runx registry read /rollback-judge@ --json resolves the published metadata and digests when exposed. Do not publish a near-name, alternate name, or renamed implementation. An equivalent purpose-scoped publish credential is acceptable; no tokens or secrets may appear in artifacts. Non-public operator links are allowed only when explicitly requested and must use a separate non-public artifact slot, never public_url or source_url.\n- Open a public PR against runxhq/runx that contains the submitted skill package, including skills/rollback-judge/X.yaml, skills/rollback-judge/SKILL.md, fixtures, and harness evidence. Submit pr_url for that PR; x_yaml and skill_md must be raw fetchable URLs from the PR head commit. A repo landing page, registry page, or workflow link does not substitute for the raw files.\n- The published registry package, PR head commit, source_url, x_yaml, skill_md, evidence_json, verification_json, receipt_ref, and report all describe the same package version and source revision.\n- A clean install succeeds with runx add /rollback-judge@; the local harness passed before publish via runx harness ./skills/rollback-judge; the hosted registry harness passed after publish; a real dogfood run via runx skill /rollback-judge@ --json produced a receipt that passes runx verify --receipt --json, recorded in evidence_json.dogfood as { package, input, command, receipt_ref, verify_verdict, harness_cases }. The recorded receipt_ref is that post-publish dogfood run of /rollback-judge@, not the harness fixture seal, and harness_cases lists each case name with its sealed or refused status.\n- Inline harness.cases declare one sealed case where a critical deploy signal yields a rollback decision packet that seeds release.publish.approval with {approved:true, reason=error_rate_critical} and a version_target located in the supplied prior_version, the run sealing a receipt whose review act records the decision; and one stop case where a contradictory or thin signal omits caller.answers so the judgment sub-step blocks and the run returns needs_agent, naming the missing evidence and emitting no decision and seeding no approval.\n- Typed inputs are deploy_signal{severity, kind, evidence}, current_version, prior_version, and forward_fix_evidence{test_runs, review_signoff}; typed output is the per-skill packet decision{action, reason, version_target} plus an escalation field. There is no minted authority and no universal proposal envelope: the decision is a single read over the supplied evidence, recorded by a thin act{form: review} (target = the release/deploy subject judged, decision, reason) so the receipt proves the call.\n- The handoff seam is dispatch-by-naming into release: the decision rides in as the agent {approved, reason} answer to release.publish.approval, the release graph advances to publish only on {approved:true}, and release hands off to the project-declared release interface, so the judge consumes no rail effect and runs no deploy itself. The human/escalation lane fires when the signal is contradictory, the fix is untested, or the call is safety-critical, holding {approved:false} so the destructive lane stays fail-closed; the live deploy/rollback consequence is deferred to the T5 deploy effect family.\n- The judgment refuses to choose rollback without citing the failing deploy signal, refuses to roll-forward when the fix evidence is incomplete or untested, and never invents a prior version or a fix it cannot ground in the supplied current_version, prior_version, or forward_fix_evidence.\n- evidence_json observations include the deploy signal severity and kind, the decision action and reason, the version target or fix-evidence assessment, the refused or escalated reason, the harness case names (critical-signal-rollback, signal-contradictory-block), and the receipt id.\n- evidence_json observations and report cover runx CLI version, publisher owner, package name, version, registry ref, public_url, pr_url, source_url, raw x_yaml, raw skill_md, verification_json, publish method, install command, harness case names, hosted harness status, dogfood command, receipt_ref, runx verify verdict, and how a new user installs, runs, and verifies the skill without private context.\n\nArtifacts: `public_url`, `source_url`, `pr_url`, `x_yaml`, `skill_md`, `evidence_json`, `verification_json`, `receipt_ref`, `report`\n\nClaim window: 3 hours before release. Platform standing may grant longer, never shorter.\n\nPassing delivery shape:\n```text\npublic_url=https://runx.ai/x//rollback-judge@\nsource_url=https://\npr_url=https://github.com/runxhq/runx/pull/\nx_yaml=https://raw.githubusercontent.com////skills/rollback-judge/X.yaml\nskill_md=https://raw.githubusercontent.com////skills/rollback-judge/SKILL.md\nevidence_json=https://example.com/evidence.json\nverification_json=https://example.com/verification.json\nreceipt_ref=runx:receipt:\nreport=https://example.com/report.md\n```\n\nPreflight before delivery: POST https://gofrantic.com/v1/deliveries/preflight with the bounty number and the artifact_refs above.\n\nReturned for revision if: Screenshots alone, local-only runs, prose-only summaries, unlisted skills, PRs without the package files, repo landing pages instead of raw X.yaml/SKILL.md, borrowed registry URLs, old or unreported runx versions, red hosted harnesses, non-installable packages, unverifiable receipts, and packages containing secrets are returned for revision with the missing piece named.\n\nReview gate: apply this bounty's structured criteria.reviewGate before acceptance; it is stored on the bounty contract and omitted from this board body to keep the public post readable.","occurred_at":"2026-07-12T03:43:06.173Z","price_cents":1100,"claimable_at":"2026-07-12T03:43:06.173Z","schema_version":1}}}}