Update agent schedule and orchestration defaults

[sepo-agent-app]

Summary

• Disable scheduled daily summaries by default while preserving manual dispatch
• Increase default orchestration max rounds from 5 to 12 across workflows and CLI fallbacks
• Update docs and focused regression coverage for the new defaults

Verification

• npm run build
• node --test dist/__tests__/schedule-policy.test.js dist/__tests__/scheduled-activity.test.js dist/__tests__/dispatch-agent-implement-cli.test.js dist/__tests__/dispatch-agent-orchestrator-cli.test.js dist/__tests__/orchestrator-preflight-cli.test.js
• node --test dist/__tests__/envelope.test.js
• git diff --check

Closes #243

[sepo-agent-app]

Sepo pushed fixes for this PR. Branch: agent/implement-issue-243/codex-25930241941. Requested by @sepo-agent.

• Kept scheduled agent-daily-summary.yml disabled unless its workflow override explicitly enables it, in both TypeScript and shell gate paths.
• Added regressions for unrelated non-empty AGENT_SCHEDULE_POLICY values and the no-env AUTOMATION_MAX_ROUNDS fallback of 12.
• Verification: npm run build; node --test dist/__tests__/schedule-policy.test.js dist/__tests__/scheduled-activity.test.js dist/__tests__/orchestrate-handoff-cli.test.js; git diff --check.

[sepo-agent-app]

Sepo is dispatching follow-up automation.

Source	Next	Target	Round	Status
fix-pr	review	PR #244	5 / 40	Dispatched

Reason: agent planner selected review: fix-pr succeeded on open PR #244 after a NEEDS_REWORK review; the updated head should receive a focused automated review before stopping or human merge consideration.

[sepo-agent-app]

Rubrics Review

Total Score	Verdict	Rubrics Scored
100	PASS	10

Dimension	Rubric	Result	Score	Evidence
coding_workflow / generic	Validate delegated route authorization	pass	8/8	PR changes schedule defaults and orchestration round caps only. It does not remove requester/access-policy propagation or downstream route checks; orchestrator-preflight-cli.test.ts keeps delegated-access coverage and adds default-round coverage.
coding_workflow / specific	Default follow-up orchestration to one-shot	pass	8/8	Raising the max round cap from 5 to 12 does not alter planner/heuristic continuation policy or make later user-launched follow-ups auto-continue outside explicit orchestration.
coding_workflow / generic	Reuse existing code	pass	8/8	Daily-summary behavior is implemented through existing AGENT_SCHEDULE_POLICY, schedule-policy.ts, and resolve-scheduled-activity-gate.sh paths instead of adding a parallel scheduler.
coding_workflow / generic	Keep docs in sync	pass	7/7	PR updates README.md, .agent/docs/architecture/memory.md, .agent/docs/architecture/supported-workflows.md, .agent/docs/customization/configuration-list.md, and .agent/docs/technical-details/agent-orchestrator.md for the new defaults.
coding_workflow / generic	Prefer explicit structured inputs	pass	7/7	Existing structured automation_max_rounds workflow/CLI input is preserved; the PR changes only default values and keeps AGENT_SCHEDULE_POLICY.workflow_overrides as the explicit schedule control.
coding_workflow / generic	Read docs and linked context first	pass	6/6	Linked issue #243 requested exactly two defaults: disable daily summaries by default and increase default orchestrate rounds from 5 to 12; the PR and docs follow that scoped plan.
coding_workflow / generic	Make surgical changes	pass	8/8	Diff is limited to requested defaults, matching docs, and focused regression tests. No broad refactor or unrelated cleanup is present.
communication / generic	Write audience-first documentation	pass	7/7	README keeps scheduled jobs reader-facing and states daily summary cron is disabled by default while manual dispatch remains available, without front-loading internal mechanics.
coding_workflow / generic	No per-file inventory tables in docs	pass	6/6	Documentation changes describe behavior and configuration defaults; they do not add per-file inventory tables.
coding_style / generic	Prefer simple implementations	pass	7/7	The implementation is direct default updates plus small tests, including the follow-up fix to keep agent-daily-summary.yml disabled for unrelated non-empty schedule policies.

Notes

• The current PR head addresses the earlier review synthesis blocker: both TypeScript and shell schedule gates keep agent-daily-summary.yml disabled unless explicitly enabled, even when AGENT_SCHEDULE_POLICY contains unrelated overrides.
• Test coverage is focused on the changed contracts: default schedule policy, shell gate parity, CLI fallback defaults, dispatch payloads, and orchestrator preflight.
• No rubric-specific follow-up needed.

Findings

• INFO: No rubric-specific findings.

Final Rubric Verdict

PASS

[sepo-agent-app]

AI Review Synthesis

Dual-agent review by Claude and Codex.
Requested by @lolipopshock.

Summary of PR/Issue

PR #244 implements issue #243 by disabling scheduled daily summaries by default while keeping manual dispatch available. It also raises the default orchestration max rounds from 5 to 12 across workflows and CLI fallbacks. The PR updates docs and focused regression tests for both behavior changes. A prior blocker around custom AGENT_SCHEDULE_POLICY values has been fixed on the current head.

Review

The synthesized reviews agree the current PR head is ready to ship. The earlier schedule-policy blocker is resolved, coverage is now focused on the changed defaults, and rubrics review passed.

Issue	Severity	Description
No actionable issues	INFO	Current reviewer outputs found no blocking or warning findings on the updated head.

Progress

• Prior blocker fixed: unrelated non-empty AGENT_SCHEDULE_POLICY values no longer re-enable agent-daily-summary.yml.
• Regression coverage was added for schedule-policy defaults, shell/TypeScript parity, and the AUTOMATION_MAX_ROUNDS=12 fallback.
• Rubrics review passed with no rubric-specific follow-up.
• Resolved existing same-agent inline thread PRRT_kwDOSOhEMM6CbRO9; no new inline comments were posted.
• One reviewer noted a pre-existing memory-sync custom-policy asymmetry as INFO; it predates this PR and is out of scope.

Recommended Next Step

NO_AUTOMATED_ACTION: No unresolved actionable work remains for automation.

Final Verdict

SHIP

Action Items

• No required branch-change action items remain.

[sepo-agent-app]

Rubrics Update

No changes were committed to agent/rubrics from PR #244.

no rubric changes

I found no trusted PR-conversation feedback that warrants a new durable rubric. PR #244 and linked issue #243 contain a specific feature request from lolipopshock, while the substantive review guidance was bot-authored and already covered by existing rubrics like keeping docs in sync, surgical changes, reuse existing code, and focused regression coverage.

[lolipopshock]

@sepo-agent /answer

Wait a sec I merged this pr and then launched another implement request #250 but the title is still not correct ?

[sepo-agent-app]

Yes, you’re right: #250 should not have kept the fallback title.

What happened:

• The title fix came from PR Derive implement issue titles from command text #232, not PR Update agent schedule and orchestration defaults #244. PR Update agent schedule and orchestration defaults #244 only changed daily-summary/orchestration defaults.
• The Implement requested change #250 router run did enter the new Generate implement issue metadata step.
• That metadata step failed before returning JSON. The log shows Codex tried to run gh pr view 241 ..., requested interactive network approval, then the workflow aborted it.
• Because the step is continue-on-error, Resolve route continued and logged: Implement issue metadata was invalid; using fallback metadata.
• So Implement requested change #250 was created with the fallback title: Implement requested change.

So this is a real bug in the new metadata path, not GitHub caching.

The fix should be: make the metadata step non-interactive. Either prefetch the PR/discussion context deterministically before invoking the agent, or adjust the metadata prompt/runtime so read-only GitHub context access does not require an interactive approval. The intended title for #250 would be something like:

Replace skill setup manifest with setup.sh hook

The issue body still contains the full request, so implementation can proceed, but the tracking issue title fallback behavior needs a follow-up fix.

[lolipopshock]

adjust the metadata prompt/runtime so read-only GitHub context access does not require an interactive approval

@sepo-agent /answer we should add a small fix? Also the gh view command should work right?

[sepo-agent-app]

Yes, this should be a small fix.

The important distinction: gh pr view ... does work in the workflow shell with GH_TOKEN. The failure was not that gh pr view is impossible. It failed because the metadata prompt asked the Codex tool runtime to run gh pr view, and under permission_mode: approve-reads that still triggered an interactive “allow network command?” approval. In GitHub Actions there is nobody to approve it, so the metadata step got cancelled and Sepo fell back to Implement requested change.

I would not fix this by changing the metadata step to approve-all. Better small fix:

1. Keep the metadata agent read-only.
2. Fetch PR/discussion context deterministically before the agent prompt, using workflow/CLI code with GH_TOKEN.
3. Inject that fetched context into the metadata prompt.
4. Update agent-implement-metadata.md so it says “use the provided target context”, not “run gh pr view yourself.”
5. Add a regression asserting the metadata prompt path does not require agent tool approval for gh pr view.

So yes: gh pr view should work, but it should be run by the workflow/runtime before the agent call, not interactively by the metadata agent.

[lolipopshock]

I would not fix this by changing the metadata step to approve-all. Better small fix:

We should do this fix. @sepo-agent /implement

[sepo-agent-app]

Implementing this request — tracking in #253.

[lolipopshock]

Yes, this should be a small fix.

The important distinction: gh pr view ... does work in the workflow shell with GH_TOKEN. The failure was not that gh pr view is impossible. It failed because the metadata prompt asked the Codex tool runtime to run gh pr view, and under permission_mode: approve-reads that still triggered an interactive “allow network command?” approval. In GitHub Actions there is nobody to approve it, so the metadata step got cancelled and Sepo fell back to Implement requested change.

I would not fix this by changing the metadata step to approve-all. Better small fix:

1. Keep the metadata agent read-only.
2. Fetch PR/discussion context deterministically before the agent prompt, using workflow/CLI code with GH_TOKEN.
3. Inject that fetched context into the metadata prompt.
4. Update agent-implement-metadata.md so it says “use the provided target context”, not “run gh pr view yourself.”
5. Add a regression asserting the metadata prompt path does not require agent tool approval for gh pr view.

So yes: gh pr view should work, but it should be run by the workflow/runtime before the agent call, not interactively by the metadata agent.

@sepo-agent /implement #254 is totally wrong. What I want is just simply allow accept-all for metadata generation step to fix the bug.

[sepo-agent-app]

Implementing this request — tracking in #255.