efo report for 2026-05-15

[github-actions]

Agent activity for EBISPOT/efo is high in this 14-day window. Copilot authored 9 PRs — 7 merged, 1 open awaiting human review (PR #2694), and 1 closed without merging (PR #2690). All activity was agent-initiated; there were 0 explicit agent summons. The main pattern: Copilot opens PRs autonomously, then responds quickly to reviewer feedback (QC failures, merge conflicts, ID clashes). No scan errors detected.

[github-actions]

Qualitative Review — efo, 2026-05-15

What's Working

Copilot is highly productive in this repo. Seven PRs merged in the window, all Copilot-authored, covering a range of ontology tasks: term imports (PR #2691), new term additions (PR #2692, PR #2695, PR #2696, PR #2671), bulk obsoletions (PR #2693), and infrastructure improvements (PR #2698). The self-correction loop is genuinely functional — when reviewers flag issues, Copilot responds and re-commits within minutes. On PR #2693, Copilot handled a QC failure caused by 27 dangling MONDO terms independently after being asked; on PR #2698, it removed an unnecessary secret token when questioned by a reviewer.

What's Not Working / Recurring Friction

ID collision management is the most consistent failure mode. Three separate PRs encountered ID clash issues in this window:

• PR #2671 initially selected an ID already used by another PR and had to be corrected twice (once in March, once again in May when a reviewer re-checked).
• PR #2694 had ID clashes with other open PRs that required a mid-review fix.
• PR #2693 required a full re-check at merge time.

The agent appears to check IDs at PR creation but doesn't account for the moving target of concurrently-open PRs, leading to repeated human prompting.

QC failures are caught late. Both PR #2691 and PR #2693 had QC failures triggered by upstream MONDO deprecations. The agent fixes them when asked, but doesn't proactively identify that an upstream release might have changed the validation landscape.

Incomplete initial commits. On PR #2693, a reviewer noted that the initial commit contained only a plan checklist with no actual changes to efo-edit.owl. The agent recovered, but this suggests the initial commit workflow sometimes publishes an incomplete state.

PR #2690 was closed without merging, superseded by PR #2693. The dossier doesn't explain why the earlier attempt was abandoned rather than revised, but the two PRs cover overlapping scope (duplicate term obsoletions), suggesting the agent restarted rather than repaired.

Open Item

PR #2694 (desmosterol measurement terms) is open and awaiting human review. The ID clashes were corrected in a follow-up commit; no further blockers are flagged.

Potential Missed Opportunities

Three open issues with no agent involvement were surfaced:

Issue #2682 — bulk NTR for cholesterol precursor terms is the most actionable. It follows the same pattern as PR #2694 (a bulk term request template with a structured attachment), and Copilot has demonstrated this exact workflow works. Worth picking up once PR #2694 is merged to avoid compounding ID collisions.

Issue #2639 — specimen slide ID import is close to actionable but blocked on a human decision: the maintainer (zoependlington) is currently deciding whether to import OBIB:0002003 or create a new EFO term. The agent should not act until that decision is resolved.

Issue #2626 — disease progression trait pattern is a design-level request requiring a new ontology pattern proposal. It is open-ended and not bounded enough for a confident agent recommendation based on the current dossier.