cell-ontology report for 2026-05-19

[github-actions]

Report date: 2026-05-19 | Lookback: 14 days

Copilot is actively authoring new cell-type term PRs, but no agent-touched PRs merged in the window. All 7 agent-authored PRs are open and stalled. The CLARA review bot (Claude-backed) is operational and being invoked regularly via /clara. Recurring curator friction centers on species specificity and branch staleness. One concrete unaddressed issue (#3642) is a potential low-friction opportunity.

[github-actions]

Review: cell-ontology — 2026-05-19

What seems to be working

• Copilot is generating new term PRs at a steady pace. Seven agent-authored PRs were opened or active in the 14-day window, all as new cell-type term requests (NTRs) — hippocampal interneurons, dentate gyrus lineage cells, macrophages, and more.
• CLARA (the Claude-backed ontology reviewer) is functioning. It was invoked via /clara on at least four PRs this period (PR #3631, PR #3637, PR #3645, PR #3614) and is completing reviews within 3–14 minutes.
• Copilot responds correctly to direct correction requests. On PR #3604, it changed a synonym annotation type from hasRelatedSynonym to hasExactSynonym when asked. On PR #3631, it reverted problematic changes promptly when the human asked it to.
• The NTR → PR pipeline is functional end-to-end. Issue #3644 (parvalbumin-positive basket cell NTR) was filed at 14:09 UTC and a Copilot PR (PR #3645) was open by 14:14 UTC — a 5-minute turnaround. Similarly Issue #3636 (bistratified interneuron) is already addressed by PR #3637, opened the same day.

What seems not to be working

• Zero merges. Despite 7 agent-authored PRs and active curator engagement, nothing merged in the 14-day window. All 7 items are classified as stalled.
• No agent summons via @copilot. All 7 Copilot contributions are self-initiated PR opens. Curators are not using @copilot to direct the agent mid-review; they are using #gogoeditdiff and /clara for CI tools but not routing correction tasks back through Copilot (with a couple of exceptions on PR #3604 and PR #3631).
• PR #3610 is 13 commits behind main and blocked from progress by branch staleness. The CI comment flags this but there is no automated or agent-driven branch-update step.

Recurring friction

• Species specificity. Multiple PRs have received curator review requests to relax mouse-specific protein markers to species-general terms and flag taxon scope in definitions. This has come up on PR #3624 (AVPV/PeN kisspeptin neuron, still under review as of May 14), PR #3610 (dentate gyrus terms, requested April 30, still open), and PR #3614 (iron-recycling macrophage). This is a recurring correction pattern across Copilot-authored NTR PRs and suggests Copilot's term drafts are defaulting to mouse-specific evidence without adequately flagging taxon scope.

• Import management scope error. On PR #3631, Copilot was asked to "fix the issue" (a CI failure caused by a deprecated NCBITaxon:6657 term) and responded by modifying merged_import.owl and non_native_classes.txt — the wrong approach. The correct fix was to sync with an upstream branch that had already resolved this in PR #3633. Copilot over-solved by touching import infrastructure it should not have modified. The human had to instruct it to revert. This is a scoping/context failure: Copilot did not know about the parallel upstream fix.

• Stall accumulation. The pipeline produces PRs faster than they are being reviewed and merged. Five of the seven stalled PRs have been open for 2–4 weeks with curator feedback left unaddressed. This may be a review capacity issue rather than an agent failure, but it means the agent's output is building up without landing.

Potential missed opportunities

The dossier surfaces three issues under "Potential Missed Opportunities," but two of them are not actually missed:

• Issue #3644 is already addressed by PR #3645, opened the same day.
• Issue #3636 is already addressed by PR #3637.

One issue appears genuinely unaddressed:

• Issue #3642 — a data-type annotation bug: rdfs:comment on CL_4023106 is typed as xsd:boolean instead of the expected string type. This is a concrete, bounded fix (a single annotation correction in the OWL file) with no ambiguity about what needs to change. It would be a low-friction starting point for an agent-authored PR, though the dossier does not include the full issue body so confidence in the scope is moderate.