Support orchestrated implement from PRs and discussions

[sepo-agent-app]

Summary

• Allow /orchestrate on discussions and enable planner preflight for discussion targets.
• Let planner-selected implement from PRs/discussions create a tracking issue first, then dispatch agent-implement.yml against it.
• Update orchestrator prompts/docs and add focused regression coverage.

Tests

• npm run build
• node --test dist/__tests__/triage.test.js dist/__tests__/orchestrator-preflight-cli.test.js dist/__tests__/handoff.test.js dist/__tests__/orchestrate-handoff-cli.test.js dist/__tests__/envelope.test.js
• YAML parse for agent-router.yml and agent-orchestrator.yml
• git diff --check

Closes #227

[lolipopshock]

why we added so many lines in .agent/src/cli/orchestrate-handoff.ts? we can just have a simple extension based on what we have?

@sepo-agent /answer

[sepo-agent-app]

Short answer: yes, conceptually this should be a simple extension. The line count grew because the PR implemented the missing “PR/discussion /orchestrate -> implement” path inside orchestrate-handoff.ts instead of extracting the already-existing non-issue /implement tracking-issue pattern into a shared helper.

Why some extra logic is still needed:

• agent-implement.yml is issue-based: it requires issue_number, not a PR/discussion target. See agent-implement.yml.
• Existing explicit /implement from PR/discussion already creates a tracking issue, dispatches implement against that issue, then links back. See agent-router.yml.
• Orchestrated handoff dedupe/status comments are issue-comment based, so once PR/discussion orchestration selects implement, we need a stable tracking issue before the normal handoff marker flow can work.
• The later added lines were mostly hardening: reuse an existing tracking issue on retry, avoid duplicate link-backs, escape copied Sepo control markers, and support discussion comment lookup. That lives around orchestrate-handoff.ts.

I agree the shape can be cleaner. The better version is:

1. Keep the small policy changes in triage.ts, orchestrator-preflight.ts, and handoff.ts.
2. Move the non-issue implementation tracking issue creation/reuse/link-back logic into a shared helper, e.g. .agent/src/implementation-tracking.ts.
3. Reuse that helper from both explicit /implement and orchestrated implement.
4. Leave orchestrate-handoff.ts with only the orchestration-specific call:

if (decision.nextAction === "implement" && targetKind !== "issue") {
  decision.targetNumber = ensureImplementationTrackingIssueForTarget(...);
}

So I would not remove the tracking issue/dedupe/marker safety entirely, but I do think we should refactor this PR so orchestrate-handoff.ts gets a small extension rather than owning all of that glue directly.

[lolipopshock]

Short answer: yes, conceptually this should be a simple extension. The line count grew because the PR implemented the missing “PR/discussion /orchestrate -> implement” path inside orchestrate-handoff.ts instead of extracting the already-existing non-issue /implement tracking-issue pattern into a shared helper.

Why some extra logic is still needed:

• agent-implement.yml is issue-based: it requires issue_number, not a PR/discussion target. See agent-implement.yml.
• Existing explicit /implement from PR/discussion already creates a tracking issue, dispatches implement against that issue, then links back. See agent-router.yml.
• Orchestrated handoff dedupe/status comments are issue-comment based, so once PR/discussion orchestration selects implement, we need a stable tracking issue before the normal handoff marker flow can work.
• The later added lines were mostly hardening: reuse an existing tracking issue on retry, avoid duplicate link-backs, escape copied Sepo control markers, and support discussion comment lookup. That lives around orchestrate-handoff.ts.

I agree the shape can be cleaner. The better version is:

1. Keep the small policy changes in triage.ts, orchestrator-preflight.ts, and handoff.ts.
2. Move the non-issue implementation tracking issue creation/reuse/link-back logic into a shared helper, e.g. .agent/src/implementation-tracking.ts.
3. Reuse that helper from both explicit /implement and orchestrated implement.
4. Leave orchestrate-handoff.ts with only the orchestration-specific call:

if (decision.nextAction === "implement" && targetKind !== "issue") {
  decision.targetNumber = ensureImplementationTrackingIssueForTarget(...);
}

So I would not remove the tracking issue/dedupe/marker safety entirely, but I do think we should refactor this PR so orchestrate-handoff.ts gets a small extension rather than owning all of that glue directly.

@sepo-agent /orchestrate

[lolipopshock]

@sepo-agent /answer

I thought the goal of #294 (comment) is to make the PR simpler, but it adds more lines??

[sepo-agent-app]

Yes, your read is fair.

The refactor made the architecture simpler, but not the net diff smaller.

At current PR head 3b256955:

• .agent/src/cli/orchestrate-handoff.ts got much smaller versus the previous heavy version: about 2124 lines -> 1739 lines.
• Compared to main, though, it is still slightly larger: about 1662 -> 1739.
• The total PR grew because the logic moved into a new shared helper:
  • implementation-tracking.ts adds the shared create/reuse/link-back/marker-safety behavior.
  • ensure-implementation-tracking.ts lets the workflow call that helper.
  • New tests add a lot of lines too.

So the change did address the specific concern from your comment: orchestrate-handoff.ts no longer owns most of the tracking issue logic. The orchestrated path is now basically the intended small call:

if (decision.nextAction === "implement" && targetKind !== "issue") {
  dispatchTargetNumber = ensureImplementationTrackingIssueForTarget(...)
}

See orchestrate-handoff.ts.

But if the goal is “make the whole PR smaller”, then yes, this still misses that bar. It centralized behavior, but kept all the hardening: marker escaping, retry reuse, duplicate link-back prevention, discussion support, and tests.

The smallest follow-up direction would be: keep the shared helper for tracking issue creation/reuse, but avoid moving every link-back/threading behavior into it. The latest review already found this tradeoff: explicit /implement lost original response-threading, so preserving the old post-response.js path for explicit /implement may be simpler than expanding the helper further.

[sepo-agent-app]

Sepo pushed fixes for this PR. Branch: agent/implement-issue-227/codex-26126733366. Requested by @sepo-agent.

• Escaped every <!-- sepo- opener in copied tracking issue content, including incomplete marker openers without a closing -->.
• Added generated tracking issue regressions for incomplete Sepo marker openers in request text and target body.
• Verified with npm run build, focused node --test ..., and git diff --check.

[sepo-agent-app]

Sepo is dispatching follow-up automation.

Source	Next	Target	Round	Status
fix-pr	review	PR #294	11 / 40	Dispatched

Reason: agent planner selected review: fix-pr succeeded after addressing the latest review action item, so one automated review should verify the marker-escaping fix and regressions before stopping.

[sepo-agent-app]

Rubrics Review

Total Score	Verdict	Rubrics Scored
99	PASS	13

Dimension	Rubric	Result	Score	Evidence
coding_workflow / generic	Validate delegated route authorization	pass	8/8	Router passes requester/auth context into orchestrator dispatch, and preflight blocks planner use when delegated capability authorization fails: agent-router.yml, orchestrator-preflight.ts.
coding_workflow / generic	Generate context-derived artifact metadata	pass	7/7	Tracking issues are generated from target metadata, requester, planner reason, request/context, and base inputs rather than copying the slash command as metadata: implementation-tracking.ts.
coding_workflow / generic	Reuse existing code	pass	8/8	Explicit non-issue /implement and orchestrated non-issue implement both use the shared tracking helper/CLI: agent-router.yml, orchestrate-handoff.ts.
coding_workflow / generic	Minimal defaults, native escape hatches	pass	8/8	The PR keeps the common path minimal: non-issue implement normalizes through a tracking issue, while base_pr / base_branch remain explicit typed escape hatches.
coding_workflow / generic	Prefer explicit structured inputs	pass	7/7	The helper and handoff path use typed fields such as targetKind, trackingScope, baseBranch, basePr, response kind, and threaded IDs; conflicting base inputs stop clearly: implementation-tracking.ts, orchestrate-handoff.ts.
coding_workflow / generic	Prefer agent-derived decisions	pass	7/7	Planner policy and deterministic handoff validation decide PR/discussion orchestration behavior without adding a new repo setting: agent-orchestrator.md, handoff.ts.
coding_workflow / generic	Separate decision gates from action authorization	pass	8/8	Planner decisions remain constrained by target kind: discussions can dispatch only implement, while PRs can dispatch review, fix-pr, or separate implement: handoff.ts.
coding_workflow / generic	Keep docs in sync	pass	7/7	README and architecture docs now describe PR/discussion orchestration and shared tracking issue behavior: README.md, supported-workflows.md.
coding_workflow / generic	Make user-facing automation idempotent	pass	7/7	The helper checks trusted threaded/top-level link-backs and trusted issue-body markers before creating, then skips duplicate link-backs: implementation-tracking.ts, implementation-tracking.ts.
coding_workflow / generic	Surface stateful failures	pass	7/7	Tracking issue failures stop the handoff, set outputs, log the reason, and attempt a visible stop comment: orchestrate-handoff.ts.
coding_workflow / generic	Comment on accepted workflow stops	pass	6/6	Discussion stop dedupe has focused coverage for trusted final comments, avoiding silent stops and duplicate status noise: orchestrate-handoff-cli.test.ts.
coding_workflow / generic	Surgical changes	pass	8/8	The file list is scoped to orchestration policy, shared tracking helper/CLI, workflow wiring, docs/prompts, and focused tests for the PR/discussion implement path.
coding_style / generic	Prefer simple implementations	partial	6/7	The call sites are now simple and shared, matching the requested shape. The helper is still sizable because it centralizes create/reuse/link-back/threading/marker-safety behavior in one module.

Notes

• The current PR head satisfies the main rubric concern from the request: tracking issue create/reuse/link-back logic moved into .agent/src/implementation-tracking.ts, and orchestrate-handoff.ts now makes a small route-specific call.
• The prior idempotency and marker-safety gaps are covered by focused tests, including incomplete <!-- sepo- marker openers, threaded PR review-comment retries, and discussion link-back dedupe.
• Verification run: npm run build and focused node --test ... passed with 230 passing tests.
• No rubric-specific follow-up needed.

Findings

• INFO: No blocking rubric-specific issues found. The only tradeoff is helper size, but the complexity is now centralized and reused by both explicit /implement and orchestrated implement.

Final Rubric Verdict

PASS

[lolipopshock]

Superseded by the smaller focused replacement PR #302.

[sepo-agent-app]

AI Review Synthesis

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

Summary of PR/Issue

• PR Support orchestrated implement from PRs and discussions #294 addresses Orchestrator small bugs: #227 by allowing /orchestrate on discussion targets and letting planner-selected implement run from PRs and discussions.
• Non-issue implement handoffs now create or reuse a shared implementation tracking issue before dispatching agent-implement.yml.
• The PR updates route policy, planner validation, workflow wiring, prompts, docs, and focused regression tests.
• Follow-up fixes addressed the earlier marker escaping, tracking issue reuse, threaded link-back, and discussion stop-comment dedupe concerns.

Review

The current PR head a33e26fec8b7923dc4b7919a85a23507b4638180 is ready to ship. The code review and rubrics review agree there are no unresolved BLOCKING or WARNING findings; the only remaining note is INFO-level helper size tradeoff.

Issue	Severity	Description
No actionable issues remain	INFO	Current reviewer outputs report SHIP/PASS, and the current diff supports that conclusion. Prior findings are fixed and covered by focused tests.

Progress

• Issue Orchestrator small bugs: #227’s requested minimal behavior is implemented for PR and discussion /orchestrate paths.
• Prior marker-safety issues are fixed by escaping every <!-- sepo- opener in copied tracking-issue content.
• Tracking issue reuse and link-back dedupe now cover top-level and threaded PR/discussion surfaces.
• Re-fetched inline comments and review threads; resolved stale same-agent thread PRRT_kwDOSOhEMM6DVLEr.
• No new inline comments, replies, or comment minimizations were added.

Recommended Next Step

NO_AUTOMATED_ACTION: no unresolved required branch-change work remains.

Final Verdict

SHIP

Action Items

• No required branch-change work remains.