rickylabs · rickylabs · Aug 13, 2026 · Aug 13, 2026 · Aug 13, 2026 · Aug 13, 2026
@@ -0,0 +1,43 @@
+use harness
+
+# #1644 acceptance-guidance completion
+
+Resume the existing `harness-evidence-and-verdict-tooling` implementer in the same Codex thread.
+The coordinator interrupted the packaging turn before any commit because final reconciliation found
+one exact acceptance gap. Preserve and inspect the current uncommitted packaging artifacts; do not
+discard valid work blindly.
+
+## SKILL
+
+Re-read and follow:
+
+- `.agents/skills/netscript-harness/SKILL.md`
+- `.agents/skills/netscript-tools/SKILL.md`
+- `.agents/skills/netscript-pr/SKILL.md`
+- `.agents/skills/netscript-deno-toolchain/SKILL.md`
+
+Coordinator control head `33626b1f4` amends the binding leaf contract. Exactly one additional edit
+surface is now authorized: `.agents/skills/netscript-pr/SKILL.md`. Live #1621 explicitly requires
+the skill to state plainly that only markdown checkboxes are close-gated and that a plain-bullet
+Acceptance section has no mirrorable targets and therefore takes no `acceptance-evidence` block.
+Add that concise operator guidance in the existing machine-convention section. Do not edit any
+other new product, workflow, tool, or test surface.
+
+Record the coordinator amendment and supersession of the earlier read-only decision in leaf
+drift/worklog/context. Update draft PR #1644 so S3 and its matching Definition-of-Done row become
+truthful only after the guidance commit exists. Preserve the implementation and Tier-A results.
+
+Treat the already-generated final receipts at `b21424c44` as pre-guidance evidence, not the final
+immutable set. Commit the acceptance-complete guidance and run-artifact updates first. Then run
+exactly one final structured `check`, `test`, and `quality-job` cycle at that acceptance-complete
+implementation head. Package those receipts in one evidence-only commit and push by explicit
+refspec. It is expected that receipt `actualGitHead` identifies the validated parent implementation
+head while the final branch tip is the evidence-only packaging commit; record that relationship
+explicitly. Do not create an infinite receipt/self-reference loop and do not rerun the gates merely
+because the receipt-only commit changes HEAD.
+
+Stop at the formal IMPL-EVAL handoff. Do not launch Claude, Fable, OpenRouter, DeepSeek, Minimax, or
+any substitute evaluator. The formal opposite-family gate is held until the native allowance reset
+at Saturday 2026-08-15 00:00 Europe/Zurich. Do not merge, publish, mark ready, or alter central
+cluster state. Return the exact implementation head, evidence head, receipt outcomes/durations,
+remote reconciliation, and `BLOCKED: awaiting native opposite-family IMPL-EVAL after reset`.
@@ -0,0 +1,31 @@
+# Native reset-gate dispatch
+
+This directory is the coordinator-owned source of truth for the six formal gates parked until
+Saturday 2026-08-15 00:00 Europe/Zurich. `dispatch.json` fixes the order, exact remote source heads,
+worktrees, routes, and output artifacts.
+
+Rules:
+
+- Do not launch any entry before `launchAfter`.
+- The Codex milestone coordinator remains the only active coordinator before the reset.
+- Dispatch exactly one evaluator at a time and reconcile its result before starting the next.
+- Use the exact per-entry native Claude Opus 5 route in `dispatch.json`: low effort for bounded
+  genuinely easy review and medium for the retained substantive gates.
+- Do not launch Fable mechanically. Fable 5 is available only after the coordinator records a
+  genuinely architectural PLAN question or exceptional complex implementation-review need and amends
+  the entry. OpenRouter, DeepSeek, Minimax, AGY, and silent substitutions remain unauthorized.
+- Start `/remote-control` in every native evaluator session and record its session, bridge, PID, and
+  cwd in the verdict artifact so the owner can see the lane.
+- Resolve the PR and remote branch independently. A mismatch with `sourceHead` is a hard refusal,
+  not permission to evaluate a nearby commit.
+- Evaluators write verdict evidence only. They do not implement, merge, publish, change labels,
+  update central state, or take an expensive-gate lease.
+- The coordinator verifies the verdict commit, updates the leaf head/phase, renders status, and
+  validates cluster state before any next transition.
+
+The order deliberately clears the harness-tooling IMPL-EVAL first, then the second ready-to-merge
+fix, before opening four plan-gated implementation queues. Each retained gate has a specific
+existing hold, invalid/advisory prior gate, or complexity record; this is not a mechanical PLAN-EVAL
+quota for later work. All six leaves are Wave 0 and the central dependency DAG contains no edge
+between them. Broad source/CI/hold reconciliation is shared in the coordinator record, but harness
+law still requires one fresh evaluator session per retained formal gate.
@@ -0,0 +1,30 @@
+use harness
+
+## SKILL
+
+Read `AGENTS.md`, the full `netscript-harness`, `netscript-tools`, `netscript-pr`,
+`netscript-doctrine`, `netscript-deno-toolchain`, and docs-relevant project skills. Then read the
+PLAN-EVAL protocol, plan gate, verdict definitions, central leaf contract, complete run, live issue
+#1551, PR #1652, and coordinator scope comments.
+
+Act only as fresh formal PLAN-EVAL cycle 1. The reset boundary has passed. Use the owner-authorized
+native Claude Opus 5 **low** route and immediately enable `/remote-control`; record session ID,
+bridge ID, PID, cwd, requested route, and observed route. No silent substitute; Fable 5 remains
+unassigned and requires a coordinator amendment recording genuine architectural or exceptional
+implementation-review necessity.
+
+Worktree `/home/codex/repos/netscript-007-docs-comparison`; branch `docs/comparison-docs-programme`;
+immutable base `01e0960494c95ce56eb35892c211a095eb13e6ed`; exact remote PR/source head
+`d35cbca30872d1f55118d63437638e93270c2ac3`. Fetch and independently resolve PR and branch; refuse
+any mismatch.
+
+Independently judge all Plan-Gate rows, the 0.0.7-sized slice of the 17-deliverable programme,
+methodology/navigation scope, immutable evidence manifest and measurement procedure, deferred case
+study/migration roadmap, source/claim freshness, terminology, locked decisions, risks, explicit
+deferrals, and exact docs/check/test gate set. Check that evidence paths can prove claims without
+turning future comparisons into untracked scope. Do not author docs or implement.
+
+Write `plan-eval.md` with exactly `PASS` or `FAIL_PLAN`. Commit only it, push explicitly to the
+existing branch, and post one structured PR comment with verdict, evaluated head, evaluator commit,
+and remote-control identity. Keep PR #1652 draft at `status:plan-eval`; do not implement, merge,
+publish, relabel, or mutate central state.
@@ -0,0 +1,151 @@
+{
+  "schemaVersion": 1,
+  "milestone": "0.0.7",
+  "launchAfter": "2026-08-15T00:00:00+02:00",
+  "timezone": "Europe/Zurich",
+  "concurrency": 4,
+  "concurrencyScope": "per-topic-orchestrator",
+  "perOrchestratorConcurrency": 1,
+  "ownerOverride": {
+    "recordedAt": "2026-08-14T22:41:15.000Z",
+    "coordinator": "codex-root-0.0.7",
+    "topicOrchestrators": "native Claude Opus 5 / high with Remote Control",
+    "implementationLeaves": "daemon-attached WSL Codex GPT-5.6-SOL with effort matched to documented complexity",
+    "evaluationDefault": "native Claude Opus 5 / low through high; low only for genuinely easy bounded review",
+    "planEvalPolicy": "open only for an existing formal hold, unresolved architectural decision, or demonstrated complexity",
+    "fablePolicy": "reserve Fable 5 for recorded genuinely architectural PLAN questions or exceptional complex implementation review",
+    "costPolicy": "right-size effort, deduplicate shared review inputs, preserve every required independent evidence contract",
+    "queuePolicy": "serialize within each topic orchestrator; docs, internals, fixes, and features may each run one evaluator concurrently",
+    "supersedes": "the 2026-08-14T22:13:00.000Z Sonnet route matrix and the pre-reset blanket Fable matrix"
+  },
+  "routePolicy": {
+    "provider": "native-claude",
+    "remoteControlRequired": true,
+    "freshSessionPerGate": true,
+    "oppositeFamilyRequired": true,
+    "substitutesAllowed": false,
+    "fablePolicy": "not pre-dispatched; coordinator amendment must record genuine architectural or exceptional implementation-review necessity",
+    "forbiddenProviders": ["openrouter", "opencode", "antigravity"]
+  },
+  "entries": [
+    {
+      "order": 1,
+      "leafId": "harness-evidence-and-verdict-tooling",
+      "phase": "impl-eval",
+      "pr": 1644,
+      "branch": "fix/harness-evidence-and-verdict-tooling",
+      "worktree": "/home/codex/repos/netscript-007-harness-evidence",
+      "sourceHead": "4d9fb196765cbf1a6bc7eaa7c18ec82b237ab89f",
+      "implementationParent": "634b257ea1afcedb2d7f1da486d8c9e9432a2a86",
+      "runDir": ".llm/runs/release-0.0.7-internals--orchestration/slices/harness-evidence-and-verdict-tooling",
+      "brief": "harness-evidence-and-verdict-tooling.md",
+      "output": "evaluate.md",
+      "route": {
+        "provider": "native-claude",
+        "model": "Claude Opus 5",
+        "cliModel": "claude-opus-5",
+        "effort": "medium",
+        "rationale": "substantive cross-tooling implementation review; Opus medium matches its documented complexity"
+      }
+    },
+    {
+      "order": 2,
+      "leafId": "legacy-port-pin-sweep",
+      "phase": "impl-eval",
+      "pr": 1643,
+      "branch": "fix/legacy-port-pin-sweep",
+      "worktree": "/home/codex/repos/netscript-007-leaf-legacy-port-pin-sweep",
+      "sourceHead": "e6ba15ec6414c0a42b1f9870791131162ea71c36",
+      "runDir": ".llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0",
+      "brief": "legacy-port-pin-sweep.md",
+      "output": "evaluate.md",
+      "route": {
+        "provider": "native-claude",
+        "model": "Claude Opus 5",
+        "cliModel": "claude-opus-5",
+        "effort": "low",
+        "rationale": "genuinely easy bounded implementation review with complete Tier-A and gate evidence; Opus low is sufficient"
+      }
+    },
+    {
+      "order": 3,
+      "leafId": "rfc-plugin-cli-contribution",
+      "phase": "plan-eval",
+      "cycle": 2,
+      "pr": 1651,
+      "branch": "docs/rfc-plugin-cli-contribution",
+      "worktree": "/home/codex/repos/netscript-007-features-1502",
+      "sourceHead": "12276e6d86403ed1340ef79a963e87d401d643e9",
+      "runDir": ".llm/runs/docs-rfc-plugin-cli-contribution--1502",
+      "brief": "rfc-plugin-cli-contribution.md",
+      "output": "plan-eval.md",
+      "route": {
+        "provider": "native-claude",
+        "model": "Claude Opus 5",
+        "cliModel": "claude-opus-5",
+        "effort": "medium",
+        "rationale": "existing formal architectural hold requires substantive independent reasoning; Opus medium is sufficient for this evidence contract"
+      }
+    },
+    {
+      "order": 4,
+      "leafId": "quality-scan-allowance-rail",
+      "phase": "plan-eval",
+      "cycle": 2,
+      "pr": 1653,
+      "branch": "chore/quality-scan-allowance-rail",
+      "worktree": "/home/codex/repos/netscript-007-quality-rail",
+      "sourceHead": "09dfb092dccf7f843b9270295047d674a8187362",
+      "runDir": ".llm/runs/release-0.0.7-internals--orchestration/slices/quality-scan-allowance-rail",
+      "brief": "quality-scan-allowance-rail.md",
+      "output": "plan-eval.md",
+      "route": {
+        "provider": "native-claude",
+        "model": "Claude Opus 5",
+        "cliModel": "claude-opus-5",
+        "effort": "medium",
+        "rationale": "substantive coupled ownership and fail-closed quality-policy PLAN-EVAL"
+      }
+    },
+    {
+      "order": 5,
+      "leafId": "scaffold-generated-output-correctness",
+      "phase": "plan-eval",
+      "cycle": 2,
+      "pr": 1654,
+      "branch": "fix/scaffold-generated-output-correctness",
+      "worktree": "/home/codex/repos/netscript-007-leaf-scaffold-generated-output-correctness",
+      "sourceHead": "5b3c6fcf21b0b4947a770d8e67ea5cc8082724d5",
+      "runDir": ".llm/runs/fix-scaffold-generated-output-correctness--0.0.7-wave0",
+      "brief": "scaffold-generated-output-correctness.md",
+      "output": "plan-eval.md",
+      "route": {
+        "provider": "native-claude",
+        "model": "Claude Opus 5",
+        "cliModel": "claude-opus-5",
+        "effort": "medium",
+        "rationale": "substantive grouped generator/runtime boundary PLAN-EVAL without running the expensive gate"
+      }
+    },
+    {
+      "order": 6,
+      "leafId": "comparison-docs-programme",
+      "phase": "plan-eval",
+      "cycle": 1,
+      "pr": 1652,
+      "branch": "docs/comparison-docs-programme",
+      "worktree": "/home/codex/repos/netscript-007-docs-comparison",
+      "sourceHead": "d35cbca30872d1f55118d63437638e93270c2ac3",
+      "runDir": ".llm/runs/docs-comparison-docs-programme--1551",
+      "brief": "comparison-docs-programme.md",
+      "output": "plan-eval.md",
+      "route": {
+        "provider": "native-claude",
+        "model": "Claude Opus 5",
+        "cliModel": "claude-opus-5",
+        "effort": "low",
+        "rationale": "genuinely easy bounded docs-only PLAN-EVAL with immutable evidence and no product implementation; Opus low is sufficient"
+      }
+    }
+  ]
+}
@@ -0,0 +1,36 @@
+use harness
+
+## SKILL
+
+Read `AGENTS.md`, `.agents/skills/netscript-harness/SKILL.md`,
+`.agents/skills/netscript-tools/SKILL.md`, `.agents/skills/netscript-pr/SKILL.md`,
+`.agents/skills/openhands-handoff/SKILL.md`, and `.agents/skills/netscript-deno-toolchain/SKILL.md`
+completely. Then read the harness evaluator protocol, verdict definitions, lane policy,
+agent-handoff rules, `6-cli-tooling` profile, the whole leaf run directory, live issues
+#1561/#1563/#1621, PR #1644, and its structured comments.
+
+Act only as the fresh separate formal IMPL-EVAL. The reset boundary has passed. Use the
+owner-authorized native Claude Opus 5 **medium** route and immediately enable `/remote-control`;
+record session ID, bridge ID, PID, cwd, requested route, and observed route. No silent substitute;
+Fable 5 remains unassigned and requires a coordinator amendment recording genuine architectural or
+exceptional implementation-review necessity.
+
+Repository `rickylabs/netscript`; worktree `/home/codex/repos/netscript-007-harness-evidence`;
+branch `fix/harness-evidence-and-verdict-tooling`; immutable base
+`01e0960494c95ce56eb35892c211a095eb13e6ed`; exact remote PR/source head
+`4d9fb196765cbf1a6bc7eaa7c18ec82b237ab89f`; acceptance-complete implementation parent
+`634b257ea1afcedb2d7f1da486d8c9e9432a2a86`. Fetch and independently resolve the PR and branch.
+Refuse if either source head differs. Receipts intentionally attest the implementation parent;
+verify both SHAs and do not demand evidence-child self-reference.
+
+Evaluate the complete authorized diff and all evidence independently, including empty evidence,
+block attribution, zero-checkbox ordering/guidance, dry-run failure, exact wrapped verdict tokens,
+fenced/template exclusions, provenance states, workflow matchers, lock hygiene, and markdown-only
+close-gate semantics. Verify the only binding gates are `check`, `test`, and `quality-job`. Run only
+small independent checks needed for confidence. Do not implement or take an expensive gate.
+
+Write the run's `evaluate.md` from the harness template with exactly one verdict: `PASS`,
+`FAIL_FIX`, `FAIL_RESCOPE`, or `FAIL_DEBT`. Commit only that artifact and push explicitly to the
+existing branch. Post one structured PR comment with phase, verdict, evaluated source and parent,
+evaluator commit, and remote-control identity. Keep the PR draft at `status:impl`; do not merge,
+publish, relabel, or mutate central state.
@@ -0,0 +1,32 @@
+use harness
+
+## SKILL
+
+Read `AGENTS.md` and the full `netscript-harness`, `netscript-tools`, `netscript-pr`,
+`netscript-cli`, `netscript-doctrine`, `netscript-deno-toolchain`, `jsr-audit`, `aspire`, and `rtk`
+project skills. Then read the IMPL-EVAL protocol/verdict definitions, complete leaf run, live issue
+#1243, PR #1643, coordinator scope comments, full diff, Tier-A review, and receipts.
+
+Act only as the fresh separate formal IMPL-EVAL. The reset boundary has passed. Use the
+owner-authorized native Claude Opus 5 **low** route and immediately enable `/remote-control`; record
+session ID, bridge ID, PID, cwd, requested route, and observed route. No silent substitute; Fable 5
+remains unassigned and requires a coordinator amendment recording genuine architectural or
+exceptional implementation-review necessity.
+
+Worktree `/home/codex/repos/netscript-007-leaf-legacy-port-pin-sweep`; branch
+`fix/legacy-port-pin-sweep`; immutable base `01e0960494c95ce56eb35892c211a095eb13e6ed`; exact remote
+PR/source head `e6ba15ec6414c0a42b1f9870791131162ea71c36`. Fetch and resolve both PR and branch
+independently; refuse any mismatch. This exact head supersedes the stale `af3dca0f5` value in the
+older leaf-local `evaluate-prompt.md`.
+
+Independently verify the auth command has no silent localhost:4437 default, requires explicit
+`--stream-url`, fails before the session adapter when omitted, and gives actionable Aspire endpoint
+discovery guidance. Verify manifest/copy port fields remain coordinator-classified compatibility
+metadata, broad formatting remains mechanically isolated, lock/JSR/publish evidence is honest, and
+there is no hidden behavioral or scope drift. Run only the smallest non-expensive checks needed; do
+not run scaffold.runtime, Aspire, Docker, or publish and do not implement.
+
+Write `evaluate.md` with exactly one harness verdict: `PASS`, `FAIL_FIX`, `FAIL_RESCOPE`, or
+`FAIL_DEBT`. Commit only it and push explicitly to the existing branch. Post one structured PR
+comment with phase, verdict, evaluated head, evaluator commit, and remote-control identity. Keep PR
+#1643 draft at `status:impl`; do not merge, publish, relabel, or mutate central state.
@@ -0,0 +1,31 @@
+use harness
+
+## SKILL
+
+Read `AGENTS.md`, the full `netscript-harness`, `netscript-tools`, `netscript-pr`,
+`netscript-deno-toolchain`, `netscript-doctrine`, `jsr-audit`, and relevant quality-scan skills.
+Then read the PLAN-EVAL protocol, plan gate, verdict definitions, central leaf contract, complete
+run, issues #1378/#1545, PR #1653, issue #1655, and coordinator resolution comments.
+
+Act only as fresh formal PLAN-EVAL cycle 2. The reset boundary has passed. Use the owner-authorized
+native Claude Opus 5 **medium** route and immediately enable `/remote-control`; record session ID,
+bridge ID, PID, cwd, requested route, and observed route. No silent substitute; Fable 5 remains
+unassigned and requires a coordinator amendment recording genuine architectural or exceptional
+implementation-review necessity.
+
+Worktree `/home/codex/repos/netscript-007-quality-rail`; branch `chore/quality-scan-allowance-rail`;
+immutable base `01e0960494c95ce56eb35892c211a095eb13e6ed`; exact remote PR/source head
+`09dfb092dccf7f843b9270295047d674a8187362`. Fetch and independently resolve PR and branch; refuse
+any mismatch. Evaluate the repaired head, not advisory head `c573beda...`. Preserve the existing
+advisory `plan-eval.md` as `plan-eval-cycle-1-advisory.md`, then write the new canonical
+`plan-eval.md`.
+
+Independently judge all Plan-Gate rows and specifically verify the seven-record ownership decision,
+exact authorized scanner-test/consumer/generated-asset/debt surfaces, strict Workers no-increase
+baseline owned by #1655, measured budget, issue prose, RED-first proof, generated-asset freshness,
+full JSR audit plan, and honest gate map. Do not implement.
+
+The new artifact must contain exactly `PASS` or `FAIL_PLAN`. Commit only the history rename and new
+verdict artifact, push explicitly to the existing branch, and post one structured PR comment with
+cycle, verdict, evaluated head, evaluator commit, and remote-control identity. Keep PR #1653 draft
+at `status:plan-eval`; do not implement, merge, publish, relabel, or mutate central state.