[codex] Branch 9N OpenCode result report model by Ian747-tw · Pull Request #103 · Ian747-tw/NexusLoop

Ian747-tw · 2026-07-07T19:57:23Z

Summary

Adds durable OpenCode result-report records for one launched OpenCode session.
Supports preview, dry-run, record, list, show, latest, and summary surfaces across runtime, TUI, fake runtime, and E2E.
Records executor result evidence only; it does not accept results, complete missions, ingest research, create checkpoints, or prompt/control OpenCode.

Result report model

Adds bounded result kinds: completion_report, partial_report, failure_report, inconclusive_report, blocked_report, status_report, and unknown.
Maps report kinds to report-only dispositions such as reported_done, reported_partial, and reported_failed.
Stores changed files, tests, test results, artifacts, metrics, claims, known failures, follow-ups, confidence, and linked evidence IDs as bounded previews.

Result disposition/review-state behavior

Completion, partial, failure, inconclusive, and blocked reports require Commander or human review.
status_report defaults to review_not_required, unless an explicit review queue is requested.
Accepted/rejected review states are blocked because result acceptance belongs to a future branch.

Launch/progress/evidence anchoring behavior

Preview/record requires a resolvable planned session and 9D launch in launch_started or launched state.
Optional 9E/9F/9G/9H/9I/9L/9M links are validated against the same session/launch chain.
Progress completion/failure evidence can provide bounded defaults where appropriate.

Preview/dry-run/record/list/latest/summary behavior

Preview/list/get/latest/summary are read-like and do not auto-start runtime/OpenCode.
Dry-run uses the same validation path and appends no events.
Non-dry record appends one bounded opencode_result_report_recorded event through the metadata write path.

Duplicate behavior

Multiple distinct result reports are allowed over time.
Exact duplicate report hashes for the same session/launch are blocked, including dry-run.
Duplicate detection scans durable records beyond the public list cap.

Event persistence/redaction behavior

Events are persisted to .nxl/events.jsonl with bounded/redacted payloads.
Payloads exclude raw logs, full diffs, file contents, raw OpenCode output, provider output, full event logs, research.db dumps, prompt transcripts, checkpoint payloads, and mission mutation payloads.

Fake-vs-real runtime behavior

Fake runtime supports deterministic preview, dry-run, record, duplicate handling, list/show/latest/summary, and fake linked progress defaults.
Neither fake nor real result-report paths launch OpenCode, prompt OpenCode, call providers, write research.db, or mutate missions.

Runtime commands/API surface

runtime.preview_opencode_result_report
runtime.record_opencode_result_report
runtime.list_opencode_result_reports
runtime.get_opencode_result_report
runtime.latest_opencode_result_report
runtime.opencode_result_report_summary

TUI result-report surface

Adds /opencode-result-report-preview, /opencode-result-report-dry-run, /opencode-result-report, /opencode-result-reports, /opencode-result-report-latest, /opencode-result-report-show, and /opencode-result-report-summary.
Adds aliases /result-report-preview, /result-report-dry-run, /result-report, /result-reports, /result-report-latest, and /result-report-summary.
Renders the OpenCode result reports section with evidence links, review state, false side-effect flags, and explicit out-of-scope notes.

No-mission-completion guarantee

Result reports are executor evidence only.
mission_mutated=false is persisted and rendered.
No mission/session completion, success, failure, proposal, review, or apply mutation is performed.

No-researchdb-ingestion guarantee

research_db_written=false and checkpoint_created=false are persisted and rendered.
No research findings are ingested or promoted.
No checkpoint or result-review acceptance event is created.

No-provider/no-OpenCode-prompt/no-process-control guarantees

Result-report commands do not call MiniMax or any provider.
They do not call MCPs or online research.
They do not send prompts to OpenCode, launch additional sessions, or pause/resume/stop/kill processes.

Authority registry update

Adds safe-read authority records for preview/dry-run/list/latest/show/summary commands.
Adds metadata-write authority for /opencode-result-report and /result-report.
Authority notes state result reports are metadata only and do not complete missions, ingest research, call providers, prompt OpenCode, control processes, mutate mission/proposal/review/apply records, or create checkpoints.

Targeted validation output

cd agentcore/runtime && bun test && bun run typecheck
624 pass
0 fail
4835 expect() calls
Ran 624 tests across 2 files. [45.50s]
$ tsc --noEmit

cd agentcore/tui && bun test && bun run typecheck
274 pass
0 fail
3629 expect() calls
Ran 274 tests across 7 files. [5.42s]
$ tsc --noEmit

uv run pytest tests/integration/cli/test_tui_launcher.py tests/integration/cli/test_tui_entrypoint.py -q
.......                                                                  [100%]
7 passed in 1.24s

uv run pytest tests/e2e_user/scenarios/test_opencode_result_report_tui.py -q
.                                                                        [100%]
1 passed in 78.18s (0:01:18)

uv run pytest tests/e2e_user/scenarios/test_command_authority_inventory_tui.py -q
.                                                                        [100%]
1 passed in 111.85s (0:01:51)

uv run pytest tests/e2e_user/scenarios/test_opencode_wake_action_execution_tui.py tests/e2e_user/scenarios/test_opencode_progress_heartbeat_tui.py tests/e2e_user/scenarios/test_opencode_wake_supervisor_execution_tui.py -q
...                                                                      [100%]
3 passed in 194.62s (0:03:14)

Full regression status

Full historical E2E was not run; this branch did not broadly change runtime-client lifecycle, adapter startup/shutdown, process management, shared event-store behavior, research backend storage/projection behavior, or scheduler/wake execution beyond the result-report metadata surface.
OpenCode adapter smoke was not run because adapter startup/shutdown and live process-control code were not touched.

Secret/redaction behavior

Secret-looking report, outcome, metric, artifact, and evidence text is redacted before state, snapshot, and event persistence.
Raw logs, full diffs, file contents, provider output, raw OpenCode output, full event logs, and research.db dumps are blocked.

Explicit out-of-scope list

Commander result review, result acceptance/rejection, research.db writes/ingestion, research promotion, mission/proposal/review/apply mutation, checkpoint creation, provider calls, MCP/online research, OpenCode prompts, process control, automatic wake action execution, raw diff parsing, artifact upload/storage, result approval gates, and follow-up mission creation.

Notes about Branch 9O Commander result review gate

9N intentionally leaves every completion/partial/failure/inconclusive/blocked report in a review-needed state.
9O should add the explicit Commander/human result review gate that can inspect these reports and decide acceptance/rejection without overloading the result-report metadata event.

Ian747-tw · 2026-07-07T19:57:29Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f050dd38ec

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T20:03:06Z

+  }
+
+  private async findDuplicate(reportHash: string, sessionId: string, launchId?: string): Promise<OpenCodeResultReportRecord | undefined> {
+    return (await this.list({ session_id: sessionId, launch_id: launchId, limit: MAX_LIST })).find((record) => record.report_hash === reportHash)


Scan all prior reports when deduplicating

When a session/launch has more than 100 result reports, this duplicate check only searches the capped public list, so an identical older report outside the latest 100 records is no longer found and can be appended again. Since result reports are durable executor evidence for later Commander review, duplicate prevention should scan the uncapped event projection for the matching session/launch hash rather than reusing the display limit.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-07-07T20:03:06Z

+    wake_execution_id: optional(input.wakeExecutionId ?? input.wake_execution_id ?? input.wake_execution),
+    wake_action_execution_id: optional(input.wakeActionExecutionId ?? input.wake_action_execution_id ?? input.wake_action),


Honor advertised wake aliases in report input

The runtime client overload advertises wakeExecution and wakeAction as accepted payload aliases, but these reader lines only consume the *Id, snake_case, and lower snake aliases. API callers that supply only wakeExecution or wakeAction therefore get their linked evidence dropped and a preview/record can be blocked as missing session/launch even though the referenced wake record exists.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T20:13:22Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 09210bd7d2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T20:18:27Z

+
+    const resultKind = readResultKind(input.result_kind, evidence.progress?.kind)
+    const disposition = dispositionForKind(resultKind)
+    const reviewState = readReviewState(input.review_state, defaultReviewState(resultKind), blockers)


Require Commander review for terminal reports

When an operator supplies reviewState=review_not_required with kind=completion_report (or partial/failure/inconclusive/blocked), this call accepts the override and can_record can still remain true. Those non-status result reports are supposed to be executor evidence awaiting Commander review, so persisting them as review_not_required lets later review consumers miss exactly the terminal reports they need to adjudicate.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-07-07T20:18:27Z

+    } else throw new Error("OpenCode result report arg is unsupported")
+  }
+  if (!effect.sessionId && !effect.launchId && !effect.progressId && !effect.watchdogId && !effect.questionId && !effect.guidanceId && !effect.deliveryId && !effect.wakeExecutionId && !effect.wakeActionExecutionId) throw new Error("OpenCode result report requires session=<id>, launch=<id>, or linked evidence")
+  if (requireSummary && !effect.summary) throw new Error("OpenCode result report requires summary=<text>")


Allow progress-linked reports to use progress summaries

For /opencode-result-report and dry-run commands with only linked progress evidence, this parser throws before the runtime can apply its bounded defaults from the progress record. A user with a completion/failure progress record cannot run /opencode-result-report progress=<id> even though the runtime model supports linked progress supplying omitted report fields, so the TUI/headless surface does not actually mirror the runtime API for that advertised flow.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T20:29:21Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e8163d5825

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T20:34:57Z

+    const reviewState = readReviewState(input.review_state, defaultReviewState(resultKind), blockers)
+    if (resultKind !== "status_report" && reviewState === "review_not_required") blockers.push("terminal OpenCode result reports require Commander or human review")


Keep terminal reports in review queues

When a caller supplies reviewState=unknown or reviewState=not_ready_for_review for a non-status report, readReviewState accepts it and this guard only rejects review_not_required, so a completion/failure/partial report can be recorded without needs_commander_review or needs_human_review. That breaks the terminal-report disposition model and causes opencode_result_report_summary to omit the report from review counts.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T20:44:08Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector · 2026-07-07T20:49:59Z

Codex Review: Didn't find any major issues. Can't wait for the next one!

Reviewed commit: f49b6f89a6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Ian747-tw · 2026-07-07T20:50:01Z

@codex review the Branch 9N OpenCode result report model

Ian747-tw · 2026-07-07T20:55:58Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 19ec98627c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T21:01:29Z

+    if (!launch && sessionId) {
+      const launches = await this.options.launchGateService.list({ session_id: sessionId, limit: MAX_LIST })
+      launch = launches.find((item) => LAUNCHED_STATUSES.has(item.status)) ?? launches[0] ?? null


Resolve the latest launch status before accepting reports

When callers provide only session_id, this selects a launch_started record directly from launchGateService.list(). A failed launch writes both a start event and a later failure event, and the public list can still contain the stale launch_started record for that same launch; in that case the preview/record path treats the session as active and allows a result report even though launchGateService.get(launch_id) would return launch_failed. This should re-resolve the chosen launch id through get() (as the other OpenCode metadata services do) before applying LAUNCHED_STATUSES.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T21:13:37Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4d71e5ba03

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T21:22:52Z

+}
+
+function fakeResultReviewState(input: string | undefined, kind: string): string {
+  if (input === "needs_commander_review" || input === "needs_human_review" || input === "not_ready_for_review" || input === "review_not_required" || input === "unknown") return input


Block non-review terminal states in fake reports

When the default fake TUI runtime is used, a terminal command such as /opencode-result-report kind=completion_report review_state=review_not_required ... is accepted because the fake model returns any recognized review state here and only blocks accepted/rejected later. The real result-report service blocks terminal reports unless they are queued for Commander or human review, so the default TUI can record a completion/failure report as review_not_required, bypassing the review-needed safety contract and diverging from real runtime behavior.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-07-07T21:22:52Z

+    if (resultKind === "inconclusive_report" && !outcome && followups.length === 0 && knownFailures.length === 0) blockers.push("inconclusive_report requires outcome, followups, or known_failures")
+    if (resultKind === "blocked_report" && followups.length === 0 && knownFailures.length === 0) blockers.push("blocked_report requires followups or known_failures")
+    if (String(payload.reviewState ?? payload.review_state ?? "").match(/accepted|rejected/i)) blockers.push("accepted/rejected review states belong to a future Commander result review gate")
+    const rawValues = [summary, outcome, ...changedFiles, ...testsRun, ...testResults, ...artifacts, ...metrics, ...claims, ...knownFailures, ...followups].filter(Boolean).join("\n")


Check fake raw payloads before truncating them

In the fake runtime path, summary and the CSV fields have already gone through preview()/readCsvPayload() before this raw-payload check, so a long log or full diff with markers past the first 160 characters is truncated and can be recorded instead of blocked. Since the real runtime rejects raw result payloads, the default TUI fake client accepts out-of-scope report data and can let user-simulation coverage miss the production validation.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T21:32:03Z

@codex review the Branch 9N OpenCode result report model

Ian747-tw · 2026-07-07T21:38:43Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f6e7c27338

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T21:43:36Z

+    if (sessionId) {
+      const session = await this.options.opencodeSessionService.get(sessionId)
+      if (!session) blockers.push("session_id does not resolve to a planned OpenCode session")


Validate launch-derived sessions before recording

When callers provide only launch_id/launch=<id>, sessionId is still empty here, so the planned-session lookup is skipped; the code then copies launch.session_id later and can record a result report for any orphan/restored/corrupt launch event whose session plan is missing. Since this feature's contract requires a resolvable planned OpenCode session as well as a launch, derive the session from the launch before this check or re-run the session validation after line 166.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T21:51:50Z

@codex review the Branch 9N OpenCode result report model

Ian747-tw · 2026-07-07T21:57:23Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 310ea5e912

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-07-07T22:02:28Z

+const LAUNCHED_STATUSES = new Set(["launch_started", "launched"])
+const RAW_PATTERNS = [
+  /\n.{100,}\n.{100,}/s,
+  /(stdout|stderr|traceback|stack trace|bun test v|npm error|diff --git|@@ |\+\+\+ |--- ).{0,120}\n/i,


Block all advertised raw payload markers

When a result summary/outcome contains a short provider output, file contents, events.jsonl, or research.db snippet, this regex does not mark it as raw, so /opencode-result-report can persist exactly the payloads the blocker text says are out of scope. The human-control service already blocks these keyword classes; without the same coverage here, bounded-looking raw provider/file/event data can be recorded into opencode_result_report_recorded events instead of being rejected.

Useful? React with 👍 / 👎.

Ian747-tw · 2026-07-07T22:08:25Z

@codex review the Branch 9N OpenCode result report model

Ian747-tw · 2026-07-07T22:17:08Z

@codex review the Branch 9N OpenCode result report model

chatgpt-codex-connector · 2026-07-07T22:23:22Z

Codex Review: Didn't find any major issues. Another round soon, please!

Reviewed commit: 96631b0658

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

M9.N: add OpenCode result report model

f050dd3

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: address result report review feedback

09210bd

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: enforce result report review requirements

e8163d5

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: require review queues for terminal reports

f49b6f8

M9.N: normalize terminal result review state

19ec986

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: resolve latest launch before result reports

4d71e5b

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: align fake result report validation

f6e7c27

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: validate launch-only result report sessions

310ea5e

chatgpt-codex-connector Bot reviewed Jul 7, 2026

View reviewed changes

M9.N: block raw result report payload markers

96631b0

Ian747-tw merged commit 57e1783 into main Jul 8, 2026
2 checks passed

Ian747-tw deleted the redesign/branch9n-opencode-result-report-model branch July 8, 2026 16:38

		wake_execution_id: optional(input.wakeExecutionId ?? input.wake_execution_id ?? input.wake_execution),
		wake_action_execution_id: optional(input.wakeActionExecutionId ?? input.wake_action_execution_id ?? input.wake_action),

		const reviewState = readReviewState(input.review_state, defaultReviewState(resultKind), blockers)
		if (resultKind !== "status_report" && reviewState === "review_not_required") blockers.push("terminal OpenCode result reports require Commander or human review")

Conversation

Ian747-tw commented Jul 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Result report model

Result disposition/review-state behavior

Launch/progress/evidence anchoring behavior

Preview/dry-run/record/list/latest/summary behavior

Duplicate behavior

Event persistence/redaction behavior

Fake-vs-real runtime behavior

Runtime commands/API surface

TUI result-report surface

No-mission-completion guarantee

No-researchdb-ingestion guarantee

No-provider/no-OpenCode-prompt/no-process-control guarantees

Authority registry update

Targeted validation output

Full regression status

Secret/redaction behavior

Explicit out-of-scope list

Notes about Branch 9O Commander result review gate

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

chatgpt-codex-connector Bot commented Jul 7, 2026

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

Ian747-tw commented Jul 7, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

Ian747-tw commented Jul 7, 2026 •

edited

Loading