[eslint-refiner] ESLint Refiner — Daily Report (2026-08-26)

[github-actions[bot]]

Reviewed prefer-actions-exec-over-child-process, the newest rule in eslint-factory's import list (58 rules total, up from 12 at the last run on 2026-07-08 — a 46-day gap in this workflow's cadence). Filed 2 grounded issues; both quality-checked against the live actions/setup/js corpus rather than filed on suspicion alone.

Key metrics

• Rules registered: 58 (was 12 last run) — 45 still unreviewed.
• Issues filed today: 2 / 3 cap (quality over cap).
• Rule reviewed: prefer-actions-exec-over-child-process (flags child_process.exec/execSync/execFile/execFileSync in github-script-marked modules, suggesting @actions/exec instead).
• Grounding: live corpus grep + cross-reference of shim.cjs / setup_globals.cjs to check which globals are actually available at runtime.

Issues filed

1. False positive — dual-mode shim.cjs scripts. The rule's only scoping check is a /// <reference types="@actions/github-script" /> comment. But ~38 non-test .cjs files also require("./shim.cjs") so they can run either as a require()'d github-script module (where exec truly is a global) or as a standalone Node process such as inside the safe-outputs/mcp-scripts MCP servers (where shim.cjs only stubs core/context — never exec). For those files, child_process.execFileSync is the only correct choice; switching to @actions/exec would break the standalone path.
   Grounded at build_checkout_manifest.cjs:52/61 and merge_remote_agent_github_folder.cjs:194-215 (both carry the marker + shim.cjs, both call execFileSync for git/gh).
2. Latent false negative — retainsCallResult too broad. The "handle retained for streaming" exemption for exec()/execFile() fires for any non-bare-statement parent (await, void, logical/conditional/sequence expressions), not just genuine retention shapes (assignment, return, member-chain). await exec(cmd, cb) — a plausible copy-paste artifact from @actions/exec's real async exec() — would silently escape the rule. No live occurrence found (0 matches for await exec(...) etc. in the corpus), so this is a latent gap, cheap to close before it hides a real bug.

Grounding detail

• eslint.config.cjs registers gh-aw-custom/prefer-actions-exec-over-child-process as "warn" for all non-test .cjs files — no per-file excludes exist for the shim.cjs dual-mode population, so both grounded false positives above are live warnings today.
• Confirmed true negative: git_helpers.cjs:21 (execSync, marker present, no shim.cjs require) is correctly flaggable — this is a real migration candidate, not a rule bug, and cleanly demonstrates the rule works as designed outside the dual-mode blind spot.
• Repo clone is shallow (.git/shallow present) — git log/blame all report the same date for every rule file, so recency was determined from eslint-factory/src/index.ts's import-list append order instead.
• Prior open issues from before the 46-day gap (#43948 + sibling, #43486 + sibling, #43326, #42915, #42682/#42680, #42190/#42189/#42188, and the two filed 2026-07-08 for require-fs-sync-try-catch) were not re-verified this run — flagged in memory to re-check via gh api before citing in a future report, since they may already be closed or superseded.

Next actions

• Continue the newest-rule-first cadence: 45 rules remain unreviewed (no-throw-plain-object, no-github-request-interpolated-route, require-mkdirsync-try-catch, no-child-process-interpolated-command, no-empty-catch-block, and 40 more — full list in repo-memory state.json). Given the size of the backlog, this will take many runs; not attempting to clear it in one sitting.
• Once the shim.cjs fix lands, re-verify build_checkout_manifest.cjs:52/61 and merge_remote_agent_github_folder.cjs:194-215 are silenced.
• Re-verify stale prior-run issue numbers before citing them again.
• Strategy, findings, and history persisted to repo-memory (memory/eslint-refiner branch) for continuity.

Generated by 🤖 ESLint Refiner · claude · agent · 188.2 AIC · ⌖ 7.66 AIC · ⊞ 5.8K · ◷

• expires on Aug 26, 2026, 9:31 PM UTC-08:00