Add deliverability judge skill

skills/deliverability-judge/README.md

@@ -0,0 +1,22 @@
+# deliverability-judge
+
+A native runx skill: a `SKILL.md` contract, an `X.yaml` execution profile, and a
+`run.mjs` script. No build step and no dependencies.
+
+## Develop
+
+```bash
+runx harness . --json                       # run the harness cases in X.yaml
+runx skill . --input message=hello --json   # run the skill once
+runx history                                # inspect the signed receipt
+```
+
+Edit `run.mjs` to do the real work, and keep both harness classes in `X.yaml`:
+one happy path and one stop, error, or refusal case.
+
+## Publish
+
+```bash
+runx login --provider github --for publish
+runx registry publish .   # the registry runs the harness as the publish gate
+```

skills/deliverability-judge/SKILL.md

@@ -0,0 +1,59 @@
+---
+name: deliverability-judge
+description: Read-only deliverability posture judge that fuses sealed provider evidence into a continue/throttle/pause recommendation or a human-escalation refusal.
+source:
+  type: cli-tool
+  command: node
+  args:
+    - run.mjs
+  timeout_seconds: 30
+  sandbox:
+    profile: readonly
+    cwd_policy: skill-directory
+inputs:
+  evidence:
+    type: object
+    required: true
+    description: Sealed deliverability evidence with postmaster_report, bounce_metrics, complaint_metrics, and placement_probe.
+  policy:
+    type: object
+    required: true
+    description: Operator thresholds with min_reputation_score, max_bounce_pct, and max_complaint_pct.
+runx:
+  category: ops
+  input_resolution:
+    required:
+      - evidence
+      - policy
+---
+
+# deliverability-judge
+
+Use this skill when an operator needs a read-only judgment on whether a sending
+posture is healthy enough to continue sending.
+
+The skill fuses four sealed provider signals:
+
+- `postmaster_report.reputation_score`
+- `bounce_metrics.bounce_pct`
+- `complaint_metrics.complaint_pct`
+- `placement_probe.passed`
+
+It compares those signals against an operator policy:
+
+- `min_reputation_score`
+- `max_bounce_pct`
+- `max_complaint_pct`
+
+The output is a JSON verdict. When all signals are sealed, present, and
+non-contradictory, the verdict includes a read-only recommendation:
+
+- `continue` for healthy posture
+- `throttle` for moderate aligned risk
+- `pause` for clearly unhealthy aligned risk
+
+When signals are missing, unsealed, or contradictory, the skill refuses to emit
+a recommendation and returns an escalation record instead.
+
+This skill is read-only. It mints no authority, holds no state, emits no Effect,
+and never performs a send, throttle, payment, or operational handoff.

skills/deliverability-judge/X.yaml

@@ -0,0 +1,83 @@
+skill: deliverability-judge
+version: "0.1.0"
+
+catalog:
+  kind: skill
+  audience: public
+  visibility: public
+  role: canonical
+
+harness:
+  cases:
+    - name: sealed_healthy_signals_continue
+      runner: default
+      inputs:
+        evidence:
+          postmaster_report:
+            source: google-postmaster
+            timestamp: "2026-07-01T12:00:00Z"
+            reputation_score: 0.94
+          bounce_metrics:
+            source: esp-bounce-export
+            timestamp: "2026-07-01T12:00:00Z"
+            bounce_pct: 0.7
+          complaint_metrics:
+            source: feedback-loop
+            timestamp: "2026-07-01T12:00:00Z"
+            complaint_pct: 0.02
+          placement_probe:
+            source: seed-inbox-probe
+            timestamp: "2026-07-01T12:00:00Z"
+            passed: true
+        policy:
+          min_reputation_score: 0.8
+          max_bounce_pct: 2
+          max_complaint_pct: 0.1
+      expect:
+        status: sealed
+    - name: contradictory_signals_escalate
+      runner: default
+      inputs:
+        evidence:
+          postmaster_report:
+            source: google-postmaster
+            timestamp: "2026-07-01T13:00:00Z"
+            reputation_score: 0.96
+          bounce_metrics:
+            source: esp-bounce-export
+            timestamp: "2026-07-01T13:00:00Z"
+            bounce_pct: 9.4
+          complaint_metrics:
+            source: feedback-loop
+            timestamp: "2026-07-01T13:00:00Z"
+            complaint_pct: 0.03
+          placement_probe:
+            source: seed-inbox-probe
+            timestamp: "2026-07-01T13:00:00Z"
+            passed: true
+        policy:
+          min_reputation_score: 0.8
+          max_bounce_pct: 2
+          max_complaint_pct: 0.1
+      expect:
+        status: sealed
+
+runners:
+  default:
+    default: true
+    type: cli-tool
+    command: node
+    args:
+      - run.mjs
+    sandbox:
+      profile: readonly
+      cwd_policy: skill-directory
+    inputs:
+      evidence:
+        type: object
+        required: true
+        description: Sealed deliverability evidence.
+      policy:
+        type: object
+        required: true
+        description: Deliverability policy thresholds.

skills/deliverability-judge/evidence.json

@@ -0,0 +1,57 @@
+{
+  "summary": "deliverability-judge 0.1.0 is a read-only runx skill that fuses sealed deliverability evidence into a continue, throttle, pause, or escalation verdict.",
+  "observations": [
+    {
+      "name": "runx_version",
+      "value": "runx-cli 0.6.14",
+      "meets_requirement": true
+    },
+    {
+      "name": "package_name",
+      "value": "deliverability-judge",
+      "meets_requirement": true
+    },
+    {
+      "name": "github_star",
+      "value": "pay4love1 currently stars https://github.com/runxhq/runx, verified through the GitHub API before submission preparation.",
+      "meets_requirement": true
+    },
+    {
+      "name": "source_pr",
+      "value": "https://github.com/runxhq/runx/pull/204",
+      "meets_requirement": true
+    },
+    {
+      "name": "local_direct_runner_healthy_case",
+      "value": "node run.mjs returned verdict.state=healthy and recommendation.action=continue for sealed healthy signals.",
+      "meets_requirement": true
+    },
+    {
+      "name": "local_direct_runner_contradictory_case",
+      "value": "node run.mjs returned verdict.state=escalate and no recommendation for contradictory sealed signals.",
+      "meets_requirement": true
+    },
+    {
+      "name": "windows_runx_harness_blocker",
+      "value": "runx harness and registry publish were attempted locally on Windows, but the CLI failed in receipt/sandbox runtime paths before skill logic execution. Linux hosted harness/publish is still required for final delivery.",
+      "meets_requirement": false
+    },
+    {
+      "name": "runx_connect_auth_blocker",
+      "value": "runx login --provider github --for publish generated connect.runx.ai sessions, but the browser flow returned 404 Not Found after GitHub sign-in. The same result was reproduced by the operator manually, so registry publish could not be completed from this environment.",
+      "meets_requirement": false
+    }
+  ],
+  "commands": [
+    "runx --version",
+    "node run.mjs",
+    "runx login --provider github --for publish",
+    "runx registry publish ./skills/deliverability-judge/SKILL.md --registry https://api.runx.ai"
+  ],
+  "artifacts": {
+    "source_url": "https://github.com/runxhq/runx/pull/204",
+    "pr_url": "https://github.com/runxhq/runx/pull/204",
+    "local_output": "skills/deliverability-judge/harness-evidence/local-output.md"
+  },
+  "version": "0.1.0"
+}

skills/deliverability-judge/fixtures/contradictory_signals_escalate.json

@@ -0,0 +1,29 @@
+{
+  "evidence": {
+    "postmaster_report": {
+      "source": "google-postmaster",
+      "timestamp": "2026-07-01T13:00:00Z",
+      "reputation_score": 0.96
+    },
+    "bounce_metrics": {
+      "source": "esp-bounce-export",
+      "timestamp": "2026-07-01T13:00:00Z",
+      "bounce_pct": 9.4
+    },
+    "complaint_metrics": {
+      "source": "feedback-loop",
+      "timestamp": "2026-07-01T13:00:00Z",
+      "complaint_pct": 0.03
+    },
+    "placement_probe": {
+      "source": "seed-inbox-probe",
+      "timestamp": "2026-07-01T13:00:00Z",
+      "passed": true
+    }
+  },
+  "policy": {
+    "min_reputation_score": 0.8,
+    "max_bounce_pct": 2,
+    "max_complaint_pct": 0.1
+  }
+}

skills/deliverability-judge/fixtures/sealed_healthy_signals_continue.json

@@ -0,0 +1,29 @@
+{
+  "evidence": {
+    "postmaster_report": {
+      "source": "google-postmaster",
+      "timestamp": "2026-07-01T12:00:00Z",
+      "reputation_score": 0.94
+    },
+    "bounce_metrics": {
+      "source": "esp-bounce-export",
+      "timestamp": "2026-07-01T12:00:00Z",
+      "bounce_pct": 0.7
+    },
+    "complaint_metrics": {
+      "source": "feedback-loop",
+      "timestamp": "2026-07-01T12:00:00Z",
+      "complaint_pct": 0.02
+    },
+    "placement_probe": {
+      "source": "seed-inbox-probe",
+      "timestamp": "2026-07-01T12:00:00Z",
+      "passed": true
+    }
+  },
+  "policy": {
+    "min_reputation_score": 0.8,
+    "max_bounce_pct": 2,
+    "max_complaint_pct": 0.1
+  }
+}

skills/deliverability-judge/harness-evidence/local-output.md

@@ -0,0 +1,66 @@
+# Local Evidence
+
+Toolchain:
+
+```text
+runx-cli 0.6.14
+node v22.17.1
+```
+
+Direct runner verification was executed against the same inputs used by the
+inline harness cases.
+
+## sealed_healthy_signals_continue
+
+Expected: sealed evidence fuses into a healthy verdict and read-only
+`continue` recommendation.
+
+Observed:
+
+```json
+{
+  "verdict": {
+    "state": "healthy",
+    "confidence_window": "high",
+    "reason": "all_sealed_signals_pass_policy"
+  },
+  "recommendation": {
+    "action": "continue",
+    "signal_bindings": {
+      "postmaster_report": "google-postmaster",
+      "bounce_metrics": "esp-bounce-export",
+      "complaint_metrics": "feedback-loop",
+      "placement_probe": "seed-inbox-probe"
+    },
+    "evidence_hash": "de4d14b9583c678f096fecbc0a147a20b9f86ca17b8d297420ba9365bf718c5a"
+  }
+}
+```
+
+## contradictory_signals_escalate
+
+Expected: contradictory sealed signals refuse a recommendation and escalate.
+
+Observed:
+
+```json
+{
+  "verdict": {
+    "state": "escalate",
+    "confidence_window": "low",
+    "reason": "contradictory_signals"
+  },
+  "escalation": {
+    "reason": "Signals disagree, so no read-only recommendation is emitted.",
+    "contradictions": [
+      "high_reputation_conflicts_with_high_bounce"
+    ],
+    "signal_names": [
+      "postmaster_report",
+      "bounce_metrics",
+      "complaint_metrics",
+      "placement_probe"
+    ]
+  }
+}
+```

skills/deliverability-judge/report.md

@@ -0,0 +1,13 @@
+# deliverability-judge 0.1.0 report
+
+- Package: `pay4love1/deliverability-judge@0.1.0`
+- Pull request: https://github.com/runxhq/runx/pull/204
+- Bounty: https://gofrantic.com/bounties/65
+- runx CLI: `runx-cli 0.6.14`
+- Purpose: read sealed provider evidence for postmaster reputation, bounce rate, complaint rate, and placement probe status, then emit a read-only deliverability verdict.
+- Healthy sealed evidence returns `verdict.state=healthy` with `recommendation.action=continue`.
+- Contradictory sealed evidence returns `verdict.state=escalate` and refuses to emit a recommendation.
+- The implementation is read-only: no send, throttle, payment, state write, Effect, or operational handoff is performed.
+- Local direct runner evidence is included in `harness-evidence/local-output.md`.
+- Current blocker: `runx login --provider github --for publish` and manual `runx connect github` both reached the browser/GitHub sign-in flow, then returned 404 Not Found after sign-in.
+- Remaining finalization: recover runx publish auth, complete registry publish, hosted harness, clean install, dogfood run, and receipt verification.