Agentic CI: Issue & PR Triage Tracker

[andreatnvidia]

This issue collects weekly triage reports generated by the Agentic CI: Repository Triage workflow.

Each Monday at 10:00 UTC the workflow scans open issues and PRs, classifies them, detects staleness and duplicates, and posts a summary as a comment below.

[github-actions]

Repository Triage Report — 2026-08-31

Open issues: 54 | Open PRs: 11 | Healthy (no action needed): 27 (21 issues, 6 PRs)

---

Close as resolved (1)

#	Title	Action	Evidence	Rationale
#866	Resolve deployment-owned vLLM serving plans for Slurm	Close	PR #890 "Closes #866" merged 2026-08-28	Merged into the feat/slurm-execution integration branch, not main, so GitHub never auto-closed it.

Note: sibling Slurm issues #865, #872, #873, #880, #851–#853 were closed correctly; only #866 was missed.

Close as duplicate (1)

#	Title	Action	Evidence	Rationale
#850	Epic: Slurm batch execution v1	Close or link as child of #160	Overlaps #160 (2025-12-22, older, on-roadmap/in-progress)	Two open umbrella items track the same Slurm delivery; pick one canonical tracker.

Needs maintainer decision (7)

#	Title	Action	Evidence	Rationale
#897	Add OrcaRouter as a built-in model provider	Triage now (add/withhold triaged)	Blocks PR #898's check job	External contributor is blocked solely on the missing triaged label; decide whether to accept a new built-in provider.
#790	feat: generate an exact number of rows matching a declared criterion	Pick an implementation approach	PRs #817 and #830 are competing designs; #830 says "intent is comparison, not replacement"	Two full implementations are parked waiting on a design call.
#182	Improve validation in structured generation columns	Decide direction	needs-discussion, 213d idle	No scope agreed since January.
#677	Add first-class reranking model support	Decide direction	needs-discussion, 104d idle	New model modality; needs a product call before scoping.
#352	Ability to control order of multi-modal context along with prompts	Decide direction	needs-discussion+on-roadmap, 101d idle	On the roadmap but no agreed API shape.
#711	Allow model providers to restrict workflows to approved models	Decide direction	needs-discussion, 94d idle	Policy/enforcement boundary is unclear.
#651	Design resource-vector and provider-aware admission policies	Decide whether to run the spike	spike+plan, 102d idle	Spike never started; child of epic #645.

Ready for assignment (3)

#	Title	Action	Evidence	Rationale
#882	Support pandas>3 in data-designer	Assign	10d old, no assignee, scope clear	Three pins to relax (pandas>=2.3.3,<3 in all three package pyprojects) plus compat testing.
#885	Add keyed execution for group-scoped consistent generation	Assign	10d old, task, no assignee	Well-specified engine task with no competing work.
#862	Support es_SV (El Salvador) locale in PersonSampler	Assign	19d old, no assignee	Additive locale support; follows an existing pattern.

Stuck PR (3)

#	Title	Action	Evidence	Rationale
#898	feat: add orcarouter as a built-in model provider	Unblock by triaging #897	1 failing check: linked-issue check reports "#897 has not been triaged yet"	Not an author problem — the PR is blocked on maintainer triage of #897.
#884	Add Word document generation Colab notebook	Ask author to retitle	2 failing semantic-pull-request checks, 10d since push	Title lacks a conventional-commit prefix (e.g. docs: add Word document generation Colab notebook).
#861	feat: expose terminal failure locations	Nudge author / review	0 failing checks, 19d since last push, no reviews	Green and unreviewed; closes #860 but has gone quiet.

Duplicate PRs (1)

#	Title	Action	Evidence	Rationale
#817 / #830	engine-native record selection vs. cohort-based record retry	Pick one, close the other	Both target #790 (#817 "Closes #790", #830 "Alternative design to #817")	Deliberate A/B by the same author; both are stale (38d / 31d) awaiting the design call in #790.

Stale, consider closing (21)

#	Title	Action	Evidence	Rationale
#246	Update health-checks to account for multi-modality	Close with note	200d idle, no assignee	No traction since February.
#354	Add other common params to ChatCompletionInferenceParams	Close or assign	186d idle, no assignee	Small additive change nobody has picked up.
#362	Allow user to keep record fields that failed to parse as null	Close with note	175d idle, no assignee	External request, never scoped.
#394	Jinja2 templates cannot reference PRE_BATCH processor columns	Confirm or close	173d idle, bug, no assignee	Open bug with no repro activity — verify it still reproduces before closing.
#400	Enable non-existent tool fallback with MCP server	Close with note	172d idle, no assignee	No follow-up from reporter.
#503	Add request/response logging hook to native model client	Close or assign	143d idle, no assignee	May be partly covered by the diagnostics work in #800/#802.
#386	Add structured metadata to log records	Close or assign	143d idle, no assignee	Overlaps logging work landed in #755/#808.
#563	Skill: auto-port custom columns to plugin packages	Close with note	133d idle, no assignee	Tooling nice-to-have, no demand signal.
#573	Concrete schemas for all nemotron-personas datasets	Close or assign	129d idle, no assignee	Data-side task with no owner.
#211	Batch-level error rates for early shutdown	Reconfirm roadmap slot	123d idle, on-roadmap, no assignee	Marked on-roadmap but unowned for 4 months.
#601	tests: cover structural_impact.py pure helpers	Assign, don't close	119d idle; .agents/tools/structural_impact.py still has no test file	Still valid and small — better as a good first issue than a close.
#619	Push Data Designer datasets to Harbor Hub	Close with note	116d idle, no assignee	Integration request with no owner.
#637	Split docs preview build and deploy workflows	Close or assign	110d idle, no assignee	May be obsolete after the Fern publishing fixes (#818/#842).
#660	Document async scheduling architecture, capacity, and operations	Close or assign	109d idle, plan, no assignee	Docs plan under epic #645, never started.
#659	uv tool update removes DataDesigner plugin installs	Confirm or close	109d idle, bug, no assignee	Packaging bug — verify against current uv before closing.
#695	Report per-column idle time in end-of-run metrics	Close or fold into #727	102d idle; overlaps assigned issue #727	Narrower slice of the capacity-diagnostics work already owned by @eric-tramel.
#741	Public async runs cannot tune row-group horizon/adaptive admission	Re-scope or close	87d idle; merged #745 ("Refs #741") landed max_concurrent_row_groups	Fixed horizon is delivered; only the adaptive-admission half remains.
#753	Add README to plans/ folder	Assign, don't close	76d idle; plans/README.md still absent	Genuinely small and still needed.
#759	Conflicting guidance on lazy heavy imports	Assign, don't close	75d idle; STYLEGUIDE.md/lazy_heavy_imports.py conflict unresolved	Docs inconsistency confirmed present.
#758	Add native provider request support for non-chat plugin model calls	Close or assign	75d idle, task, no assignee	Plugin-surface work with no owner.
#760	Enforce from __future__ import annotations via ruff FA102	Assign, don't close	73d idle; no FA rules in any pyproject.toml select list	AGENTS.md invariant is still unenforced — small, mechanical fix.

Five of these (#601, #753, #759, #760, and arguably #394) are stale by age but still verifiably valid and small. Recommend good first issue rather than closing.

Repeatedly-failed fix attempts (0)

No .agentic-ci-state/ runner-state files were available in this run, so no attempted_fixes could be checked.

---

Healthy items (21 issues, 6 PRs)

Issues

• Agentic CI: Issue & PR Triage Tracker #562 Agentic CI: Issue & PR Triage Tracker — andreatnvidia — 2026-08-24 (this tracker)
• Implement the allocation-local Data Designer client worker #876 Implement the allocation-local Data Designer client worker — nabinchha — 2026-08-24 (assigned)
• Compile Slurm configuration into deterministic execution plans #875 Compile Slurm configuration into deterministic execution plans — nabinchha — 2026-08-24 (assigned, PR feat: compile deterministic Slurm run plans #893 open)
• Run, observe, and analyze Slurm benchmarks #877 Run, observe, and analyze Slurm benchmarks — nabinchha — 2026-08-25 (assigned)
• Expose Slurm services, CLI, packaging, and documentation #874 Expose Slurm services, CLI, packaging, and documentation — nabinchha — 2026-08-25 (assigned, PR feat: add Slurm service foundation #894 open)
• Persist Slurm run state, retry shards, and collect winners #869 Persist Slurm run state, retry shards, and collect winners — nabinchha — 2026-08-25 (assigned)
• Implement Slurm submission and allocation runtime #868 Implement Slurm submission and allocation runtime — nabinchha — 2026-08-25 (assigned)
• Build, inspect, and manage SQSH images for Slurm #867 Build, inspect, and manage SQSH images for Slurm — nabinchha — 2026-08-25 (assigned, PR feat: add Slurm image lifecycle jobs #896 open)
• Harden the Slurm runtime and validate sealed release artifacts #870 Harden the Slurm runtime and validate sealed release artifacts — nabinchha — 2026-08-15 (assigned)
• Expose terminal failed-column and seed-row provenance in public results #860 Expose terminal failed-column and seed-row provenance — andreatnvidia — 2026-08-12 (PR feat: expose terminal failure locations #861 open)
• Support Scaling Data Designer on SLURM #160 Support Scaling Data Designer on SLURM — kirit93 — 2026-08-03 (assigned, in-progress)
• Support more types of Model Providers like VertexAI and Bedrock #92 Support more types of Model Providers like VertexAI and Bedrock — kirit93 — 2026-08-03 (on-roadmap)
• chore: consolidate docs/ and fern/ into one documentation tree #837 chore: consolidate docs/ and fern/ into one documentation tree — andreatnvidia — 2026-07-30
• refactor: interface-layer OpenTelemetry runtime follow-ups deferred from #802 #809 refactor: interface-layer OpenTelemetry follow-ups from feat: add OpenTelemetry metrics for create jobs #802 — nabinchha — 2026-07-08
• Consolidate Data Designer user configuration into a single TOML file #694 Consolidate user configuration into a single TOML file — eric-tramel — 2026-07-08
• refactor(config): non-breaking hygiene pass on RunConfig #807 refactor(config): non-breaking hygiene pass on RunConfig — nabinchha — 2026-07-07
• Harden Agentic CI PR reviews for forked pull requests #804 Harden Agentic CI PR reviews for forked pull requests — andreatnvidia — 2026-07-07
• Public async runs do not expose capacity diagnostics needed to explain endpoint idle time #727 Public async runs do not expose capacity diagnostics — eric-tramel — 2026-06-01 (assigned)
• Epic: Async Scheduling Resource Metadata, Admission, and Observability #645 Epic: Async Scheduling Resource Metadata, Admission, Observability — eric-tramel — 2026-05-14 (assigned)
• refactor: introduce AsyncRunController and explicit sinks for async build/preview #447 refactor: introduce AsyncRunController and explicit sinks — andreatnvidia — 2026-03-25 (assigned)
• Add more stages to processors, and allow for custom processors via callbacks #159 Add more stages to processors, allow custom processors via callbacks — andreatnvidia — 2026-03-02 (assigned)

PRs

• docs(agentic-ci): fix stale param name in register_model_configs docstring #899 docs(agentic-ci): fix stale param name in register_model_configs docstring — github-actions[bot] — 2026-08-31 (green)
• feat: add Slurm image lifecycle jobs #896 feat: add Slurm image lifecycle jobs — nabinchha — 2026-08-28 (green)
• chore: strengthen code review design pass #895 chore: strengthen code review design pass — nabinchha — 2026-08-28 (green, 68 checks)
• feat: add Slurm service foundation #894 feat: add Slurm service foundation — andreatnvidia — 2026-08-29 (green)
• feat: compile deterministic Slurm run plans #893 feat: compile deterministic Slurm run plans — andreatnvidia — 2026-08-28 (green)
• fix: harden linked issue validation #881 fix: harden linked issue validation — nabinchha — 2026-08-27 (green)

Summary

• 38 items flagged for action across 7 buckets (33 issues, 5 PRs)
• 5 PRs flagged — 3 stuck (feat: add orcarouter as a built-in model provider #898, Add Word document generation Colab notebook (uses the docx processor plugin) #884, feat: expose terminal failure locations #861), 2 duplicate (feat: add engine-native record selection #817/feat(workflow): add cohort-based record retry #830)
• Highest-leverage single action: triage Add OrcaRouter as a built-in model provider #897 — it unblocks external PR feat: add orcarouter as a built-in model provider #898's only failing check
• Second: decide the feat: generate an exact number of rows matching a declared criterion #790 approach — it unblocks two complete, stale PRs at once
• 21 stale issues dominate the backlog; 5 of them are still-valid small tasks better relabeled than closed
• 27 healthy items (collapsed above)